A few months ago, I posted an essay “How I Read a Randomized Clinical Trial Before I Believe It.” Step 1 asks “What’s the Outcome and Who Cares?” The recently published STAREE study (1) illustrates why that is so important to know, and how you can be misled by a seemingly common word: Disability. Inconceivable! A composite built from MI, stroke, revascularization, or cardiovascular death is built from things a clinician can verify: an ECG, a troponin, a stent, a death certificate, and major adverse cardiovascular events. (“MACE”)-type endpoints are similar enough across cardiovascular disease (CVD) studies that you know what the writers mean by MACE-3 or MACE-4 without having to go searching for the definition. There may be minor definitional nuances between trials, but we pretty much know what they mean. However, a composite that includes an ordinary word like “disability” is different. Disability isn’t a clinical event. It’s a lay term with no inherent boundary, and a trial has to invent one before it can measure it. That invented boundary, not the word on the page, is what the result actually describes. STAREE has one endpoint of each kind.

STAREE randomized 9,971 healthy, independent adults at least 70 years old, with no history of CVD, diabetes mellitus, or dementia, to atorvastatin 40 mg or placebo. Over a median of 5.9 years, atorvastatin reduced MACE-4 (cardiovascular death, nonfatal MI or stroke, coronary revascularization) by 30% (hazard ratio 0.70, 95% CI 0.61-0.82, P<0.001), driven mostly by fewer heart attacks and revascularizations with a number needed to treat of 37. This is the kind of endpoint you can accept at face value: a robust result in a population where evidence for statin effects in primary prevention has been challenged as thin. I agree with those calling it good evidence for primary prevention in older adults.

But the trial’s other primary endpoint is problematic. It’s a composite of death, dementia, or “persistent physical disability,” meant to capture disability-free survival, and it did not show a significant difference (hazard ratio 0.94, 95% CI 0.84-1.05, P=0.25). Disability-free survival is an important endpoint, since many of today’s heart attacks are small and outcomes are much better with primary revascularizations and other therapies. That is exactly why a well-defined disability endpoint is important, and why this definition deserves scrutiny before anyone draws a conclusion from it. As expected, the statin-naysayers already have latched onto it. However, the problem is that death and dementia are real endpoints with real boundaries, but “disability” isn’t, and before I could decide what this null composite endpoint event even meant, I had to find out what STAREE meant by the word in it.

“Disability” was defined much narrower than the word suggests

First, I looked at the study design in the methods section (remember Step 2?), but the main paper never defined it. The abstract, methods, results, and discussion all use “persistent physical disability” as though it were self-explanatory. But the actual definition sits in Appendix 2: loss of a basic activity of daily living, rated “a lot of difficulty,” “unable to do,” or “requiring help,” using a modified Katz ADL, persistent for at least six months, or a clinical record of aged-care admission for disability. You cannot get this from the main paper, so we have a primary endpoint whose definition never appears in the paper reporting it. Inconceivable~

Who cares? Picture someone who has a stroke and needs help bathing, dressing, and transferring for six months, then by month seven can do all three again, but is left with a permanent limp and gait imbalance, so they walk with a walker and their speech is very hard to understand. That person does not meet this endpoint. The six ADL items resolved before six months, but the lasting impairment did not.

I don’t think most people would call that person free of disability. They had a stroke and were left with a handicap sticker in their car, need to use a walker, and have a noticeably slower and more cautious version of an independent life, frustrated by trouble speaking. That is what most people picture when they hear “disability after a stroke.” None of that reaches the bar STAREE set. What the trial actually measured is closer to severe, permanent dependence in basic self-care, and I’d have named the endpoint “severe disability” rather than plain “disability.” A null result on that narrow a definition tells you much less than the label implies.

Once you know the definition, the rest of the data reads differently

“Persistent physical disability,” defined at that high bar, still trended toward benefit: HR 0.74 (0.53-1.04). So did all-cause death: HR 0.91 (0.79-1.04). Both confidence intervals sit close enough to 1 that a modestly longer trial could plausibly have separated the curves. Dementia is the exception, essentially flat at HR 1.03 (0.87-1.21). Furthermore, the disability component had only 140 events total across both arms, and the paper says outright that this small number “limited our ability to draw conclusions about this component of the primary end point.” That’s a different finding than “no effect,” even when both get reported the same way.

The comparison group wasn’t clean either: by year 5, 19.4% of the placebo arm was taking an open-label statin, versus 9.4% of the atorvastatin arm, roughly a fifth of the “untreated” group was receiving the drug under study. That kind of drop-in biases any true effect toward the null. The CVD endpoint cleared P<0.001 anyway, meaning its true effect is probably larger than reported.

STAREE also screened over 22,000 people to enroll a population that was 98% white, independent, and free of CVD, diabetes mellitus, and dementia - that is the right group for isolating a primary prevention signal but not the frail, multimorbid patients this trial will get cited to treat.

Atorvastatin also had more medically important musculoskeletal, hepatobiliary, and diabetes-related adverse events, though overall serious adverse events were identical (2.7% each), costs worth weighing against the real benefit, especially in older patients.

The lesson, and the trial’s real verdict

STAREE is good evidence that statins prevent CVD events in healthy older adults and it updates our prior expectation that they are useful. But it is not good evidence that they fail to extend disability-free survival, because the endpoint meant to capture that, as defined, missed a lot of what most of us would call “disability.” Most of what kills or disables people this age has nothing to do with their coronary arteries: 80% of the deaths in this trial were non-cardiovascular, so a 30% cut in CVD risk was never going to move a composite that broad on its own.

I think another year or two of follow-up would have made this result clearer. Primary prevention studies enroll healthier people than we see in practice or than the secondary prevention trials that built most of what we know about preventing CVD. To be fair, 5.9 years is long by most standards, but hard endpoints like death and disability in a healthy population need longer than that to accumulate enough events to answer cleanly. Short of that, we keep getting the same thing: reductions in nonfatal events, borderline trends, and results tantalizing enough to feel like they show something while convincing almost no one. If a drug is worth prescribing to millions of people for life, we should design trials that can prove it. STAREE proved statins cut CVD events in healthy older adults, but it did not disprove that they preserve independence or quality of life, and until someone runs a trial built to actually answer that question, the honest verdict is not known. I suspect they do, but we need to run the study to be sure.

Reference:

1. Zoungas S, Wolfe R, Moran C, et al. Atorvastatin, cardiovascular events, and disability-free survival in older adults. N Engl J Med 2026. doi: 10.1056/NEJMoa2607314.

Knowing your “VO₂ max,” the maximum amount of oxygen your body can use during hard exercise, has become popular, largely because smartwatches and fitness rings now put a number on your wrist or finger (1). Peak oxygen uptake is one of the best single measures of overall fitness and a strong predictor of long-term health (2). But how that number is produced matters enormously, and most of the popular ways of getting it are far less reliable than people assume.

The problem with wearable “VO₂ max” numbers
As I discussed previously, the VO₂ max shown by a watch or ring is an estimate, not a measurement. These devices use proprietary formulas that guess your VO₂ max from things like your age, sex, body size, resting heart rate, and how your heart rate behaves during activity (1). They never actually measure the oxygen you breathe. The result is a number that is often inaccurate and highly variable from person to person. Estimates based on resting data tend to run high, on average by ~2 mL/kg/min, but with very wide scatter at the individual level, where a given person’s value can be off by ~13 mL/kg/min too low to ~17 mL/kg/min too high (1). Because that individual error is so large, these numbers are of little use for watching your own trend over time.

What a CPET is, and why it is the gold standard
A cardiopulmonary exercise test (CPET) measures the oxygen you breathe in and the carbon dioxide you breathe out through a mask while you exercise on a treadmill or bike. The mask connects to a “metabolic cart” that directly measures your VO₂ rather than estimating it, which is why it is considered the gold standard. It also can confirm you truly gave a maximal effort, something a watch or a routine stress test cannot do.

Why the “old-fashioned” treadmill estimate is only a rough guide
It is true that you can get an estimate of overall fitness from an ordinary treadmill stress test, and that the large studies linking fitness to longer survival mostly used plain treadmill tests, not CPETs (2,3). So the treadmill has real value. The catch is in how the number is calculated:

· Your “METs” (a measure of workload) are estimated from the speed and slope of the treadmill using formulas that have built-in error, especially when the test is too hard for the person taking it (3).

· The final step, multiplying METs by 3.5 to estimate VO₂ max, relies on a constant that is a population average derived mostly from healthy men. It has limited validation in women, older adults, and people with medical conditions (3,4).

· Different treadmill exercise protocols (for example, the standard Bruce vs. a gentler protocol like the modified Bruce, Balke, or ramp protocols) can give meaningfully different results in the same person, and holding the handrails inflates the number further. This is seen in clinical and research labs all over the world. Imagine how inaccurate these measures will be when you do it on your home or gym treadmill that is less well-calibrated, and performed under non-standardized conditions.

The key point: When researchers have compared the treadmill estimate against a real metabolic-cart measurement, the treadmill estimate consistently overestimates the CPET measured value. In 99 healthy, active adults tested to exhaustion, the ACSM equation overestimated measured VO₂ max by 9.8 mL/kg/min (~21%) (4). A much larger analysis of 7,983 people tested across several protocols found an overall error of 21.4 ± 24.9% (5). Yes, the standard deviation is so high that it exceeds the average error, which means that in any one person the size of the error, and sometimes even its direction, is almost unpredictable. The bottom line is that the treadmill estimate is a useful gauge of your general fitness category, but it is a rough rule of thumb rather than a precise measurement, and it is not reliable for tracking small changes over time (4,9).

The real reasons VO₂ max is worth knowing, even if the exact number is imperfect
Fitness is one of the strongest predictors of living longer and staying free of heart disease, but there are three important points. First, the relationship is not a straight line. The biggest health gains come from moving out of the lowest fitness category; above ~10 METs, additional fitness keeps helping but the extra benefit per unit gets smaller. Most reasonably active people already are well past the steepest part of the risk curve (2,6). Second, the numbers used to define “good” or “bad” come from large populations, so they describe groups better than they pin down any one individual. And third, a practical reason to test is that many people misjudge their own fitness. It is common for my patients to describe what they “used to” do (“I used to play football,” “I used to run 5Ks”) while underestimating how quickly fitness fades with age, and especially how much is lost after repeated illnesses, injuries, and periods of inactivity. An objective test replaces memory with a real starting point.

The best reason to do a CPET: training smarter
I reviewed the utility of CPET here. However, I want to emphasize that the single most useful thing that a CPET gives you is not the headline VO₂ max value. It is your two “ventilatory thresholds,” VT1 and VT2. These are the exercise intensities where your body’s metabolism shifts gears, and they only can be pinpointed by measuring your expired respiratory gases during a CPET (7). VT1 marks the top of easy, sustainable “Zone 2” training, the intensity that builds your aerobic engine (mitochondria and small blood vessels in muscle). VT2 marks the point above which effort becomes truly hard and can only be sustained briefly, the zone used sparingly to push peak performance and VO₂ max.

Setting your training zones off these personal thresholds is far more accurate than the generic “220 minus your age” heart-rate formulas. Indeed, the European Society of Cardiology explicitly advises against using the “220 minus your age” formula because the scatter around the age–heart-rate line is very wide, and heart-rate targets are further distorted by beta-blockers (8). Without knowing your own thresholds, a VO₂ max score alone is mostly a number for bragging about or for beating yourself up over, and it does not tell you how to train.

My takeaways
Be very skeptical of the VO₂ max value on your watch or ring. Personally, I do not wear any of these devices and if I did, I’d especially ignore the VO₂ max estimate. An ordinary treadmill stress test gives a useful fitness category and prognosis, but the METs x 3.5 = VO₂ max estimate is, at best, a rough guide that overestimates and is not precise in individual people (4,5).

If you genuinely want to quantify your fitness and, more importantly, train effectively, get a CPET. The real value is the personalized training zones, VT1 and VT2, that it provides. I used DexaFit here in Madison, WI and I suspect there is one near you.

I’ll end with one caveat: even the gold-standard CPET generally has to show a change on the order of 10% before it counts as a real improvement rather than day-to-day noise, which is why the far wobblier wearable and treadmill numbers are pretty useless for tracking small gains (9).

References

1. Petek BJ, Al-Alusi MA, Moulson N, et al. Consumer Wearable Health and Fitness Technology in Cardiovascular Medicine: JACC State-of-the-Art Review. Journal of the American College of Cardiology 2023;82:245-264.

2. Kokkinos P, Faselis C, Samuel IBH, et al. Cardiorespiratory Fitness and Mortality Risk Across the Spectra of Age, Race, and Sex. Journal of the American College of Cardiology 2022;80:598-609.

3. Ross R, Blair SN, Arena R, et al. Importance of Assessing Cardiorespiratory Fitness in Clinical Practice: A Case for Fitness as a Clinical Vital Sign: A Scientific Statement From the American Heart Association. Circulation 2016;134:e653-e699.

4. Dugas MO, Paradis-Deschênes P, Simard L, et al. Comparison of VO2max Estimations for Maximal and Submaximal Exercise Tests in Apparently Healthy Adults. Sports 2023;11:235.

5. Kokkinos P, Kaminsky LA, Arena R, et al. New Generalized Equation for Predicting Maximal Oxygen Uptake (From the Fitness Registry and the Importance of Exercise National Database). American Journal of Cardiology 2017;120:688-692.

6. Kokkinos P, Faselis C, Samuel IBH, et al. Changes in Cardiorespiratory Fitness and Survival in Patients With or Without Cardiovascular Disease. Journal of the American College of Cardiology 2023;81:1137-1147.

7. Bishop DJ, Beck B, Biddle SJH, et al. Physical Activity and Exercise Intensity Terminology: A Joint American College of Sports Medicine (ACSM) Expert Statement and Exercise and Sport Science Australia (ESSA) Consensus Statement. Medicine and Science in Sports and Exercise 2025;57:2599-2613.

8. Pelliccia A, Sharma S, Gati S, et al. 2020 ESC Guidelines on Sports Cardiology and Exercise in Patients With Cardiovascular Disease. European Heart Journal 2021;42:17-96.

9. Faricier R, Keltz RR, Hartley T, et al. Quantifying Improvement in V̇O₂peak and Exercise Thresholds in Cardiovascular Disease Using Reliable Change Indices. Journal of Cardiopulmonary Rehabilitation and Prevention 2024;44:121-130.

After 36 years of practicing medicine, most of which included clinical research, I have learned that things that seem obvious and make perfect sense often do not work. Sometimes they even cause harm. Since this past week was my birthday, I decided to go a little off path and connect my clinical and scientific experiences to how I think about gadgets for my bike, specifically radar.

Supplemental Oxygen

A myocardial infarction (”MI” or “heart attack”) occurs because part of the heart is not getting enough oxygen. For most of my career, we gave patients with an MI or a suspected MI supplemental oxygen via small plastic tubes in their noses. It made perfect sense and was the standard of care for nearly a century. But when oxygen was tested in the DETO2X-AMI trial, it did not reduce mortality or rehospitalization (1). Furthermore, in the AVOID trial of MI patients, supplemental oxygen was associated with increased myocardial injury and larger infarct size at six months, compared with ambient air, likely due to oxidative stress and vasoconstriction (2). Beyond MI, liberal oxygen therapy caused a dose-dependent increase in mortality in acutely ill patients (3), and an ICU trial found that liberal oxygen therapy nearly doubled ICU mortality compared with a conservative approach, with more shock, liver failure, and bloodstream infections (4).

Treating Anemia in Hospitalized Patients

Anemia reduces the oxygen-carrying capacity of blood. Because oxygen delivery is essential for tissue repair and fighting infection, it seems logical that giving blood transfusions to anemic patients would improve outcomes. Yet randomized trials consistently have shown that liberal transfusion strategies usually do not improve outcomes compared with more restrictive approaches and they are associated with harms. In the TRICC trial of ICU patients, 30-day mortality was similar, but in-hospital mortality was lower with the restrictive transfusion strategy (5). Moreover, pulmonary edema and MI were more frequent in the liberal transfusion group. Each transfusion carries risks of acute lung injury and circulatory overload, and observational data link transfusions to increased rates of infections and blood clots. Acute MI may be an exception: transfusing if the hemoglobin level is <10 g/dL may improve outcomes (6), but the broader lesson is that blood transfusions, which seem like an obvious solution to anemia, may not help and often cause harms.

Medicine is filled with examples like these. We identify a problem, find something that changes the physiology or a number we can measure, and assume that changing it will improve the outcome we actually care about. Sometimes it does, but often it does not, and sometimes the intervention has adverse, unintended consequences. After enough years in medicine, I have become skeptical of the phrase “It just makes sense.” Sometimes I flippantly respond “In G-d we trust, everyone else must show data.”

Which brings me, somewhat improbably, to bike radar.

Radar for Road Bikes: Lack of Data

Several of my cycling friends have the Garmin Varia or similar rear radar devices and rave about them. A small radar device mounted on the back of your bicycle that detects vehicles approaching from behind, sometimes hundreds of feet away, can alert you through a cycling computer. The reasoning is that if you know a car is coming earlier, you should be safer. That makes perfect sense, but is the kind of claim I have learned to test rather than accept on its face.

After reviewing the Garmin Varia 515 and similar rear radar devices in Consumer Reports and cycling reviews, I decided to pass. I didn’t do a randomized clinical trial because an unbiased funder would look at the premise (the reasoning behind the experiment) and see that it is unlikely to hold up. Here is why.

At its limit of roughly 450 feet, for a vehicle driving 55 mph, the device provides about 5-6 seconds of warning (less if they are driving faster). That is enough time to register an alert, but I do not think it meaningfully changes the decisions available to a cyclist. The options remain the same as seeing a vehicle in your rear view mirror or hearing it approach: hold your line, move right (even if that means moving off the road), or signal your presence by waving your arms. Those decisions are made as the car approaches, not from that far away. Earlier detection, in this context, adds information without adding options.

The more fundamental question is how far back you can tell whether an approaching vehicle is going to hit you or go around you, and radar cannot answer that. That determination depends on the driver’s trajectory, which only becomes apparent when the car is close enough for you to read directly, at which point you already are in decision time regardless of what the device told you a few seconds earlier.

This is where my research instincts kick in. The marketing suggests safety benefits, but the supporting literature provided by the manufacturer validates detection accuracy rather than safety outcomes. There are no data showing that radar, or even early detection, reduces crashes or injuries, a gap the marketing does not acknowledge. In medicine, we recognize this as the problem of the surrogate endpoint. Showing that an intervention changes something we can measure is not the same as showing that it improves the outcome we actually care about. A drug can lower a laboratory value without helping patients, and sometimes the action taken in response to the lab report causes harm. Similarly, just because a device can detect an approaching car sooner does not mean it will help you avoid a collision, and there is at least a reasonable chance it will increase your risk because of unforeseen consequences.

What works?

High visibility jackets and rear lights (snore!). A randomized controlled Danish trial of 6,793 cyclists reported a 47% lower rate of self-reported personal-injury bicycle crashes among cyclists assigned a high-visibility yellow jacket, relative to a control group without one, including a 55% reduction in accidents with motor vehicles (7). Another Danish study of 3,845 cyclists reported a 19% lower rate of self-reported personal-injury bicycle crashes among cyclists fitted with permanent running lights, relative to a control group without them (8).

Personally, I wear brightly colored, reflective jerseys and jackets and have a bright blinking taillight because that research is good enough for me. I also rely on a rear-view mirror I have used effectively for over 20 years. I understand that no controlled trial has tested whether bicycle mirrors reduce crash or injury rates, but I am not stupid. I have neck problems and terrible balance, so turning my head to look is not a good option. These three items provide continuous awareness and make me conspicuous to drivers without adding another layer of monitoring. I personally am not persuaded that the added cost, added weight, and the inconvenience of having another device to charge and manage are justified, particularly when I am trying to enjoy my ride and focus on the environment around me, for pleasure and for safety.

Unintended Consequences

OK, Dr Scrooge. What’s the harm?

Published research does show that alerts may introduce distraction or divided attention, and also may promote overreliance on a device with inherent limitations (9). Real-world use of bike radar devices suggests that the audible signal can be too quiet or inconsistent depending on ambient noise from motor vehicles or the environment, which necessitates visual confirmation on a screen. Then the user’s attention is diverted from the road, to the screen. Indeed, the most important safety device on my bicycle is my attention.

Furthermore, most of what can hurt a cyclist is in front of them: potholes, gravel, debris, pedestrians, other cyclists, and cars either parked or entering the roadway. Even before radar, I have seen cyclists become so absorbed in their cycling computer that they rode off the road, and others have hit parked cars in front of them because they were not paying close enough attention. And I can’t count the number of times riders I know forgot to charge their light (the one with safety evidence) but did remember to charge their radar device.

There also is the problem of alert fatigue (10). Anyone who has driven a car that constantly beeps because you are too close to something in front of you, too close to something next to you or behind you, or drifting toward a lane marker knows what eventually happens. You start ignoring the beeps. I suspect that the same thing happens with bike radar. When I ride with friends who use it, their devices seem to beep frequently, sometimes almost constantly. They continue talking and riding. When I am the one announcing “Car back!” I am not convinced they are more aware of approaching cars than I am using my $20 mirror. At some point, radar alerts can simply become part of the background and literally add real noise, but no signal.

None of this means the device has no value. It may help riders who cannot use a mirror well or who ride long solo stretches on open roads and believe it increases their awareness. But for my riding style and habits, I do not see an incremental benefit and the added complexity is not worth it. Others may decide differently, though the device should not be viewed as a safety panacea.

And that brings me back to medicine. The Garmin Varia unquestionably detects approaching cars. The more important question is whether knowing about those cars a few seconds earlier makes cyclists safer. Maybe it does, but we have made this mistake in medicine too many times to assume that because the mechanism makes sense, the outcome necessarily follows.

Medicine has taught me to ask a simple question whenever something seems obviously beneficial: Does it work because we tested it, or does it work because we can explain why it should?

Ride safely.

References

1. Hofmann R, James SK, Jernberg T, et al. Oxygen therapy in suspected acute myocardial infarction. N Engl J Med 2017;377:1240-1249.

2. Stub D, Smith K, Bernard S, et al. Air versus oxygen in ST-segment-elevation myocardial infarction. Circulation 2015;131:2143-2150.

3. Chu DK, Kim LH, Young PJ, et al. Mortality and morbidity in acutely ill adults treated with liberal versus conservative oxygen therapy (IOTA): a systematic review and meta-analysis. Lancet 2018;391:1693-1705.

4. Girardis M, Busani S, Damiani E, et al. Effect of conservative vs conventional oxygen therapy on mortality among patients in an intensive care unit: the Oxygen-ICU randomized clinical trial. JAMA 2016;316:1583-1589.

5. Hébert PC, Wells G, Blajchman MA, et al. A multicenter, randomized, controlled clinical trial of transfusion requirements in critical care. N Engl J Med 1999;340:409-417.

6. Carson JL, Brooks MM, Hébert PC, et al. Restrictive or liberal transfusion strategy in myocardial infarction and anemia. N Engl J Med 2023;389:2446-2456.

7. Lahrmann H, Madsen TKO, Olesen AV, et al. The effect of a yellow bicycle jacket on cyclist accidents. Safety Science 2018;108:209-217.

8. Madsen JCO, Andersen T, Lahrmann H. Safety effects of permanent running lights for bicycles: A controlled experiment. Accid Anal Prev 2013;50:820-829.

9. Kim S, Oviedo-Trespalacios O. Behavioural adaptation in Advanced Driver Assistance Systems (ADAS) use.Applied Ergonomics 2026; 136:104799. ISSN 0003-6870. https://doi.org/10.1016/j.apergo.2026.104799.

10. UC Davis PSNet Editorial Team. Alert Fatigue. PSNet. Rockville (MD): Agency for Healthcare Research and Quality, US Department of Health and Human Services. 2019. https://psnet.ahrq.gov/primer/alert-fatigue.

One of the gifts of social media is the democratization of science and critical analysis of the medical literature. That kind of scrutiny used to be gatekept by journal editors: a critical letter to the editor rarely was accepted, and if published, it appeared months later, relegated to the back of the journal or behind a paywall, nearly guaranteeing that no one read it. Now we can review a paper together, online, where many thousands more people will see and interact with it. So, let’s have a Journal Club. We will review an article’s highlights and then use our critical appraisal skills to analyze it so we can better understand and translate it for patient care.

I picked this paper (1) because it just was published in a top-tier cardiology journal and came from a well-respected group - the MESA (Multi-Ethnic Study of Atherosclerosis). Indeed, I am a MESA investigator, ran its carotid ultrasound reading center for over a decade, and have published with several of the co-authors who are my friends. I chose it because papers that confirm what we already expect often get the least scrutiny. As an observational paper, the framework I described in “How I Read a Randomized Clinical Trial” does not map perfectly. But some of the good habits do: read methods, tables, and figures before the abstract and conclusion, consult the supplemental data when something is not clear, and be skeptical toward anything that arrives already wrapped in a bow.

The paper

The 2026 ACC/AHA dyslipidemia guideline used the new PREVENT equations to estimate cardiovascular disease (CVD) risk and sorted people into four 10-year risk categories: low (<3%), borderline (3 to <5%), intermediate (5 to <10%), and high (≥10%), for consideration of statin therapy among those who do not have other reasons they need it (2). The guidelines’ “Calculate-Personalize-Reclassify and Reassess” method states that if there is “patient or clinical uncertainty” about starting a statin, people at borderline and intermediate risk can be “re-classified and re-assessed” using coronary artery calcium (CAC) testing (2). Rikhi et al (1) used the MESA to ask how CAC performs inside this framework. Here is their conclusion, directly quoted from the abstract:

“CAC provided its greatest clinical value when statin treatment decisions were uncertain, particularly among individuals with borderline predicted risk.”

That sentence confirms the guideline and hands clinicians a clean rule: when in doubt, get a CAC score, and it will tell you what to do. Physicians love simple rules because they make a complicated, busy job much easier.

What does “uncertain” even mean?

“Uncertain” never was defined or measured in this paper, and it could refer to at least three different things that aren’t interchangeable. First, outcome uncertainty - whether a given patient will have an event. But that is axiomatic at any predicted risk, because a 20% risk is no more “certain” than a 4% risk, only higher. Second, estimate uncertainty. That is how much confidence to place in the number itself, which depends on calibration, measurement error, and how well the derivation cohort resembles the patient in front of you; this paper never clearly addressed that. And third - what the paper seems to be implying is decisional uncertainty. That is whether, after the clinician and patient have talked, the choice remains unsettled. But that uncertainty exists in the exam room, not in a database, and was not assessed. Keep those three meanings in mind as we look at what the study measured because “uncertain” is doing almost all the work in this paper’s conclusion. After reading this paper, I think borderline and intermediate risk are being elided and relabeled as "uncertain," with risk categories substituted for uncertainty; however, those categories carry no such meaning. Indeed, a predicted risk of 4% is a definite estimate of a borderline risk over 10 years: neither high nor low.

What kind of paper is this?

This is an observational cohort study, so it cannot support causal claims, because of confounding and bias. How to appraise such studies has been covered by others (3-5), but two limitations deserve particular attention here.

The first is who enrolls. Volunteers for long-term cohort studies are healthier than the general population, and MESA explicitly selected for health: 45-84 year-olds free of clinical CVD, enrolled between 2000 and 2002. That design is right for studying how atherosclerosis develops, but it also means that absolute event rates run lower than what clinicians see in practice. This is not a generic concern about generalizability: it goes to the paper’s central finding. Whether the CAC-negative borderline group lands below the 3/1,000 person-year line depends as much on the cohort’s absolute event rates as on anything about CAC, and a healthy volunteer cohort from 25 years ago is about as friendly a setting as you could find for a subgroup to slip under that line.

The second is what happened during follow-up. This analysis covered the decade after enrollment, a period of marked increase in statin use in the US (6) and in MESA (7). In fact, MESA participants were given the results of their CAC scans, and by 2012, 1,281 participants had newly started statins (7). Who started them was not random since a high calcium score made statin treatment more likely, and statins lower event rates, so the observed rates in the calcium-positive groups are treated rates, attenuated by the very marker the study set out to evaluate. The results compare these attenuated, treated rates against a risk threshold as though they were natural history, which leads to a treatment bias.

What do the results show?

The results show what has been demonstrated previously: CAC presence and burden are associated with future CVD events across risk levels; what’s new is that the analyses used the risk categories from the 2026 guidelines and the PREVENT score rather than older risk models (1,2).

What caught my eye was the central illustration. Look at the CAC-negative and CAC-positive event rates per 1,000 person-years in every risk category: low risk (<3%): 1.1 versus 3.0. Borderline (3 to <5%): 2.7 versus 5.2. Intermediate (5 to <10%): 5.9 versus 11.3. High (≥10%): 6.2 versus 20.9. In every category, calcium-positive participants had meaningfully higher event rates, which is what a prognostic marker should do and is old news. But look how close those values are to the original predicted ranges: for the two categories the authors seem to care about, the “Borderline” risk changes by 0.3 at the low end and 0.2 at the high end, and for “Intermediate” risk, 0.9 at the low end and 1.3 at the high end.

That is tiny! And the supplement shows why. The PREVENT risk calculator was well calibrated in this cohort: the observed 10-year incidence in the borderline category was 3.9%, matching the predicted 3.9% exactly (reference 1, Supplemental Table 8). The model already was right about these people. The CAC split just divided a correctly predicted average into a slightly lower half and a slightly higher half.

Are these even findings or just arithmetic?

In any risk category, if you split people into two groups using a marker associated with events, the two subgroup rates must average back to the category’s overall rate, weighted by person-time. That’s arithmetic: splitting a category by CAC status necessarily produces one subgroup above the average and one below it.

So look again at the borderline category. Its CAC-negative rate, 2.7/1,000 person-years, sits just under the 3/1,000 line the paper treats as a relevant threshold, and its CAC-positive rate, 5.2, sits just above the 5/1,000 line. But that is what happens whenever a category’s average risk sits close to a round number that is considered a threshold. Any genuinely prognostic marker will appear to resolve uncertainty near a threshold once you go looking, because averages have to average.

As expected, the formal test for whether CAC behaves differently across PREVENT risk categories, the interaction term, was not significant (p=0.47). A separate analysis by statin eligibility found one statistically significant interaction (p=0.047), but it was unadjusted and appears to have been driven by heavily calcified people in the statin-not-recommended group, a comparison resting on only 10 events, and not even the borderline group the conclusion was built around.

The pivotal number deserves a closer look, too. The claim that a calcium score of zero moves borderline patients below the threshold rests on that 2.7, which came from only 13 events. Its 95% confidence interval runs from 1.6-4.6, spanning the threshold of 3 with plenty of room to spare. A conclusion this consequential should not balance on a point estimate that fragile.

The paper’s summary statistics tell the same story. As expected from a well-calibrated model and small shifts at the margins, the net reclassification improvement (NRI) when CAC was added was modest: 0.124, lower than earlier reports in this cohort, albeit with longer follow-up and slightly different CVD endpoints (8). That does not mean 12.4% of people were meaningfully reclassified - it sums the net share of events shifted to a higher category and nonevents shifted lower, roughly 6% each. Among the 374 participants who had events, the entire net gain came from movement between the 5 to <10% and ≥10% categories, a boundary that changes nobody’s statin eligibility, since both already call for treatment. At the two boundaries that do change eligibility, the net movement among future events ran the other way: a net 5 shifted below the 3% threshold where statins are first considered, and a net 4 below the 5% threshold where statins are recommended, meaning shifted out of treatment eligibility for people who then had events. The positive headline number lives entirely at a boundary with no treatment consequence, while at the two boundaries that decide who gets a statin, the net movement pointed away from treatment.

What the conclusion should have said

Abstracts and conclusions are the most heavily read sections of any paper. This one concludes: “CAC appears to be most informative when treatment decisions remain uncertain, particularly among individuals with borderline predicted risk.” Now the substitution is easy to see: the study measured event rates in borderline and intermediate risk categories, terms that describe the size of a risk estimate, while the conclusion claims something about decisions that “remain uncertain,” which is the state of a conversation between a clinician and a patient. The authors use these words interchangeably, but the guideline does not: it conditions CAC testing on the uncertainty that remains after the clinician and patient have talked, and the paper collapsed that into a risk category. Then the pattern offered as proof is mostly what any correlated marker would produce once you understand how averages work.

Some might say “Jim, we know that you do not like CAC testing. Please stop it already.” But to be clear, I do not like or dislike any test; I care about what a test means and using it correctly. I accept that CAC predicts events in every risk category, but I am skeptical of its utility. And I agree with the paper’s other conclusion, that a CAC of zero should not be used to withhold statins from people who already meet treatment thresholds. This paper illustrated what I said in my very first Substack post, a year ago this month: Different nets catch different fish. Today’s corollary: your net may not be built for the pond you’re fishing in.

A more precise conclusion would read something like this: CAC adds modest prognostic information, spread fairly evenly across risk categories rather than concentrated at the borderline, and its practical value within the PREVENT framework remains an open question. That version is true, but less likely to get published, and, I promise you, will never be quoted in a slide deck.

The takeaway

Reading a paper this way requires developing good habits: read the methods and tables before the abstract, consult the supplement when something is not adding up, and for observational data, be skeptical about causal claims or projected use in patients. That’s the difference between reading a paper and being told what it says.

References

1. Rikhi R, Chen H, Mirzai S, et al. The role of coronary artery calcium in statin eligibility based on the American Heart Association’s PREVENT Calculator: Insights From MESA. J Am Coll Cardiol 2026. doi: 10.1016/j.jacc.2026.05.039.

2. Blumenthal RS, Morris PB, Gaudino M, et al. 2026 ACC/AHA/AACVPR/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA guideline on the management of dyslipidemia: a report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol 2026;87:2624-2757.

3. Dahabreh IJ, Bibbins-Domingo K. Causal Inference About the Effects of Interventions From Observational Studies in Medical Journals. JAMA 2024;331:1845-1853.

4. Ramspek CL, Steyerberg EW, Riley RD, et al. Prediction or causality? A scoping review of their conflation within current observational research. Eur J Epidemiol 2021;36:889-898.

5. Grimes DA, Schulz KF. Bias and causal associations in observational research. Lancet 2002;359:248-252.

6. Sarpong EM, Zuvekas SH. Statistical Brief #459: Changes in Statin Therapy among Adults (Age ≥ 18) by Selected Characteristics, United States, 2000-2001 to 2010-2011. November 2014. Available at https://meps.ahrq.gov/data_files/publications/st459/stat459.shtml.

7. Strand LN, Young RL, Bertoni AG, et al. New statin use and left ventricular structure: Estimating long-term associations in the Multi-Ethnic Study of Atherosclerosis (MESA). Pharmacoepidemiol Drug Saf 2018;27:570-580.

8. Polonsky TS, McClelland RL, Jorgensen NW, et al. Coronary artery calcium score and risk classification for coronary heart disease prediction. JAMA 2010;303:1610-1616.

In my last post I argued that supplements marketed for heart disease do not work. The question that follows, and I hear some version of it almost every week in clinic, is “Fine, but what about everything else?” There are thousands of products on the store shelves, far too many to take up one at a time, and for most of them I have no strong opinion other than “probably a waste of money.” So rather than work through a list, I will use a single example: creatine, because it is among the few supplements with any real research data from studies that have evaluated it. If the case is thin for creatine, that tells you most of what you need to know about the rest.

Creatine

I am not a big fan of creatine. In a person with normal kidney function, it probably is harmless. My primary hesitation is that the supplement industry is regulated far less rigorously than the pharmaceutical industry. Because it lacks routine premarket FDA review of safety, efficacy, and product quality, I never am fully confident about the dose or purity of what a given patient actually is taking (1,2). One clarifying note before going further: creatine is the supplement, and creatinine is the muscle breakdown product measured in a clinical chemistry lab and used to assess kidney function. The two are connected: creatine supplementation can push your serum creatinine modestly upward, because some of it is converted to creatinine and adds to the blood pool, without any true decline in kidney function (3,4). If your serum creatinine level goes up and you take this supplement, tell your physician before embarking on a detailed work-up.

As for a possible benefit of creatine supplements, certain preparations at high doses have been shown to increase maximum weight-lifting reps, which can serve as a stimulus to muscle growth (5). For a competitive bodybuilder, that margin may be worth chasing. But for most people it sits well down the list, behind the two interventions that actually do the work: consistent resistance training and adequate protein intake. I will concede that in older adults trying to not lose muscle, for whom sarcopenia can be a genuine clinical problem, some data on resistance training combined with creatine suggest some benefit, although the data are quite weak. Two analyses of pooled randomized trials in adults roughly 55-70 years old found that creatine added to resistance training produced about 3 pounds more lean tissue mass and greater strength on chest press and leg press than training alone (6,7). That may sound impressive, but these two meta-analyses drew from largely the same small pool of trials and many of those trials were authored by the same group of investigators, so they are not truly independent confirmations. Furthermore, the individual studies were small (typically 12-40 people per arm), of very short duration (range 7-16 weeks, average around 12-13 weeks), and blinding was questionable, because creatine causes water retention and early weight gain that people can notice. Indeed, water retention can bias lean mass measurements, so some of the lean mass gain likely reflects intracellular water, not new muscle protein. The strength results were uneven, with improvements on chest press and leg press but not on knee extension, biceps curl, or isometric torque measures. This is considered low to very low-quality data.

So, I don’t dismiss the idea that creatine may help older adults who are doing resistance training, but let’s not overstate the case. Creatine may have a modest effect, based on a thin, low-quality evidence base. Consistent training and adequate dietary protein intake are the interventions that do the real heavy lifting (pun intended).

The Bigger Concern: Safety, Cost, and Distraction

The bigger problem with supplements is the unknown safety of what you actually buy on the shelf or online, together with the expense and the distraction from things that really work, which might not be as easy to implement as swallowing a pill.

If you do decide to take a supplement, look for third-party product verification, such as NSF International (formerly the National Sanitation Foundation) or USP (United States Pharmacopeia) certification. These programs test the finished product to confirm that it contains what the label says, in identity and in potency (8,9). Both screen for heavy metals, microbes, and certain toxins, and both audit the manufacturing facility for GMP (Good Manufacturing Practices) compliance. NSF “Certified for Sport” adds screening for substances banned in competitive athletics, which is worth knowing if you are subject to drug testing (8,9).

But you must understand what these certifications do not tell you. They are voluntary and paid for by the manufacturers. They test batches rather than every bottle, and they test periodically rather than continuously, so batch-to-batch variation between audits goes unmonitored. They confirm identity and potency, but they do not verify efficacy or safety claims.

And like any analytical screen, they test against a known panel of contaminants, which is useful for catching established problems, but not a guarantee against a novel adulterant that no one is looking for yet (9). What you get is a somewhat lower risk of gross contamination or outright mislabeling. They are not, however, the equivalent of FDA oversight of a prescription drug, and no seal on a bottle should be read that way.

So my answer to “what about everything else?” is close to my answer about heart disease supplements. Most - nearly all - are unnecessary or unhelpful and some are harmful. But if you still choose to take one, remember that the seal on the label is modest protection against contamination, nothing more. Supplements remain an expensive distraction from the things that truly work and are proven safe.

References

1. Geller AI, Shehab N, Weidle NJ, et al. Emergency department visits for adverse events related to dietary supplements. N Engl J Med. 2015;373(16):1531-1540. doi:10.1056/NEJMsa1504267.

2. Tucker J, Fischer T, Upjohn L, et al. Unapproved Pharmaceutical Ingredients Included in Dietary Supplements Associated With US Food and Drug Administration Warnings. JAMA Netw Open 2018;1:e183337. doi:10.1001/jamanetworkopen.2018.3337.

3. de Souza E Silva A, Pertille A, Reis Barbosa CG, et al. Effects of creatine supplementation on renal function: a systematic review and meta-analysis. J Ren Nutr 2019;29(6):480-489. doi:10.1053/j.jrn.2019.05.004.

4. Naeini EK, Eskandari M, Mortazavi M, et al. Effect of creatine supplementation on kidney function: a systematic review and meta-analysis. BMC Nephrol 2025;26:622. doi:10.1186/s12882-025-04558-6.

5. Kreider RB, Kalman DS, Antonio J, et al. International Society of Sports Nutrition position stand: safety and efficacy of creatine supplementation in exercise, sport, and medicine. J Int Soc Sports Nutr 2017;14:18. doi:10.1186/s12970-017-0173-z.

6. Devries MC, Phillips SM. Creatine supplementation during resistance training in older adults: a meta-analysis. Med Sci Sports Exerc 2014;46(6):1194-1203. doi:10.1249/MSS.0000000000000220.

7. Chilibeck PD, Kaviani M, Candow DG, et al. Effect of creatine supplementation during resistance training on lean tissue mass and muscular strength in older adults: a meta-analysis. Open Access J Sports Med 2017;8:213-226. doi:10.2147/OAJSM.S123529.

8. National Institutes of Health, Office of Dietary Supplements. Dietary Supplements: What You Need to Know. https://ods.od.nih.gov/factsheets/WYNTK-Consumer/. Accessed July 13, 2026.

9. Akabas SR, Vannice G, Atwater JB, et al. Quality Certification Programs for Dietary Supplements. J Acad Nutr Diet 2016;116:1370-1379. doi:10.1016/j.jand.2015.11.003.

Here are three more questions I hear almost every week in clinic, and how I answer them:

1. What supplements should I be taking?

2. Do I really need to see the dietitian?

3. How do I start exercising?

What supplements should I be taking?

Despite aisles of supplements in pharmacies and grocery stores, and constant social media promises of longevity and freedom from heart disease (including endorsements from doctors, real and otherwise), no supplements are proven to reduce cardiovascular disease (CVD). And this is not one of those situations where you can say "Lack of proof is not proof of lack," because there is extensive proof of lack.

No dietary supplements have been demonstrated to consistently reduce the risk of myocardial infarction, stroke, heart failure, or death. That is the reason a European cardiologist once told me that “Americans have the most expensive urine in the world.”

In the 1990s, antioxidant vitamins were widely promoted for atherosclerosis prevention until randomized clinical trials showed that they did not reduce CVD progression (1) and subsequent meta-analyses suggested that vitamin E supplementation increased overall mortality in men (2). Also in vogue were supplements intended to lower homocysteine levels, including folic acid, vitamin B6, vitamin B12, and betaine. High homocysteine levels have been associated with heart disease, but randomized trials demonstrated that lowering homocysteine with these supplements did not reduce CVD events, even among individuals with elevated levels (3).

More recently, a wide range of supplements have been promoted, including coenzyme Q10, omega-3 fatty acids, berberine, garlic supplements, magnesium, vitamin D, and many others. None convincingly have been shown to prevent CVD events in randomized clinical trials. I reviewed several of them in previous Substack posts.

One reason these claims keep recurring is that observational studies often show associations between nutrient levels and health outcomes. However, observational studies are susceptible to confounding by numerous factors, including lifestyle and health behaviors. When these hypotheses are tested in randomized trials, the benefits nearly always disappear. Vitamin D is a common example. Observational studies often link low vitamin D levels with higher CVD risk, but randomized trials have not shown that supplementation reduces CVD events (4). Vitamin D levels are strongly influenced by lifestyle, social determinants of health, nutrition, time spent outdoors, and other factors, which likely explain its associations.

So, I do not recommend supplements for CVD prevention. Instead, I recommend a healthy lifestyle with regular physical activity and a diet that is rich in vegetables, legumes, fruits, nuts, fish, and low-fat dairy, with limited intake of red meat and highly processed foods.

One partial exception is fiber. Many Americans consume far less fiber than recommended because our diets, on average, are high in processed foods, which typically are low in fiber. Increasing fiber intake can benefit both cardiovascular and gastrointestinal health. Soluble fiber, found in foods such as oats, beans, and psyllium, lowers LDL cholesterol by binding bile acids in the intestine. Insoluble fiber, found in many vegetables, legumes, and whole grains, increases stool bulk and promotes regular bowel movements. Ideally, fiber should come from whole foods, but for some patients a soluble fiber supplement such as psyllium can be helpful.

Do I really need to see the dietitian?

Yes, you probably do. Over 70% of Americans are overweight, which leads to several CVD risk factors, so those patients can benefit from meeting with a trained dietitian. Dietitians usually have deeper and more current insights into nutrition, its effects on human health, motivation, and the particular challenges we face in our toxic food environment than physicians.

But dietitians are not just for people who are overweight or who have heart disease risk factors. A dietitian can help identify less obvious sources of sodium, saturated fat, or highly processed foods, even in someone who already follows an excellent diet and has optimal CVD risk factors. I particularly favor referring patients to dietitians with expertise in preventive cardiology. Furthermore, individuals who are underweight may benefit from evaluation of nutritional adequacy, including sufficient protein and micronutrient intake.

Structured nutrition counseling has been shown to improve weight, blood pressure, and glycemic control, and dietary patterns such as the DASH and Mediterranean diets reduce CVD risk factors and events (5,6). Dietitians also are trained in motivational interviewing and in recognizing eating disorders, which are more common than many clinicians realize.

How do I start exercising?

This can be a difficult question to answer. For sedentary patients, I start by asking what activities they enjoy doing and focus on getting them moving. Identifying one or two activities that someone enjoys makes it much more likely they will continue doing them. Indeed, the largest health impact of exercise comes from helping sedentary patients start moving (7).

As a general target, I suggest about 30 to 60 minutes of aerobic exercise most days of the week, with a minimum goal of 150 minutes per week for cardiovascular conditioning and closer to 250 to 300 minutes per week for weight loss. I tell patients to exercise to the point where they sweat and feel short of breath so that they can talk but would not be able to sing (7). I recommend scheduling exercise on their calendar, ideally earlier in the day before competing demands accumulate. Furthermore, I emphasize that variety helps reduce boredom and lowers the risk of overuse injuries.

I also emphasize the importance of resistance training (8). In general, I recommend that patients participate in aerobic exercise three to four days per week, with resistance training about twice per week, but that is flexible based on patient preferences and needs. Patients should begin gradually with low resistance and increase slowly. I often recommend working with a personal trainer, at least initially, to improve accountability, optimize form, and reduce the risk of injury.

Walking, cycling, swimming, or similar activities need not be complicated and are excellent starting points. Resistance training can include traditional weights as well as activities such as yoga or Pilates. I also remind patients that exercise can cause muscle soreness, which usually is unrelated to their statin.

Total daily movement is important as well. Walking more, taking the stairs, and reducing prolonged sitting all contribute meaningfully to cardiovascular health, but they do not replace formal exercise, and too many people think that because they "move a lot" they do not need to exercise more. That is not true. For people who cannot exercise easily because of arthritis, chronic illness, or physical limitations, the goal shifts to increasing movement within safe limits, sometimes with guidance from physical therapists or exercise specialists.

Finally, people often misjudge their fitness and activity levels. Self-reported activity frequently is inaccurate; that is why some experts suggest measuring it directly (9). A treadmill or bicycle stress test can be useful to characterize functional capacity and to guide an exercise prescription. Because it can be hard to obtain a stress test for this purpose in some institutions, I sometimes recommend cardiopulmonary exercise testing through a commercial company to assess aerobic capacity and to determine individualized training zones for improving fitness.

References:

1. Brown BG, et al. Simvastatin and niacin, antioxidant vitamins, or the combination for the prevention of coronary disease. N Engl J Med 2001;345:1583-92. doi: 10.1056/NEJMoa011090.

2. Miller ER 3rd, et al. Meta-analysis: high-dosage vitamin E supplementation may increase all-cause mortality. Ann Intern Med. 2005;142:37-46. doi: 10.7326/0003-4819-142-1-200501040-00110.

3. Clarke R, et al. Effects of lowering homocysteine levels with B vitamins on cardiovascular disease, cancer, and cause-specific mortality: Meta-analysis of 8 randomized trials involving 37 485 individuals. Arch Intern Med 2010;170:1622-31. doi: 10.1001/archinternmed.2010.348.

4. Barbarawi M, et al. Vitamin D Supplementation and Cardiovascular Disease Risks in More Than 83 000 Individuals in 21 Randomized Clinical Trials: A Meta-analysis. JAMA Cardiol 2019;4:765-776. doi: 10.1001/jamacardio.2019.1870.

5. Dudzik JM, et al. The effectiveness of medical nutrition therapy provided by a dietitian in adults with prediabetes: a systematic review and meta-analysis. Am J Clin Nutr 2023;118:892-910. doi: 10.1016/j.ajcnut.2023.08.022.

6. Senkus KE, et al. Medical nutrition therapy provided by a dietitian improves outcomes in adults with prehypertension or hypertension: a systematic review and meta-analysis. Am J Clin Nutr 2024;119:1417-1442. doi: 10.1016/j.ajcnut.2024.04.012.

7. Piercy KL, et al. The Physical Activity Guidelines for Americans. JAMA 2018; 320:2020-2028. doi: 10.1001/jama.2018.14854.

8. Paluch AE, et al. Resistance Exercise Training in Individuals With and Without Cardiovascular Disease: 2023 Update: A Scientific Statement From the American Heart Association. Circulation 2024;149:e217-e231. doi: 10.1161/CIR.0000000000001189.

9. Ross R, et al. Importance of Assessing Cardiorespiratory Fitness in Clinical Practice: A Case for Fitness as a Clinical Vital Sign: A Scientific Statement From the American Heart Association. Circulation 2016;134:e653-e699. doi: 10.1161/CIR.0000000000000461.

The risk prediction part is nearly impossible at the level that matters to a patient, which is whether a particular person will have an event. The best we can do is sort people broadly into loose buckets of cardiovascular disease (CVD) risk, and even those buckets are wide and uncertain. We cannot tell an individual that they will have a heart attack, only that people who resemble them have CVD events at some rate over some time period.

The risk reduction part is where the misleading numbers get quoted. An intervention lowers average risk by some relative proportion, usually around 25% to 33% for many preventive therapies. That is a relative reduction, and keeping the difference between relative and absolute risk reduction in mind is where the whole discussion can succeed or fail. If one applies a 30% relative risk reduction to a person with a genuinely high predicted risk, say a 20% to 30% chance of an adverse event over ten years, then the absolute benefit over that ten-year window comes out to 6% to 9%. And even that figure understates the value of treatment, because risk reduction keeps accruing in the years beyond the window during which it was measured.

Sadly, many people respond to the arithmetic by declaring that over 90% of those treated get no benefit, and they mean it as an indictment of preventive care. But that claim is a retrospective count dressed up as a prospective fact. After the fact, sure, you can identify who had events and who did not. But the decision to treat is made a priori, when everyone carries the same lowered probability and no one’s outcome is known. The probability shift is a property of each person’s situation at the moment of treatment, not a lottery ticket that only the eventual event-havers cash. Counting the untouched afterward and calling the treatment wasted confuses the two frames.

It is the same error as reading a weather forecast backward. A 30% chance of rain was not wrong because the day stayed dry. The probability was a real statement about a distribution of possible days, made before you knew which one you would get. The therapy lowered everyone’s probability at the moment of the decision, and that reduction was real for each person even though only some were ever going to have the event it prevents.

This holds across preventive care, not only for statins, but also for blood pressure lowering and anticoagulation in atrial fibrillation, where the same objection surfaces in the same form and fails for the same reason. In each case, the intervention shifts a probability for everyone treated, and in each case only a minority were ever destined to have the event that the shift prevents.

None of this makes the decision automatic. The benefit is a probability shift, but so are the harms, such as bleeding on an anticoagulant, or the side effects, cost, and daily burden of any of these therapies. When the absolute benefit is modest, whether it justifies those harms is a genuine question, and the answer depends on what the individual values. That is what makes these decisions preference-sensitive rather than automatic. And preferences in this setting are more than tastes, like “I prefer vanilla over chocolate.” They are real risk assessment inputs, because the probabilities of benefit and harm are close enough that individual values legitimately tip the balance.

But the critical point is that the reduction in risk is real for every treated person at the moment of the decision, whatever the eventual outcome, because that is when the probability is shifted and that is when the choice is made. Treating that shift as nothing, on the grounds that most treated people would have done fine anyway, reads the forecast backward.

As expected, my recent posts on statins received some pushback. Some comments were outrageous, like one guy who said “Anyone who prescribes a statin should be put in front of a firing squad” (I blocked him). But others raised thoughtful concerns about the magnitude of statin benefits and in particular, the mortality benefit of primary prevention. The pushback concerns the people who are in patient Group 4 - typical 40-80 year-old adults without known cardiovascular disease (CVD), extensive subclinical atherosclerosis, type II diabetes mellitus, or very high cholesterol (i.e., LDL-C >190 mg/dL). That is, most adults who come to see their doctor for a wellness checkup. This is the patient group in whom a candid clinician-patient discussion is vital, because the decision to initiate statin therapy is inherently preference-sensitive and should be guided by the patient’s values, risk tolerance, and goals, not reflexively dictated by risk calculators or guidelines.

But “preference-sensitive” does not mean “ineffective” - far from it. It means the absolute benefit at low baseline risk is modest enough that, for some patients, it does not outweigh their concerns about cost, real or perceived risks (discussed in my previous posts here and here), the disutility of a daily pill, or the burden of being a patient (i.e., routine office visits, blood draws, refills, etc.), particularly when lifestyle change is carrying much of the health burden. That is a values judgment, and it belongs to the patient. But it is not evidence that statins are ineffective and it would be incorrect to even imply that.

The case for primary prevention does not rest on a mortality benefit

For starters, patients care about more than whether a pill postpones death. Statins powerfully reduce heart attacks, ischemic strokes, and revascularization procedures (including surgeries) (1). A nonfatal MI means an emergency hospital admission, often a percutaneous coronary intervention or coronary artery bypass graft surgery, lasting myocardial damage, increased heart failure risk, a bad diagnosis, and the oodles of medication and stigma that accompany it. A nonfatal stroke can mean living with permanent disability and lost independence. I have cared for such patients for over three decades and seen how these nonfatal events damage people’s lives and well-being. Ample evidence shows that the downstream financial cost of these events is substantial. Although some may glibly claim that all that matters is mortality, I personally am more afraid of a stroke than sudden cardiac death. My preference.

Degree of LDL-C lowering affects risk reduction

The dose-response relationship also is important to consider. The Cholesterol Treatment Trialists’ Collaboration found about a 21% proportional reduction in major vascular events for every 38.7 mg/dL reduction in LDL-C (2,3). A 2026 meta-analysis of primary prevention studies (some using ezetimibe or bempedoic acid) found a 30% reduction (95% CI 26%-33%) per 38.7 mg/dL for the composite of coronary heart disease death, nonfatal myocardial infarction, fatal and nonfatal stroke, and coronary revascularization (4). These estimates differ partly because the endpoints and populations differ, but the primary prevention estimate is the larger of the two. Greater and more sustained LDL-C reduction produces greater benefit, which is why this is a scalable intervention being judged by trials far shorter than the horizon over which people actually live.

The mortality benefit

And statins do have a mortality benefit. The USPSTF 2022 systematic review of 18 trials (n>85,000 participants) found a significant reduction in all-cause mortality (RR 0.92, 95% CI 0.87 to 0.98, I² = 0%) (1). Furthermore, a meta-analysis of 40 trials (n>94,000) found an 11% reduction in all-cause mortality and a 20% reduction in CVD mortality (5). WOSCOPS, the landmark primary prevention trial, showed that five years of pravastatin produced a legacy effect, with a 13% reduction in all-cause mortality and a 21% reduction in CVD death that persisted fifteen years after the trial ended (6). A meta-analysis of time to benefit found that the absolute risk reduction for major CVD events grows from roughly 0.3 events per 100 people at one year to about 2.5 per 100 at five years (7), so a three- to five-year trial systematically understates the lifetime benefit.

JUPITER was a large trial of rosuvastatin in people with normal LDL-C and hsCRP of 2.0 mg/L or higher (8). About half of US adults have an hsCRP at or above that level (9). It was stopped early after a 44% reduction in major CVD events and a significant 20% reduction in all-cause mortality (8). Stopping a trial early can inflate outcome estimates and the trial’s other limitations have been discussed, but HOPE-3, ASCOT-LLA, and MEGA are other large primary prevention trials that point in the same direction and have been pooled in systematic reviews and meta-analyses (1,4,5).

How some people think about the stochastic benefit of preventive therapy

Finally, a critical point about the mortality benefit with statins is that looking at averages over time flattens the fact that statin benefits are stochastic. In a brilliant study, UK investigators developed a method to evaluate the probability distribution of lifespan gain in primary prevention users, and then surveyed 396 young to middle-aged people to assess how they evaluated the potential benefit of preventive therapy (10). They showed from lifetables that: (i) younger people gain the most lifespan from initiation of primary prevention therapy, (ii) a vast majority gain no lifespan, but (iii) those who do gain lifespan gain much more than the group average, and also showed that (iv) when presented with probabilistic information, many individuals have a personal preference for certainty of a small gain or for a chance of a large gain, not necessarily the most likely outcome (10). To make that granular, they presented the example in the image below:

For a 50-year-old man with average CVD risk taking a prevention intervention (like a moderate-intensity statin) expected to lower that risk by 30%, the average lifespan gain was estimated to be 7 months, but 93% would not gain any lifespan and 7% would gain an average of 8.25 years. One-third of people preferred a 2% probability of a 10-year gain to the certainty of one year. These data have profound implications for shared decision-making.

Conclusion

None of this decides the question for a typical primary prevention patient, and it is not meant to; it only settles the premise. In primary prevention, statins are effective at preventing important CVD events. However, the absolute benefit - particularly for all-cause or CVD mortality - is modest in those at low baseline risk of CVD. An honest conversation with the patient includes both facts. How one weighs reducing nonfatal events against the disutility of treatment is preference-sensitive. The patient’s values resolve it, once the evidence has been stated accurately rather than dismissed.

References

1. Chou R, Cantor A, Dana T, et al. Statin Use for the Primary Prevention of Cardiovascular Disease in Adults: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA 2022;328:754-771. doi:10.1001/jama.2022.12138

2. Mihaylova B, Emberson J, Blackwell L, et al; Cholesterol Treatment Trialists’ (CTT) Collaboration. The Effects of Lowering LDL Cholesterol With Statin Therapy in People at Low Risk of Vascular Disease: Meta-Analysis of Individual Data From 27 Randomised Trials. Lancet 2012;380:581-590. doi:10.1016/S0140-6736(12)60367-5

3. Collins R, Reith C, Emberson J, et al. Interpretation of the Evidence for the Efficacy and Safety of Statin Therapy. Lancet 2016;388:2532-2561. doi:10.1016/S0140-6736(16)31357-5

4. Kalra DK, Ray KK, Bajaj A, et al. Low-Density Lipoprotein Cholesterol Lowering and Risk of Major Adverse Cardiovascular Events in Primary Prevention Trials: A Meta-Analysis. J Clin Lipidol 2026;20:738-749. doi:10.1016/j.jacl.2026.02.006

5. Yebyo HG, Aschmann HE, Kaufmann M, et al. Comparative Effectiveness and Safety of Statins as a Class and of Specific Statins for Primary Prevention of Cardiovascular Disease: A Systematic Review, Meta-Analysis, and Network Meta-Analysis of Randomized Trials With 94,283 Participants. Am Heart J 2019;210:18-28. doi:10.1016/j.ahj.2018.12.007

6. Ford I, Murray H, McCowan C, et al. Long-Term Safety and Efficacy of Lowering Low-Density Lipoprotein Cholesterol With Statin Therapy: 20-Year Follow-Up of West of Scotland Coronary Prevention Study. Circulation 2016;133:1073-1080. doi:10.1161/CIRCULATIONAHA.115.019014

7. Yourman LC, Cenzer IS, Boscardin WJ, et al. Evaluation of Time to Benefit of Statins for the Primary Prevention of Cardiovascular Events in Adults Aged 50 to 75 Years: A Meta-analysis. JAMA Intern Med 2021;181:179-185. doi:10.1001/jamainternmed.2020.6084

8. Ridker PM, Danielson E, Fonseca FAH, et al. Rosuvastatin to Prevent Vascular Events in Men and Women With Elevated C-Reactive Protein. N Engl J Med 2008;359:2195-2207. doi:10.1056/NEJMoa0807646

9. Mensah GA, Arnold N, Prabhu SD, et al. Inflammation and Cardiovascular Disease: 2025 ACC Scientific Statement: A Report of the American College of Cardiology. J Am Coll Cardiol 2026;87:1381-1404. doi:10.1016/j.jacc.2025.08.047

10. Finegold JA, Shun-Shin MJ, Cole GD, et al. Distribution of Lifespan Gain From Primary Prevention Intervention. Open Heart 2016;3:e000343. doi: 0.1136/openhrt-2015-000343

This week I will address additional concerns patients ask me about statins and their possible side effects concerns:

• Won’t this give me diabetes?
• Doesn’t this medicine cause brain fog?
• Is this going to damage my liver?
• Should I take coenzyme Q10?

Won’t this give me diabetes?

Statins can cause a very small increase in blood sugar levels. When a patient’s fasting glucose level rises above 126 mg/dL, we diagnose diabetes mellitus. But there are three key points.

First, the excess risk is extremely small, so small that the signal is detected only in very large studies with thousands of people in each arm. In individual patients, there is no detectable signal because the change is within normal human and lab test variation. We are talking about a very small average increase in hemoglobin A1c (a measure of glucose control over the past 90-120 days) of about 0.06-0.08% (1).

Second, the effect on diabetes diagnosis mostly is seen in people who already are close to the diagnostic threshold. In other words, people who have prediabetes or significant insulin resistance may be nudged across the line by a very small increase in glucose.

Third, and perhaps most important, people with diabetes or prediabetes derive substantial cardiovascular disease (CVD) and mortality benefit from using statins (2). In the TNT study and other analyses, those with higher baseline metabolic risk (such as diabetes or pre-diabetes) derived the largest absolute benefits in CVD event reduction (3).

So the take-home message is this: the increase in blood sugar is too small to detect in an individual patient, and if it occurs, it tends to identify someone who stands to gain the most from therapy. As part of routine care, I monitor fasting glucose and hemoglobin A1c, as indicated. The best way to prevent diabetes is achieving an optimal weight, getting regular exercise, and following a healthy diet. Some medications can raise blood sugar levels and contribute to diabetes mellitus risk in a clinically meaningful way, but not statins, whose benefits vastly outweigh any possible risks.

Doesn’t this medicine cause brain fog?

Despite anecdotal reports of “brain fog” with statins, large randomized clinical trials and systematic reviews have not demonstrated even the slightest signal that statins cause cognitive impairment or decline. In fact, by reducing stroke and atherosclerosis of the carotid and intracranial arteries, statins are associated with lower rates of vascular cognitive impairment.

Earlier in my career, I conducted studies using pravastatin and simvastatin in adults over age 65, assessing multiple cognitive domains (4). We found no evidence of cognitive harm. Subsequent larger studies have shown neutral effects, with some suggesting mild benefit (5). Some statins such as rosuvastatin and pravastatin don’t even get across the blood–brain barrier.

Beyond education, I reassure my patients that if they are not feeling well, I want to know because I am their physician. But statins do not cause brain fog or progressive cognitive decline.

Is this going to cause liver damage?

That is a reasonable question since we check liver function tests after starting statins. What are we so worried about, and is it warranted? Back in the 1990s and early 2000s, we worried about possible statin-associated liver issues, but over time, we have learned that statins are not hepatotoxic in the way we once feared. It turns out that mild elevations in transaminases (certain liver function tests) are common in people with lipid disorders and often fluctuate due to hepatic steatosis (fat in the liver). Transaminases also can rise slightly when LDL cholesterol levels fall substantially, but clinically significant liver injury from statins is rare (<1 in 100,000) (6).

After more than 38 years on the market, there is no evidence that statins cause chronic liver damage in routine use. For that reason, routine periodic liver enzyme monitoring beyond baseline testing no longer is recommended in most guidelines. Indeed, hepatology societies support the use of statins in patients with mild to moderate transaminase elevations and in those at risk for metabolic-associated steatotic liver disease (“fatty liver”), because statins reduce CVD events and overall mortality in this population (7). It simply is not something we worry about in modern practice. We monitor to detect associated problems, not problems caused by the statins themselves.

Should I take coenzyme Q10?

In general, I do not recommend taking coenzyme Q10 (CoQ10). CoQ10, also known as “ubiquinone,” is an essential cofactor in mitochondrial energetics. It was hypothesized that among people taking statins, lower levels of CoQ10 might contribute to muscle symptoms. Indeed, coenzyme Q10 is carried in LDL particles (8). However, studies examining muscle CoQ10 levels in people taking statins have produced conflicting results. The LIFESTAT study found that muscle coenzyme Q10 levels were unaltered even though mitochondrial respiration decreased (9). Thus, the entire mechanistic rationale for supplementation may not hold up.

Furthermore, studies examining the effects of CoQ10 supplementation on statin-associated muscle symptoms have shown mixed and generally weak results (8). That is because many reported statin-associated muscle symptoms are driven by the nocebo effect, as I reviewed previously. Overall, if CoQ10 helps some patients, it likely is working through a placebo mechanism.

Having said all that, coenzyme Q10 seems benign. I am not enthusiastic about adding supplements or extra pills, but if a patient really wants to take it, I tell them that 100 to 200 mg once or twice daily is reasonable (8). But to be clear: I do not recommend it routinely and I make sure patients understand that the evidence does not support CoQ10 use as a standard treatment.

References:

1. Cholesterol Treatment Trialists’ (CTT) Collaboration. Effects of statin therapy on diagnoses of new-onset diabetes and worsening glycaemia in large-scale randomised blinded statin trials: an individual participant data meta-analysis. Lancet Diabetes Endocrinol 2024;12:306-319. doi: 10.1016/S2213-8587(24)00040-8.

2. Cholesterol Treatment Trialists’ (CTT) Collaborators. Efficacy of cholesterol-lowering therapy in 18,686 people with diabetes in 14 randomised trials of statins: a meta-analysis. Lancet 2008; 371:117-25. doi: 10.1016/S0140-6736(08)60104-X.

3. Deedwania P, et al. Reduction of low-density lipoprotein cholesterol in patients with coronary heart disease and metabolic syndrome: analysis of the Treating to New Targets study. Lancet. 2006 Sep 9;368(9539):919-28. doi: 10.1016/S0140-6736(06)69292-1.

4. Carlsson CM, et al. Health-related quality of life and long-term therapy with pravastatin and tocopherol (vitamin E) in older adults. Drugs and Aging 2002;19:793-805.

5. Goldstein LB, et al. Aggressive LDL-C lowering and the brain: Impact on risk for dementia and hemorrhagic stroke: A scientific statement from the American Heart Association. Arterioscler Thromb Vasc Biol 2023;43:e404-e442. doi: 10.1161/ATV.0000000000000164.

6. Newman CB, et al. Statin safety and associated adverse events: A scientific statement from the American Heart Association. Arterioscler Thromb Vasc Biol 2019;39:e38-e81. doi: 10.1161/ATV.0000000000000073.

7. Rinella ME, et al. AASLD Practice Guidance on the clinical assessment and management of nonalcoholic fatty liver disease. Hepatology 2023;77:1797-1835. doi: 10.1097/HEP.0000000000000323.

8. Raizner AE, et al. Coenzyme Q10 for patients with cardiovascular disease: JACC Focus Seminar. J Am Coll Cardiol 2021;77:609-619. doi: 10.1016/j.jacc.2020.12.009.

9. Dohlmann TL, et al. Statin Treatment Decreases Mitochondrial Respiration But Muscle Coenzyme Q10 Levels Are Unaltered: The LIFESTAT Study. J Clin Endocrinol Metab 2019;104:2501-2508. doi: 10.1210/jc.2018-01185.

Testosterone replacement therapy (TRT), promoted as a medical shortcut to renewed vitality, is surging in popularity (1). What once was confined to specialty clinics has entered everyday practice. Health podcasts promote it, patients arrive at my clinic on it, and I even was advised to consider it for my sore muscles, despite my personal history of prostate cancer. As a cardiologist, I worry that its risks outweigh its benefits.

For more than two decades, wellness clinics, online communities, and even some physicians and pharmaceutical companies have marketed TRT as a quick fix for vague symptoms such as fatigue, irritability, low mood, or decreased sexual interest. The phrase “Low T” has become a marketing slogan and an all-purpose explanation for aging-related complaints. This blurring of lines between true hypogonadism (a real medical disease) and normal aging has driven a surge in prescriptions.

Testosterone as Legitimate Therapy

Testosterone is a powerful prescription drug with specific, evidence-based indications. It is legitimate therapy for men with well-documented hypogonadism, a condition defined by both symptoms and consistently low serum testosterone levels. Classic causes include pituitary disease, testicular injury, or genetic syndromes. In such men, TRT can restore sexual function, improve bone density, and enhance quality of life. These are real benefits, supported by decades of clinical experience and randomized trial evidence.

But that is not how testosterone is typically being used in 2026. Instead, it is increasingly marketed and prescribed as a lifestyle drug for fatigue, sexual concerns, muscle-building, the vague promise of “anti-aging,” or the particularly odious construct of “biohacking.” When I caution against treating testosterone like a toy, this is what I mean: it is not a supplement, a shortcut, or a harmless biohack. It is a powerful hormone and at supraphysiologic doses - the kind taken by young men at gyms to “get shredded” or by biohackers experimenting with hormones - it is even more troubling.

These men are exposing themselves to all the risks of TRT plus additional dangers from higher doses, non-medical formulations, and lack of clinical supervision. Specific concerns include even higher risks of CVD, polycythemia, and blood clots, as well as infertility (due to reduced sperm count and induced hypogonadism), neuropsychiatric complications (such as mood swings, “roid rage,” and dependence), and liver tumors. Claims of safety often rely on selective readings of short, methodologically weak studies.

Inappropriate Prescribing and Overuse

Despite strict diagnostic criteria, testosterone prescribing has risen dramatically in middle-aged and older men, fueled by lifestyle marketing and direct-to-consumer advertising. Sadly, many men receiving therapy never had properly documented hypogonadism. In a study of more than 61,000 men aged 40 years and older, over one-quarter never even had their serum testosterone level measured before treatment (2). This widespread use falls far outside evidence-based medicine.

Prescribing it to men with a single low testosterone level is similarly inappropriate. The American Urological Association 2024 guideline explicitly states that a total testosterone <300 ng/dL alone does not define testosterone deficiency, and that making a diagnosis in the absence of signs and/or symptoms increases the likelihood of a false diagnosis and reduces the potential benefit of therapy (3). The Endocrine Society similarly recommends testosterone therapy only for men with symptomatic testosterone deficiency and that clinicians should refrain from measuring testosterone levels in patients who are asymptomatic and do not exhibit signs related to low testosterone (4). In fact, about 30% of men with an initial testosterone concentration in the hypogonadal range will have a normal level on repeat measurement, so it needs to be measured twice, while fasting, between 7-10 AM, after adequate sleep, and in the absence of acute illness, since all of these factors can transiently lower levels (4).

Such prescribing might be forgivable if TRT were perfectly safe, but it is not. As a cardiologist, I am most attuned to the risks of cardiovascular disease (CVD).

Cardiovascular Disease Risks of TRT

In 2025, the New England Journal of Medicine published a review suggesting that TRT appears safe from a CVD standpoint (5). But the idea that TRT has been “proven safe” is false and largely based on spin from an over-hyped study called the TRAVERSE Trial (6). Before diving into it, two basic truths of medical research are worth recalling:

1. No study exists in a vacuum. We interpret new results through the lens of prior evidence and biological plausibility.

2. All studies have limitations. Most clinical research trials enroll highly selected participants and monitor them far more carefully than happens in real practice. That can make their results difficult to generalize.

Pharmaceutical industry sponsorship further complicates interpretation, as company-sponsored studies are more likely to report favorable results.

Biologically, we have long known that testosterone causes erythrocytosis (higher red blood cell counts), increases risk of venous blood clots, and lowers HDL cholesterol levels. Before TRAVERSE, randomized trial evidence was mixed but concerning: A meta-analysis of 27 randomized trials including nearly 3,000 men found 54% higher odds of CVD-related events in men receiving testosterone compared with placebo, with stronger associations in non-industry-funded studies (7). The Testosterone Trials (8) were small and only followed patients for a year, so it is not surprising that they did not detect an excess of major adverse CVD events. But in an atherosclerosis imaging sub-study (9), researchers identified a clear, worrisome increase in coronary artery plaque volume on CT scans, including soft plaques which are more likely to cause heart attacks. Not all studies have shown adverse effects on surrogate markers of CVD risk, but taken together, these data suggested caution.

The TRAVERSE Trial: Too Many Limitations to Be Reassuring

TRAVERSE was a large, industry-sponsored study designed to evaluate the cardiovascular safety of TRT (6). It enrolled more than 5,200 men with confirmed hypogonadism and elevated CVD risk. On the surface, the results looked reassuring. But a closer look shows why it did not resolve the question of safety.

• Short treatment duration. The average testosterone exposure was only 21.7 months. Since men take TRT for many years, this is a grossly inadequate test of long-term safety.

• High discontinuation rates. More than 60% of participants stopped therapy, leaving little true TRT exposure.

• Very careful monitoring. Participants had frequent testosterone checks and dose adjustments to keep levels tightly controlled at what turned out to be low-normal serum levels of testosterone. In real life, follow-up is sporadic, adherence is looser, and patients often have much higher serum levels of testosterone, so the outcomes are not generalizable.

• Noninferiority design. The trial was structured so that even with a clinically meaningful increase in events, TRT could still be declared “not worse” than placebo. This is far from proof of safety.

But even under these idealized conditions, men in TRAVERSE on TRT experienced higher rates of atrial fibrillation, venous thromboembolism, and acute kidney injury. That is not a clean bill of cardiovascular health.

The TRAVERSE Trial: A Clinical Perspective

The TRAVERSE Trial cannot be read as a green light for routine testosterone use, especially in the real-world men I see every day, who are often older, overweight, or have diabetes mellitus or established CVD. For them, the risks are considerable. And most cases of “low T” I see are driven by obesity, which suppresses testosterone but is itself treatable. Weight loss improves testosterone levels, mood, energy, and sexual function with far fewer risks. Many of my preventive cardiology patients asking about TRT would do far better addressing lifestyle factors first.

As a prostate cancer survivor, I also am concerned about possible stimulation of occult cancer. While the evidence is debated, the uncertainty reinforces the need for caution.

Conclusion

Testosterone is a legitimate therapy for men with true hypogonadism. But it is not a lifestyle drug, a shortcut to vitality, or a safe experiment for aging men or young athletes, and it has not been proven safe outside of its medical indications. Clinicians must resist cultural and commercial pressures to medicalize aging and instead uphold evidence-based practice. And patients - whether in the clinic, at the gym, or listening to the latest podcast - deserve to hear that testosterone is a therapy, not a toy. It is a hormone with risks, not a hack, and treating it otherwise is dangerous.

References

1. US Pharm 2025;50:18. https://www.uspharmacist.com/article/nationwide-patterns-in-testosterone-replacement-therapy#

2. Baillargeon J, et al. Screening and Monitoring in Men Prescribed Testosterone Therapy in the U.S., 2001–2010. Public Health Rep 2015;130:142.

3. Mulhall JP, et al. Evaluation and management of testosterone deficiency: AUA guideline (2024 update). At https://www.auanet.org/guidelines-and-quality/guidelines/testosterone-deficiency-guideline. Update of original guideline published in J Urol 2018; 200:423.

4. Bhasin S, et al. Testosterone Therapy in men with hypogonadism: an Endocrine Society Clinical Practice Guideline. J Clin Endocrinol Metab 2018;103:1715-1744.

5. Bhasin S, et al. Testosterone Treatment in Middle-Aged and Older Men with Hypogonadism. N Engl J Med 2025;393:581.

6. Lincoff AM, et al for the TRAVERSE Investigators. Cardiovascular Safety of Testosterone-Replacement Therapy. N Engl J Med 2023;389:107.

7. Xu L, et al. Testosterone therapy and cardiovascular events among men: a systematic review and meta-analysis of placebo-controlled randomized trials BMC Med 2013;11:108.

8. Snyder PJ, et al. Effects of Testosterone Treatment in Older Men. N Engl J Med 2016;374:611.

9. Budoff MJ, et al. Testosterone Treatment and Coronary Artery Plaque Volume in Older Men With Low Testosterone. JAMA 2017;317:708.

Because my practice increasingly has focused on cardiovascular disease (CVD) prevention, I spend a lot of time talking about statins. This piece addresses three questions that come up in almost every one of my clinic sessions:

• Do I need to be on this medication for the rest of my life?
• Can my cholesterol get too low?
• Why can’t I just take red yeast rice instead?

Do I need to be on this medication for the rest of my life?

For patients with established atherosclerotic CVD, such as those with prior heart attack, stroke, coronary artery stent, or bypass surgery, the answer unequivocally is “yes.” Similarly, for patients with advanced subclinical atherosclerosis identified on imaging, the answer is “yes.”

The reason is straightforward. Numerous secondary prevention trials have shown major reductions in recurrent CVD events as well as survival benefits among patients with established atherosclerotic CVD who are treated with statins. In the Heart Protection Study, major vascular events were reduced from 25.2% to 19.8% over five years (1). The effects of statin therapy were even more dramatic in the 1,263 participants aged 75-80 years at entry: vascular event rates were reduced from 32.3% to 23.1%, an absolute reduction of 9.2% and a number needed to treat of approximately 11 over five years (1).

For standard primary prevention - what I called “group 4” in my previous essay on treatment of dyslipidemia - the honest answer is “most likely, yes.” But since that decision is preference-sensitive, it is more flexible. If a patient’s overall risk changes meaningfully, we can reconsider use of a statin and the dose. For example, if their diet changes dramatically and their LDL cholesterol (LDL-C) is well below goals, we can reassess, but that does not happen very often. Similarly, if they develop a comorbidity that is more likely to limit their lifespan than CVD, we can reduce the dose or stop therapy. Context and patient preference drive primary prevention decisions.

Can cholesterol get too low?

The practical answer is “no.” For starters, we do not really know what “too low” even means. When I was in training, if the total cholesterol was below 160 mg/dL, that sometimes was viewed as too low; similarly if the LDL-C was below 70-80 mg/dL. Today, those are targets we shoot for and in many people, we aim for even lower targets.

With use of more potent medications such as PCSK9 inhibitors, it is common to see total cholesterol levels under 100 mg/dL and LDL-C levels under 50 mg/dL, sometimes even under 20 mg/dL (though LDL-C estimates are less accurate when very low). Across large randomized trials and genetic studies, we do not see evidence of harm among people with very low LDL-C levels (2,3). There are extensive data showing that people with LDL-C levels below 20 mg/dL and total cholesterol levels between 70 and 100 mg/dL do well from a CVD standpoint.

But I don’t want to come across as too cavalier. If total cholesterol drops below 70 to 80 mg/dL on combination therapy, I am willing to cut back, not because we see a signal of harm, but because the incremental benefit may be small and the number of people who have lived for decades at those levels is not as vast as those who have had treatment that led to LDL-C values in the 40s and 50s mg/dL range.

Genetically very low LDL-C generally is safe and protective from atherosclerotic CVD (3). Individuals with ApoB, PCSK9, or ANGPTL3 loss-of-function mutations have lifelong LDL-C levels in the 10-40 mg/dL range and markedly lower CVD risk, with no consistent signal of excess cancer, neurocognitive disease, or hemorrhagic stroke in large human genetic datasets. (NB: A brief caveat for lipid “nerds” - familial hypobetalipoproteinemia from truncated apoB variants can be associated with hepatic steatosis, fat-soluble vitamin deficiency, and in severe biallelic forms, neurologic and malabsorptive syndromes. Those rare conditions are not relevant to clinically achieved LDL-C lowering)

I also remind patients that healthy infants are born with LDL-C levels between 30-70 mg/dL, which remain low during periods of maximal growth and development, further supporting the safety of very low LDL-C levels. Observationally, very low LDL-C has been linked to hemorrhagic stroke, but genetic data and PCSK9 inhibitor trials do not show a convincing causal signal at LDL-C levels achieved clinically. If a risk exists, it appears exceptionally small and context-dependent. The body maintains large cholesterol pools and synthesizes cholesterol as needed for cell membranes, steroid hormones, bile acids, and other essential functions.

Why can’t I just take red yeast rice instead?

Red yeast rice contains monacolin K, which is chemically identical to lovastatin (4). In other words, it is a low-potency statin. It is metabolized by CYP3A4 and has drug interaction potential, including with grapefruit juice, certain antibiotics, antivirals, and other medications. More importantly, it is not regulated by the FDA, like prescription medications. Research shows substantial variation in active ingredient content from lot to lot, which means patients experience variable and unpredictable LDL-C and total cholesterol reductions (5).

If you are willing to take red yeast rice, you should be willing to take a statin. It is safer, more predictable, and usually inexpensive to take a regulated, standardized statin tablet. There is nothing inherently more “natural” about red yeast rice than earlier-generation statins like pravastatin and lovastatin, some of which were originally derived from fungal compounds.

References:

1. Heart Protection Study Collaborative Group. MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin in 20,536 high-risk individuals: a randomised placebo-controlled trial. Lancet 2002;360:7-22. doi: 10.1016/S0140-6736(02)09327-3.

2. Giugliano RP, et al. Clinical efficacy and safety of achieving very low LDL-cholesterol concentrations with the PCSK9 inhibitor evolocumab: a prespecified secondary analysis of the FOURIER trial. Lancet 2017;390:1962-1971. doi: 10.1016/S0140-6736(17)32290-0.

3. Dron JS, et al. Association of Rare Protein-Truncating DNA Variants in APOB or PCSK9 With Low-density Lipoprotein Cholesterol Level and Risk of Coronary Heart Disease. JAMA Cardiol 2023;8:258-267. doi: 10.1001/jamacardio.2022.5271.

4. Dujovne CA. Red Yeast Rice Preparations: Are They Suitable Substitutions for Statins? Am J Med 2017;130:1148-1150. doi: 10.1016/j.amjmed.2017.05.013.

5. Gordon RY, et al. Marked variability of monacolin levels in commercial red yeast rice products: buyer beware! Arch Intern Med 2010;170:1722-7. doi: 10.1001/archinternmed.2010.382.

I gave up my pedometer a few weeks ago and am much happier than I expected. My pedometer lives in my iPhone, which was a problem. I have been trying to avoid being on my phone so much, but the only way I could see my step count was to pick it up. Smartphones distract people from living life and suck people into the alternate, unhealthy world of the internet and social media; that is what they are built to do. It turns out that the tool I was using to encourage physical activity was pulling me back into the sedentary habit I had been trying to limit, and that should have been reason enough to stop.

But following my steps also made my life worse in ways that, to be honest, are embarrassing for someone like me, who studies cardiovascular disease prevention and sees cardiology patients for a living. For example, sometimes I invented reasons to walk if the step count was a few hundred short of a round number, like 7,000, or a made-up goal, like 10,000. Since inadequate sleep is one of my personal health risks, it certainly did not make any sense to stay up later walking around the house to get in a few hundred more steps. And when I left my phone in the other room or in the car, I found myself counting steps in my head and using that tally to update the pedometer count, instead of simply being present in the world around me. That may sound obsessive, but I have friends and patients who do these sorts of things, too, because they think it is “healthy.”

Finally, I am an active person. I already know that I have a stable pattern of taking 6,000 to 7,000 steps a day, and I also cycle, swim, and do resistance training about five times a week. So I asked myself the obvious question: what was counting my steps actually doing for me?

How an ad slogan became a health prescription

The 10,000-step target has no physiological basis. The origin story is a Japanese pedometer from the mid-1960s called the manpo-kei, which loosely means “the ten-thousand-step meter.” It was a catchy round number attached to a consumer device, not a threshold discovered by exercise scientists. The number survived because it was memorable, specific, and easy to put on a screen. Over time, a marketing target came to feel like a biological one (1).

To be clear, this origin story does not mean walking is unimportant. Walking is among the safest and most useful things a person can do, and for some people, it is the only form of exercise they can do safely. But this story means that 10,000 steps is not a commandment or a physiological line between health and failure.

What the observational data show about people who walk more

Large observational cohorts consistently show that people who take more steps have lower mortality during follow-up. In a prospective cohort study of over 16,000 women, mortality declined as daily steps increased and appeared to level off around 7,500 steps per day (2). A meta-analysis of 15 cohorts found that the association flattened around 6,000 to 8,000 steps per day for adults 60 years and older and around 8,000 to 10,000 steps per day for younger adults (3). But the curve is steep at the bottom and flatter in the middle. Moving from 2,000 to 5,000 steps/day is associated with larger health improvements than moving from 8,000 to 11,000. That should be obvious physiologically, but the consumer version of pedometry hides it. The screen gives one number and the internet supplies one target.

There also is the unavoidable problem of reverse causation. Some of the association between step count and mortality almost certainly reflects the fact that people who are healthier are more able to walk. People who are ill, frail, depressed, or developing disease walk less. If an observational study sorts people by step count at baseline, the lowest step-count group will include more people whose low activity is a consequence or marker of illness, not necessarily the cause of future illness. People with higher step counts also may differ from people with lower step counts in many ways that are hard to measure, including baseline health, mobility, mood, social support, and access to safe places to walk. Researchers try to handle this by adjusting for known conditions and excluding early events; the associations often persist, albeit weaker, but observational data cannot fully separate walking as a cause of better health from walking as a sign of better health.

Pedometers can help inactive people walk more

In a meta-analysis of pedometer interventions to increase walking, pedometers worked. Pedometer users increased their steps by about 2,000 to 2,500 steps/day over 4 to 5 months (4). The subsequent PACE-UP trial showed that a pedometer-based walking program could produce modest increases in activity among people who were not meeting physical activity goals. The step-count increases declined by one year but still were present (5).

These studies suggest that for someone who is not very active, a pedometer can convert an abstract instruction like “move more” into something concrete and actionable. It can reveal that a day spent mostly sitting is even more sedentary than it felt. For that person, the number may be useful not because 10,000 steps is magic, but because feedback can change behavior. But the behavioral benefit seems to be strongest at the lower end of the activity distribution. Once a person has a generally active life, the marginal value of step counting gets much smaller.

What the number misses

It also is vital to remember that step counts are not the same as physical activity. Pedometers can make anyone who bikes, swims, rows, lifts weights, or does a lot of non-ambulatory work look mediocre (6). A swim and a strength session may leave a person with a disappointing step number, but a phone that bounced in the car all day can make them look awesome. The number on your phone feels precise, but it is not necessarily measuring the thing we actually care about. Furthermore, no large trial has randomized people to a specific daily step target and shown fewer cardiovascular disease events or deaths as a primary endpoint. The randomized trials show that pedometers can modestly increase activity and improve some cardiovascular disease risk factors over the short and medium term. But the hard-outcome data mostly come from observational associations, which are useful but clinically limited.

When tracking changes the behavior it is supposed to measure

I think the more interesting problem is not accuracy. It is what the number does to the person looking at it. Once movement becomes a score, it no longer is neutral; it can become a credit, a debt, or a judgment. I experienced my step count as yet another distracting demand on my attention and my time, but many others use it as license to change other behaviors. In the E-MECHANIC randomized trial, supervised exercise increased caloric intake in many participants and partly offset the expected weight-loss effect (7). That does not mean exercise causes overeating; it means that the human response to exercise is not straightforward: some people move more and then eat more, sit more, or treat activity as something they have earned the right to spend.

A different wearable trial gets even closer to my concern. Participants who were given artificially deflated step count feedback (i.e., they were told they had taken fewer steps than they actually had) felt less adequate and subsequently experienced worse affect, self-esteem, mental health, and blood pressure despite no actual reduction in physical activity (8). In other words, the number on the screen changed how people experienced their health, independent of what their bodies actually had done.

Neither finding proves that pedometers are bad, but they do illustrate the downsides of step counting. A device that makes an inactive person more active may be useful, but a device that makes an already active person feel inadequate, check their phone more often, sleep less, or convert movement into an accounting exercise may be doing something else.

Why I stopped counting

I am not against walking. I recommend it often and enjoy it myself. I am not against pedometers for people who are inactive, recovering from illness, or looking for a simple prompt to move more. But I am against confusing a metric with a goal (cf. Goodhart’s Law) and confusing achieving that goal with being healthy.

For me, step counting had stopped providing useful information. I already knew I was in a healthy range of daily walking and I exercised regularly. I already had the habits the pedometer was supposed to encourage, so instead of adding health, it added phone checking, mental arithmetic, sleep disruption, and a subtle sense that an active day was incomplete unless a device approved of it.

So I stopped, and the only thing I gave up was the pretense that a number generated by a 1960s marketing idea, made more persuasive by observational curves, and never validated as a universal treatment target was telling me something I needed to know. For some people, a pedometer is a useful tool. For me, it had become one more machine telling me to leave real life and return to the screen.

References

1. Tudor-Locke C, Bassett DR Jr. How many steps/day are enough? Preliminary pedometer indices for public health. Sports Med. 2004;34(1):1-8. doi:10.2165/00007256-200434010-00001

2. Lee I-M, Shiroma EJ, Kamada M, Bassett DR Jr, Matthews CE, Buring JE. Association of step volume and intensity with all-cause mortality in older women. JAMA Intern Med. 2019;179(8):1105-1112. doi:10.1001/jamainternmed.2019.0899

3. Paluch AE, Bajpai S, Bassett DR Jr, et al. Daily steps and all-cause mortality: a meta-analysis of 15 international cohorts. Lancet Public Health. 2022;7(3):e219-e228. doi:10.1016/S2468-2667(21)00302-9

4. Bravata DM, Smith-Spangler C, Sundaram V, et al. Using pedometers to increase physical activity and improve health: a systematic review. JAMA. 2007;298(19):2296-2304. doi:10.1001/jama.298.19.2296

5. Harris T, Kerry SM, Limb ES, et al. Effect of a primary care walking intervention with and without nurse support on physical activity levels in 45- to 75-year-olds: the PACE-UP cluster randomised clinical trial. PLoS Med. 2017;14(1):e1002210. doi:10.1371/journal.pmed.1002210

6. Strath SJ, Kaminsky LA, Ainsworth BE, et al. Guide to the assessment of physical activity: clinical and research applications: a scientific statement from the American Heart Association. Circulation. 2013;128(20):2259-2279. doi:10.1161/01.cir.0000435708.67487.da

7. Martin CK, Johnson WD, Myers CA, et al. Effect of different doses of supervised exercise on food intake, metabolism, and non-exercise physical activity: the E-MECHANIC randomized controlled trial. Am J Clin Nutr. 2019;110(3):583-592. doi:10.1093/ajcn/nqz054

8. Zahrt OH, Evans K, Murnane E, et al. Effects of wearable fitness trackers and activity adequacy mindsets on affect, behavior, and health: longitudinal randomized controlled trial. J Med Internet Res. 2023;25:e40529. doi:10.2196/40529

There is a particular kind of medical wishful thinking that is almost impossible to argue against in the exam room. A patient comes in having read about a new blood test that screens for fifty different cancers at once. Her neighbor had it done, and it identified pancreatic cancer before she had any symptoms or lab findings. Her brother-in-law had it done, got a clean result, and feels relieved. She wants it. The logic is just common sense, right? More testing means earlier detection means earlier treatment means longer life. What kind of doctor argues against that? Doctors like me, who have done the math.

Multi-cancer early detection (MCED) tests are becoming very popular: CancerGuard (Exact Sciences) and Galleri (GRAIL) are the most prominent. They detect fragments of tumor-derived DNA in the bloodstream using DNA methylation patterns (biochemical modifications) and other tumor-associated signals that differ between cancer and normal cells. They are being marketed directly to consumers and ordered by physicians despite lacking FDA approval, guideline endorsement, or a randomized controlled trial demonstrating that they reduce cancer mortality (1, 2). The intuition driving their adoption is understandable, but the mathematics underlying their performance in asymptomatic screening populations argues for considerable caution in both directions.

When the Test Is Positive

The leading MCED tests report specificity around 99.5%, meaning that among people who truly do not have cancer, the test correctly returns a negative result 99.5% of the time. That number sounds extraordinary, but in the context of population-level cancer screening, it’s actually pretty weak.

Why? Because the prevalence of asymptomatic cancer is low. In an asymptomatic screening population of people around age 50, the prevalence of any cancer detectable by an MCED test is ~0.5%, which means that only five people per thousand actually have cancer. The other 995 do not. A test with 99.5% specificity correctly identifies roughly 994 of those 995 cancer-free individuals as negative, but falsely flags ~5 as positive. With a sensitivity (the percent of people with cancer who the test identifies) around 50% (which is a reasonable aggregate figure across cancer types and stages), the test detects about 2 or 3 of the 5 true cancers. Among the 7 or 8 positive results generated, 5 are false positives. The positive predictive value (PPV), meaning the probability that a positive result reflects a real cancer, is only 30-40%.

The table below shows how PPV shifts with prevalence, holding test performance constant:

The test has not changed across any row of this table. Its sensitivity and specificity are fixed properties of the test and how it performs. What changes is the population: as prevalence falls, the enormous denominator of cancer-free individuals generates false positives that overwhelm the small numerator of true cancers. This is the Bayesian relationship between specificity and PPV, and it is inescapable. It’s math and it’s true, like water is wet and 1+1 = 2.

And it was shown to be true in a study of real people. In the PATHFINDER study, a prospective evaluation in an asymptomatic screening population (all of whom were over 50 and whose average age was 63 years), only 92 of 6,621 participants (1.4% screen positive rate) had a positive result, and only 35 were confirmed to have cancer (0.5% prevalence, leading to a PPV of 38% (3). Nearly half the signal-positive participants underwent biopsy or surgery (3). The math predicted what happened almost exactly.

There is an additional wrinkle: when a positive MCED result is followed by a PET-CT that reveals nothing, it remains genuinely unclear whether the patient is cancer-free or has a tumor too small to localize (4), which means the patient is left with uncertainty and worry - plus they are out a lot of money and have received additional radiation exposure and needle sticks.

But Even True Positives Are Not Simple

Suppose the test finds a real cancer. Our intuition says: good, we caught it early, now we can treat it. But there are at least three reasons this logic does not follow as reliably as it seems.

The first is stage distribution. MCED tests are better at detecting advanced disease than early disease. In the PATHFINDER study, less than half of the MCED-detected cancers were stage I or II, compared with nearly 3/4 detected by usual care (3). In a large real-world series of more than 100,000 Galleri tests, only 28% of signal-detected cancers were stage I or II, but nearly half were stage IV (5). Sensitivity for stage I disease in the CCGA validation cohort was approximately 17%, compared with roughly 90% for stage IV (6). A test that preferentially detects cancers that already are advanced has inverted the rationale for screening.

The second is lead-time bias. Even when an MCED test detects a genuinely early-stage cancer, earlier detection does not automatically translate into longer survival; it can simply move the clock forward without changing when, or whether, the patient dies of that cancer (7).

The third is tumor biology: not all cancers that shed detectable ctDNA are cancers that will kill the patient on a relevant timescale. Some are slow-growing with natural histories measured in decades, so detecting them earlier does not extend life, but it does extend the period of knowing you have cancer, which can be its own harm. At the other extreme, the cancers where early detection might matter most, such as pancreatic, ovarian, and small-cell lung cancer, either are detected too late by MCED tests given the tests’ low stage I sensitivity, or have biology so aggressive that even a genuine stage shift does not reliably change outcomes. The cancers most amenable to early-detection benefit tend to be the ones already captured by existing single-cancer screening programs.

When the Test Is Negative

So what does all of this mean for the neighbor who got a clean result and feels reassured? Not so much.

With an aggregate sensitivity ~50%, a negative MCED result misses roughly half of the cancers present in the tested population. For stage I cancers, where detection might actually change outcomes (see previous section), the sensitivity falls to ~17%, meaning the test misses more than four out of five early-stage cancers. Although the negative predictive value (NPV, the probability that a person with a negative test truly does not have cancer) is high in absolute terms at ~99.75%, the probability of not having cancer before the test already was 99.5% in this population, so the test adds only ~0.25 percentage points of reassurance.

The practical implication is that a negative MCED result provides no justification for deferring established screening programs with demonstrated mortality benefit, such as colonoscopy, mammography, or low-dose CT for lung cancer in eligible smokers. Feeling relieved about a negative test is understandable, but it is a misreading of what the test result means.

The Evidence Still Has to Catch Up

No randomized trial has demonstrated that any MCED test reduces cancer mortality (1, 2). Although “lack of proof is not proof of lack,” the onus of proof is on the new test, and the math of screening in an asymptomatic population is not great. An independent model of the NHS-Galleri trial, the first RCT designed to test MCED screening with a stage-based primary endpoint, estimated a 6-9% reduction in 5-year cancer mortality under favorable assumptions (8), but models are hypotheses, not evidence. Indeed, the NHS-Galleri trial of nearly 143,000 UK participants aged 50–77 years recently reported results: it failed to meet its primary endpoint of reducing stage III–IV cancer diagnoses among 12 prespecified cancers in the screening arm, though a reduction in stage IV diagnoses was observed as a prespecified secondary endpoint (9,10).

As of today, the FDA has not approved any MCED test for population screening. Ordering an MCED test in an asymptomatic screening population means accepting an ~2/3 false-positive rate on positive results, low sensitivity for early-stage disease, no demonstrated mortality benefit, and the risk that a negative test result will be misinterpreted as reassurance it never was designed to provide.

As physicians, we are admonished not to practice anecdotal medicine. But to patients all that matters is what happens to them (or to their family member or friend) and the meaning they assign to it. The neighbor who caught an early pancreatic carcinoma: that is the expected outcome for ~1/3 of positive results in a screening population. It is good, but it is not clear that it ultimately helped, though I hope it did. The brother-in-law with the clean result who feels great should still get his colonoscopy. Both stories are real, but neither tells you what you need to know. The math does.

References

1. Kahwati LC, Avenarius M, Brouwer L, et al. Multicancer Detection Tests for Screening: A Systematic Review. Ann Intern Med 2025. doi:10.7326/ANNALS-25-01877.

2. Hoffman RM, Wolf AMD, Raoof S, et al. Multicancer Early Detection Testing: Guidance for Primary Care Discussions With Patients. Cancer 2025;131:e35823.

3. Prajapati H, Mody M, Kakadiya J, et al. Potential and pragmatism: Clinical and public health implications of multi-cancer early detection (MCED) screening via cfDNA in asymptomatic adults. J Clin Oncol 2026;44(Suppl 16):e22570.

4. Carr DJ, Welch HG. Assessing the Clinical Utility of Liquid Biopsies Across 5 Potential Indications From Therapy Selection to Population Screening. JAMA Intern Med 2023;183:1144–1151.

5. Matrana M, Shukla V, Kingsbury D, et al. Real-World Data and Clinical Experience From Over 100,000 Multi-Cancer Early Detection Tests. Nat Commun 2025;16:9625.

6. Klein EA, Richards D, Cohn A, et al. Clinical Validation of a Targeted Methylation-Based Multi-Cancer Early Detection Test. Ann Oncol 2021;32:1167–1177.

7. Welch HG, Dey T. Testing Whether Cancer Screening Saves Lives: Implications for Randomized Clinical Trials of Multicancer Screening. JAMA Intern Med 2023;183:1255–1258.

8. Gogebakan KC, Lange J, Owens L, et al. Clinical Significance of a Multicancer Screening Trial With Stage-Based End Points. JAMA Netw Open 2025;8:e2536247.

9. Galleri MCED Test Fails to Meet Primary Endpoint in Large Trial. Cancer Discov 2026 May 30:OF1. doi: 10.1158/2159-8290.CD-NW2026-0060.

10. The ASCO Post Staff. Annual Galleri Screening Reduced Stage IV Cancer Diagnoses but Missed Primary Endpoint in First Randomized MCED Trial. Posted: 6/2/2026; last updated: 6/4/2026. https://ascopost.com/news/june-2026/annual-galleri-screening-reduced-stage-iv-cancer-diagnoses-but-missed-primary-endpoint-in-first-randomized-mced-trial/

My first review paper and book chapter were about Lp(a) excess and its historical treatments, including niacin, estrogen, and neomycin (1,2). So I am very excited that later this summer, we will get our first look at the cardiovascular disease (CVD) outcome data for pelacarsen, an Lp(a)-lowering therapy. The Lp(a) HORIZON trial is testing pelacarsen, compared with placebo, in people with established CVD and elevated Lp(a) (3). This study has been eagerly awaited because Lp(a) is a genetically determined, largely unmodifiable CVD risk factor, and we still do not know whether lowering it reduces cardiovascular events (2,4), despite compelling Mendelian randomization evidence (5) and strong biological plausibility. The field is primed to believe, which is precisely the moment to be careful.

In my last post, I outlined how I read a randomized clinical trial before I decide whether it should change my practice. When the Lp(a) HORIZON study is reported, I will apply every one of those steps, but a few will deserve particular attention.

First, what was the background lipid-lowering therapy and was it optimized? According to the design paper, patients were treated aggressively by the standards at the time the study was designed, with ~77% on high-intensity statins, but only ~56% on ezetimibe and ~11% on a PCSK9 inhibitor. The median baseline LDL-C was 65.7 mg/dL. But by the standards of the recent 2026 ACC/AHA Dyslipidemia guidelines, many, if not most, patients with established atherosclerotic CVD now would be treated toward an LDL-C goal of <55 mg/dL (6). Thus, if patients in Lp(a) HORIZON were not already on high-intensity statins, ezetimibe, and a PCSK9 inhibitor when indicated, then any apparent benefit from pelacarsen may partly reflect residual undertreated apoB-related risk, rather than Lp(a)-lowering specifically. Indeed, if apoB fell meaningfully in the treatment arm, isolating the Lp(a) effect from a more general lipoprotein-lowering effect will be difficult, because even modest reductions in LDL-C after correcting for Lp(a)-cholesterol or in apoB could affect event rates (7). A key question will be whether achieved apoB and LDL-C levels are similar between groups, or whether differences in apoB explain part of the observed effect.

Second, what are the absolute event rates in each arm, how large are the between-arm differences, and which event types are reduced? In Lp(a) HORIZON, the primary endpoint is a composite of cardiovascular death, nonfatal MI, nonfatal stroke, or urgent coronary revascularization requiring hospitalization (3). These endpoints sit on a hierarchy: cardiovascular death and nonfatal stroke matter much more to patients than nonfatal MI, which matters more than revascularization. Also, all-cause mortality is a secondary endpoint worth examining carefully, but it is a high bar to clear in 3 to 6 years. I hope for a markedly positive trial, but suspect that if the study is positive, the benefits will be driven primarily by reductions in nonfatal events. Then we will debate in what circumstances we should use the novel Lp(a)-lowering agents rather than further apoB reduction with existing agents, as we await trials like OCEAN(a)-Outcomes (TIMI 75) with olpasiran, among other studies.

I will share a full analysis when the data are available. But the reason to raise these questions now, before the trial reports, is that the moment of maximum enthusiasm is the moment of maximum vulnerability to misreading evidence. We have been here before.

The statin + niacin story

The HDL-Atherosclerosis Treatment Study (HATS) enrolled 160 patients with coronary artery disease and low HDL-C. In the key comparison, simvastatin plus niacin was associated with a 90% relative reduction in the composite of death, myocardial infarction, stroke, or revascularization (8). The authors were cautious about the small sample, but the cardiology community embraced the results and niacin’s reputation as a lipid therapy with hard CVD outcome data was established.

It wasn’t until a decade later that the assumption of benefit for adding niacin to a statin was challenged, when the AIM-HIGH study of >3,400 patients was stopped early for futility (9). And two years later, HPS2-THRIVE (10) settled the question in >25,000 patients: niacin added to statins provided no reduction in major vascular events and was associated with an increase in serious adverse events. There were some baseline differences between participants in these three studies, but the question about niacin’s usefulness was settled, and the answer was the opposite of what HATS had suggested.

The phenomenon of effect-size inflation (the “winner’s curse”) operates most powerfully when event counts are small, because trials that randomly deviate toward larger-than-true effects are more likely to cross significance thresholds and get published. HATS was a useful hypothesis-generating study that got mistaken for confirmatory evidence. The lesson is not that small trials are useless: it is that implausibly large effects in small samples deserve proportionally large skepticism.

What to watch for in August

The anticipation around Lp(a) HORIZON is legitimate, and I share it: I would love another treatment in my CVD risk-reducing arsenal, especially for people with high Lp(a). But I am cautious. Obviously, Lp(a) HORIZON is much larger than HATS, but high excitement and plausible biology are precisely when overinterpretation is most likely.

When the data are published, I will read the methods, the tables, and the supplements before I read the conclusions. I will look at the absolute event rates in each arm. I will ask which event types were reduced and by how much. I will ask what happened to apoB. I will ask whether the follow-up was long enough for the biology to express itself in hard endpoints. And I will remember that the trials that hold up do not need spin, because the data tell their story.

References:

1. Stein JH, Rosenson RS. Lipoprotein Lp(a) excess and coronary heart disease. Arch Intern Med 1997;157(11):1170-6.

2. Rosenson RS, Stein JH. Lipoprotein(a). UpToDate. Wolters Kluwer.

3. Cho L, Nicholls SJ, Nordestgaard BG, et al. Design and Rationale of Lp(a)HORIZON Trial: Assessing the Effect of Lipoprotein(a) Lowering With Pelacarsen on Major Cardiovascular Events in Patients With CVD and Elevated Lp(a). Am Heart J 2025;287:1-9. doi: 10.1016/j.ahj.2025.03.019.

4. Tsimikas S. A Test in Context: Lipoprotein(a): Diagnosis, Prognosis, Controversies, and Emerging Therapies. J Am Coll Cardiol 2017;69(6):692-711. doi: 10.1016/j.jacc.2016.11.042.

5. Burgess S, Ference BA, Staley JR, et al. European Prospective Investigation Into Cancer and Nutrition–Cardiovascular Disease (EPIC-CVD) Consortium. Association of LPA Variants With Risk of Coronary Disease and the Implications for Lipoprotein(a)-Lowering Therapies: A Mendelian Randomization Analysis. JAMA Cardiol 2018;3(7):619-627. doi: 10.1001/jamacardio.2018.1470. PMID: 29926099.

6. Blumenthal RS, Morris PB, Gaudino M, et al. 2026 ACC/AHA/AACVPR/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA Guideline on the Management of Dyslipidemia. J Am Coll Cardiol. Published online March 13, 2026. doi:10.1016/j.jacc.2025.11.016.

7. Yeang C, Karwatowska-Prokopczuk E, Su F, et al. Effect of Pelacarsen on Lipoprotein(a) Cholesterol and Corrected Low-Density Lipoprotein Cholesterol. J Am Coll Cardiol 2022;79(11):1035-1046. doi: 10.1016/j.jacc.2021.12.032.

8. Brown BG, Zhao XQ, Chait A, et al. Simvastatin and niacin, antioxidant vitamins, or the combination for the prevention of coronary disease. N Engl J Med 2001; 345(22):1583-92. doi: 10.1056/NEJMoa011090.

9. AIM-HIGH Investigators; Boden WE, Probstfield JL, Anderson T, et al. Niacin in patients with low HDL cholesterol levels receiving intensive statin therapy. N Engl J Med 2011; 365(24):2255-67. doi: 10.1056/NEJMoa1107579.

10. HPS2-THRIVE Collaborative Group; Landray MJ, Haynes R, Hopewell JC, et al. Effects of extended-release niacin with laropiprant in high-risk patients. N Engl J Med 2014;371(3):203-12. doi: 10.1056/NEJMoa1300955.

Step 1. What’s being compared to what?

The first step is to determine what the intervention is and what it is being compared against. The intervention usually is easy to identify and often is in the title of the article, but the comparator is just as important: Is it placebo or standard of care? These are the first considerations because the most important thing you’re going to look at is the difference between the treatment arm and the comparator arm regarding the outcomes (see next section). If the comparator is a placebo, you’re going to be more likely to see a difference, and that is reasonable in some contexts. But usually the best comparator is the standard of care, and not just any standard of care: the best standard of care. A classic trick in trials of new antihypertensive therapies, including renal denervation therapy, is to compare a new agent to an arm that is reasonable enough to pass for standard of care, but is well below best care, such as low doses of short-acting drugs given once a day, or therapies that are not particularly effective, like beta-blockers, thiazide diuretics, or once daily losartan.

The comparator doesn’t have to be a suboptimal drug to be a bad comparator - it could cause harm and make the intervention look better than it really is. In REDUCE-IT, the placebo was mineral oil, which raised LDL cholesterol, apoB, and inflammatory markers in the control arm. What looked like a home run for icosapentaenoic acid - a large reduction in cardiovascular disease events - may have been partly a function of the control arm getting worse rather than the treatment arm improving. The effects may be real, but the magnitude almost certainly is overstated.

Simultaneous Step 1. What’s the outcome and who cares?

Equally important are the outcomes and who cares about them. I immediately ask, are these objective or subjective, because objective outcomes - like death, or adjudicated myocardial infarction or stroke - are much less sensitive to bias. The next tier includes related clinical events such as ischemia-driven revascularization, hospitalizations for heart failure, or transient ischemic attacks – they have subjective components that get amplified if the study is not blinded. These endpoints can be biased in ways that may not be equally distributed between arms, even in a randomized trial. And lower tier outcomes - like changes in surrogate endpoints such as laboratory values or imaging measures may be one or more steps removed from outcomes patients actually care about.

I also ask who cares about the interventions and outcomes the most. Sometimes studies are designed to sell products, and they measure outcomes that patients don’t really care about. A classic example is using hospitalization for heart failure as an endpoint without looking at all-cause hospitalization. Hospitalization for heart failure is important, but you need context. Given the multimorbidity of most of our heart failure patients and the subjectivity of who gets hospitalized versus outpatient therapy, I’d want to know total hospitalizations. If they have fewer heart failure hospitalizations but more hospitalizations for kidney failure, falls, or infections, that apparent benefit might be completely lost or even offset.

I also like to understand the motivations of the study funders and authors. Are they trying to sell a product? Are they trying to improve longevity and quality of life? I can’t read people’s minds, but sometimes you can tell by the study design what they’re trying to do.

Step 2. Study design, blinding, and funding

The considerations above depend heavily on study design. For a new intervention, particularly a permanently implanted device, I want to see evidence of benefit on hard clinical outcomes or safety endpoints that matter to patients. The burden of proof is on the new intervention, not on a well-studied standard of care. For example, the CHAMPION-AF study was designed as a non-inferiority study, although I believe it should have been designed as a superiority study. It also included a bleeding-related endpoint that excluded post-procedural bleeding, even though patients care about bleeding whenever it occurs.

On the subject of blinding, it’s vital to know whether the study was double-blinded, single-blinded, or not blinded at all. There are few good excuses for not doing a double-blinded study, and it is concerning that articles get published examining the effect of a device or procedure on clinical outcomes without blinded placebo or sham control arms. Subjective or symptom-driven outcomes without a placebo control are highly biased, due to the well-known phenomenon of subtraction anxiety, in which a patient who enters a study expecting to receive a beneficial intervention doesn’t get it, becomes symptomatic, and drops out, creating a biased study sample, as well as placebo and nocebo effects. Recent examples include the TRILUMINATE Study of tricuspid valve repair, which showed improvements in symptoms among people who received a device intervention for tricuspid valve regurgitation but had no control arm. An even better example is renal denervation therapy, which showed a large reduction in blood pressure in the unblinded SIMPLICITY-HTN2 trial, but only a small, non-significant effect when tested against a sham procedure in SIMPLICITY-HTN3.

I also ask whether the follow-up was long enough for the intervention to have its full effect and for clinically meaningful events to accumulate, because trials that are too short can show early signals of large effect or no effect that do not hold in the long run.

And finally, I ask who funded the study, because everyone is biased - sometimes the bias is intellectual, sometimes it is financial. That bias shapes how studies are designed, written, and reported. Indeed, the study design often reveals intent more reliably than a disclosure statement.

Step 3. The study population

I then turn to Table 1 and ask who was enrolled, whether the study population makes sense for the intervention and outcomes, and whether the participants resemble the patients for whom I would consider the intervention. Next, I look at the recruitment flow diagram (usually Figure 1), because it tells me how many were screened and rejected to get into the study. That is important, because sometimes enrollment is so precisely tuned that the sample no longer reflects clinical reality.

The numbers in Table 1 tell a lot – and give a lot away. The recent VESALIUS-CV study enrolled patients with a mean LDL cholesterol of 122 mg/dL, despite reportedly intensive lipid-lowering therapy at baseline. Nearly all had diabetes mellitus or non-occlusive atherosclerotic vascular disease and, based on their demographics and countries of enrollment, likely were clinical patients rather than screen-detected cases. Those details matter because they substantially affect the external validity of the study, particularly for those who used it to justify lower lipid targets and population-based atherosclerosis screening.

Step 4. The absolute event rates

Then I turn my attention to the results and focus mainly on the absolute event rates within and between arms, and I ask myself: Do they make sense? Are the differences clinically important? Over the length of the study, what is the difference in absolute terms? I care much less about relative risks because people live absolute lives and have absolute events, not relative ones.

The effect size is vastly more important than the p-value, which is as much a function of sample size and variability as it is of the point estimates. I look at the confidence intervals, but I always return to the absolute event rates in each arm and ask myself how fragile they are. If by chance a small number of people in each arm had or didn’t have events, would it have changed the statistical outcome?

Step 5. How I actually read the paper

To get all of this information, you can’t just read the abstract, the introduction, and the conclusion. Those sections often are filled with spin, especially the abstract and the study conclusions. I read the methods, the tables and figures, and then go back and look at the results. I largely ignore the discussion until I’ve thought carefully about the data and weighed it against my prior convictions about the intervention and the biological plausibility, in particular, whether the event rates and the effect size are plausible. Then I read the conclusions, and I’m very cautious about spin.

If my conclusions differ from the authors’, or if anything seems even slightly out of the ordinary, I pull up the supplements. It requires extra clicks and downloading, but a vast amount of information is buried there. Often the key point that unlocks unexpected data can be found there. It’s also where additional analyses requested by reviewers are buried, where you can find details about the countries in which patients were enrolled, and where other findings are tucked away that may run against the message the authors are trying to convey.

Conclusion

Randomized clinical trials that have hard endpoints and large absolute effect sizes don’t need spin because the data tell the story. Learning to read past the abstract and the authors’ conclusions and focusing instead on the methods, the tables, and the supplements takes practice, but it doesn’t take long once it becomes habit. The payoff is that you stop outsourcing your clinical judgment to other people’s interpretations of their own work.

Most wearable metrics are not actionable for healthy people, but they easily can convince someone that something is “off” - and that is a real harm. The problem not only is that these devices are imprecise, but users, including physicians, rarely examine what their validation data actually show.

Heart Rate Variability: Measurement Variability Is Greater Than Your Day to Day Changes

Heart rate variability (HRV) is one of the primary selling points for consumer wearables. The Oura ring, for example, markets HRV as a window into “stress levels, overall health and well-being,” and suggests that day-to-day changes “can help you know how to approach your day.” They claim that the median nocturnal RMSSD (“root mean squared sum of differences,” their HRV measure) in adults is roughly 35–38 ms, it declines with age, and that the normal range spans from <20 to >200 ms across individuals. Because of that wide range, Oura encourages users to track their personal baseline, which is reasonable, in principle. But the question is whether the device can actually resolve meaningful changes against that baseline.

To answer that question, it is important to distinguish between two statistical concepts that often are conflated: correlation and agreement. Correlation measures whether two variables move together (whether high values on one tend to correspond to high values on the other). Agreement measures whether two devices produce the same numbers. A device that consistently reads 15 ms below a gold-standard ECG on every measurement would have a perfect correlation of 1.0 and still be systematically wrong. Correlation is necessary but not sufficient for clinical interchangeability. The appropriate test is a Bland–Altman analysis (1), which plots the difference between devices against the gold standard (reference) value and reports limits of agreement (i.e., the range within which 95% of individual differences fall).

The most rigorous independent validation of the Oura ring against a medical-grade ECG (2) found a high correlation for nocturnal RMSSD (r=0.96 for overnight averages), but that is misleading. For overnight averages, the 95% limits of agreement for RMSSD span ~35 ms total, centered on a mean bias of about -16 ms, which means that the ring systematically underestimates RMSSD with individual measurements varying by an additional ±17.5 ms around that bias. For 5-minute windows, which is what the Oura actually provides the user, the limits of agreement widen to approximately ±30 ms around a similar bias. In practical terms, the ±5-10 ms wiggles you see from day to day fall entirely within measurement scatter and normal biological variability. The limitation is not just hardware, but the combination of photoplethysmography, motion artifacts, and physiologic variability, none of which the device can distinguish from a genuine autonomic signal. At the individual level, these devices cannot reliably separate small changes from noise, particularly in short time windows.

Watch-Derived VO₂ max

VO₂ max illustrates the same problem. Estimated VO₂ max is reported by many modern smartwatches. VO₂ max is one of the strongest predictors of cardiovascular and all-cause mortality (3), but it is hard to measure even under ideal conditions: it requires a metabolic cart, expired gas analysis of oxygen and carbon dioxide, and continuous ECG monitoring during a maximal effort. A consumer watch cannot do any of that. Instead, it runs a proprietary algorithm (weighted mainly toward activity level and age, as best can be determined from the limited information manufacturers disclose) and produces a number that many users worry about when it falls. Agreement between watch-estimated and laboratory-measured VO₂ max is poor at the individual level (4). The number is not a physiologic measurement; it is a model output, and it should be read accordingly. “My watch says my VO₂ went down” is a common message I get from friends and patients who are unnecessarily worried. If you want to check your VO₂ max, you need a cardiopulmonary stress test, and that can be useful for some people, especially if you are training, as I discussed here.

Atrial Fibrillation: Promising, But Context Is More Important than the Watch Output

Atrial fibrillation (AF) provides the clearest example of why pre-test probability governs interpretation in diagnostic medicine and why wearable alerts cannot be evaluated in the abstract.

The 2023 ACC/AHA AF guidelines concluded that mass population screening with smartwatch apps only rarely detects new AF, and that for screening to be clinically useful it would need to improve outcomes and be cost-effective, neither of which has been established (5). In low-risk populations where AF prevalence may be <5%, the positive predictive value for smartwatch detection runs ~34%, meaning that about two out of every three alerts are false positives (6). Said another way, in the young people the watches are marketed to, a warning for AF is much more likely to be a false positive than a true positive.

A 2025 systematic review and meta-analysis found that smartwatches achieved a pooled sensitivity of 95% and specificity of 97% for AF detection (7). However, these pooled figures are misleading in a screening context because most validation studies excluded inconclusive tracings and were conducted in populations enriched for AF. A recent ACC Scientific Statement noted that diagnostic accuracy is variable and often lower than manufacturer reports (8). Indeed, the diagnostic yield of subsequent ambulatory ECG monitoring after an irregular pulse notification was only 32–34% (9). The distinction between detecting AF when it is present during a recording versus screening for AF in the general population remains critical.

The consequences of these alerts extend beyond inconvenience. Rosman et al. described a patient with paroxysmal AF and no psychiatric history who developed new-onset health anxiety triggered by excessive smartwatch monitoring, performing 916 ECGs in one year and generating 12 ambulatory clinic and emergency department visits despite appropriate anticoagulation and low arrhythmia burden (10). Kaplan et al. argued that the safety of wearables and the potential health risks associated with their use have gone “concerningly unexamined,” noting that evidence of efficacy does not mean an absence of harm (11).

But excessive concern is only one side of the problem. Wearables also can create false reassurance. I once cared for a patient who felt nauseated and short of breath, but whose Apple Watch showed normal oxygen saturation, a normal pulse, and no AF. He concluded that he was fine. In fact, he was having a heart attack and presented twelve hours later, after losing consciousness. He was lucky. The watch had not failed. It had accurately reported several parameters that were never designed to rule out myocardial infarction, but he interpreted their normality as evidence that nothing serious was happening.

There is, however, a more defensible use case: monitoring patients with established AF, particularly after ablation. Here the pre-test probability is high, the clinical question is specific, and the signal is more likely to be real. The guidelines explicitly recognize this role (5).

Anticoagulation decisions guided by smartwatch data alone are unsupported but are being investigated. Current FDA clearance language specifies that smartwatch AF burden algorithms are “not intended for medical decision making” (9). In a patient with known AF history or prior ablation, wearable-detected recurrence is worth taking seriously when corroborated by standard assessment and more reliable continuous monitoring (such as a Zio patch or implantable loop recorder). Similarly, its absence with supportive medical monitoring can help decision making, but I’d never rely on a watch alone. In the general population, screening is not guideline-supported and the false-positive burden is substantial.

Take Home

Here is the irony: whether your overnight RMSSD is 42 or 38 ms, whether your watch-estimated VO₂ max ticked up or down, whether your sleep score is 85 or 71, the advice is identical: exercise regularly, eat well, manage stress, and sleep consistently. None of these recommendations change because a wearable assigns them a number. For many users, the measurements create an impression of precision without providing information that meaningfully alters decision making.

Wearables can serve a legitimate purpose in specific clinical contexts: step counting for sedentary patients who benefit from behavioral feedback, heart rate monitoring during structured exercise training, and perhaps post-ablation rhythm surveillance in established AF. Outside those contexts, they often generate more data than we can interpret reliably.

The gap between what these devices measure and what users believe they reveal is not simply a hardware problem awaiting a software update. Better sensors may improve measurement accuracy, but they cannot supply clinical context, establish pre-test probability, or determine whether a given number matters. Those are problems of interpretation, not technology.

References

1. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1(8476):307–310.

2. Cao R, et al. Accuracy assessment of Oura Ring nocturnal heart rate and heart rate variability in comparison with electrocardiography. J Med Internet Res 2022;24:e27487.

3. Kokkinos P, et al. Cardiorespiratory fitness and mortality risk across the spectra of age, race, and sex. J Am Coll Cardiol 2022;80:598–609.

4. Petek BJ, et al. Consumer Wearable Health and Fitness Technology in Cardiovascular Medicine: JACC State-of-the-Art Review. J Am Coll Cardiol 2023;82(3):245-264.

5. Joglar JA, Chung MK, et al. 2023 ACC/AHA/ACCP/HRS Guideline for the Diagnosis and Management of Atrial Fibrillation. J Am Coll Cardiol 2024;83:109–279.

6. Ford C, et al. Comparison of 2 smart watch algorithms for detection of atrial fibrillation and the benefit of clinician interpretation: SMART WARS study. JACC Clin Electrophysiol 2022;8:782–791.

7. Barrera N, et al. Accuracy of smartwatches in the detection of atrial fibrillation: a systematic review and diagnostic meta-analysis. JACC Advances 2025;4:102133.

8. Varma N, et al. Promises and perils of consumer mobile technologies in cardiovascular care. J Am Coll Cardiol 2024;83:611–631.

9. Spatz ES, et al. Wearable digital health technologies for monitoring in cardiovascular medicine. N Engl J Med 2024;390:346–356.

10. Rosman L, Gehi A, Lampert R. When smartwatches contribute to health anxiety in patients with atrial fibrillation. Cardiovasc Digit Health J 2020;1:9–10.

11. Kaplan DM, et al. Wear with care: a call for empirical investigations of adverse outcomes of consumer health wearables. Mayo Clin Proc Digital Health 2023;1(3):413–418.

When a patient has made it to 85 years old in reasonable health, their instinct - and often their physician’s - is to redouble prevention efforts, optimize every number, and close every gap. I want to argue the opposite. If you have made it to 85 and are healthy and living independently, you have won the game of life. The appropriate response is not more medicine. It is recognizing what got you there and being very careful not to break it. We have precious few interventions that can reliably extend an 85-year-old’s lifespan (let alone their health span) but an infinite number of ways we can mess it up.

What winning the game actually means

Average life expectancy at birth in the United States is roughly 78 years (1). A healthy 85-year-old has outlived that mark by nearly a decade, and someone born in 1940, when life expectancy at birth was closer to 63 years, has outlived what the actuarial tables of their birth year would have predicted by more than two decades (2). Something is working, likely their genetics and lifestyle behaviors, acting together with a huge dose of good fortune, none of which we fully understand. That humility should inform everything that follows. This is a patient who succeeded at survival - not one who failed prevention - and interventions calibrated for a 58-year-old in a randomized clinical trial do not apply to them in any straightforward way.

It also helps to remember what we know about how treatments work. Most interventions benefit a small number of people substantially and a moderate number modestly. The averages obscure that distribution. We cannot predict who is in which group, and in someone who has already beaten the actuarial tables, the odds that any given preventive intervention will be the thing that helps them, rather than one of the many things that gets in the way, are not what they were at age 55. Treatments that transform outcomes in high-risk middle-aged patients often offer marginal benefit and real potential harms to someone who reached their mid-80s in good health.

Time-to-benefit and who was in the trials

Most of the landmark prevention trials that underpin our guidelines enrolled patients with median ages in the late 50s or early 60s, often with explicit upper age cutoffs. We then apply those findings to patients a decade or two older, with more comorbidities, more medications, and shorter time horizons. This is a foundational problem.

Time-to-benefit illustrates the problem. Statins in primary prevention take two to three years to show cardiovascular disease (CVD) risk benefits, and longer to show the full effect, including survival improvement. For a healthy 87-year-old without established CVD, those numbers may still favor treatment. For someone with lung disease, mild cognitive impairment, or metastatic cancer, it almost certainly does not. That needs to be discussed explicitly with our patients.

Where we can hurt you

The ways medicine can harm someone who has won the game are numerous, and they do not always even seem like harm in the moment. Here are some examples:

1. Dietary obsession. The relentless pursuit of a heart-healthy diet in someone already lean and functional can drive protein restriction, loss of muscle mass, impaired balance, and falls. In older people, falls can start a spiral of injury, debility, and complications that lead to death. The intervention meant to extend life shortens it. Among older adults who sustain a serious fall (i.e., one requiring emergency transport or resulting in a fracture) roughly 20-33% die within the following year (3,4).

2. Tight control of blood sugar. In older adults with diabetes, aggressive targets increase hypoglycemia risk substantially. Hypoglycemia is associated with CVD events, falls, cognitive decline, and death (5,6). The ACCORD trial demonstrated this directly, showing that intensive glycemic control (target HbA1c <6.0%) increased all-cause mortality compared with standard therapy, leading to early termination of the intensive arm (5). Any long-term microvascular benefit from the lower number is almost certainly offset well before it accrues.

3. Polypharmacy. Add an antihypertensive, a sleep aid, or even a simple antihistamine to someone already on five medications and you have created a fall waiting to happen, not a prevention success.

4. Aggressive screening. Particularly in older patients, more screening often generates incidental findings, which can lead to anxiety, and downstream costs including procedures whose harms are real and whose benefits, at that age and life expectancy, are theoretical.

None of this is inevitable, but it happens regularly because the clinical impulse is to act, which decision scientists call “action bias.”

Action bias

Ironically, many patients complain that their doctor just listened and “did nothing.” That “nothing” often demonstrates evidence-based and humanistic restraint and listening often is the best “something” they could have done. Physicians feel pressure to act. So do patients and their families. Inaction in medicine can be read as neglect or indifference, even when watchful waiting is the correct and evidence-based choice. It is the same tendency that makes goalkeepers dive on penalty kicks even though staying in the center of the goal is statistically the better choice. Diving looks like effort. Standing still looks like giving up. One of my most impactful mentors, Dr. Jesse Hall, then Director of the Medical ICU at University of Chicago Hospitals, often said, “Don’t just do something, stand there.”

Indeed, there is a specific harm in converting a healthy older adult into a medical project. They have to get the pill, pay for the pill, take the pill each day at the right time, have follow-up doctor visits, blood draws, imaging tests, and check their blood pressure twice a day and weight each morning, consulting the guidance each time. Each one seems reasonable and takes only a few minutes a day. But when you are older, those minutes feel like they matter even more. They collectively medicalize what should be healthy years and increase the work of living them.

The decision rule I try to apply is straightforward: will this test result change management in a way that meaningfully improves prognosis or quality of life? If yes, order it. If we are optimizing something that is already good enough, we should leave it alone. Time spent in waiting rooms is time taken from a life.

Careful medicine is not nihilism

I hope you know I am not advocating for nihilism or abandoning health care for older patients. There are interventions that work at advanced age. Blood pressure reduction improves CVD outcomes and reduces risk of cognitive impairment (7,8). But both findings require the same caveat: the SPRINT study that guided our recommendations excluded patients with prior stroke, heart failure, diabetes mellitus, and severely reduced kidney function, as well as institutionalized patients, patients with significant alcohol use, and those already on complex medication regimens (precisely the patients who populate the oldest-old in clinical practice). The trial population was healthier, more independent, and less medicated than most 85-year-olds seen in practice, so applying its targets without that context is a form of evidence misuse rather than evidence-based care. Patients with established CVD benefit from treatments like lipid-lowering therapy regardless of age, but personal preference and goals of care always reign supreme, especially in dealing with people who have won the game and hope to stay on the winning side as long as possible. For those patients, another medical aphorism applies: “The enemy of good is better.” In the context of an older adult who is doing well, the hardest prescription can be doing nothing additional medically and suggesting that they go live their life.

References

1. Murphy SL, Kochanek KD, Xu J, Arias E. Mortality in the United States, 2023. NCHS Data Brief. 2024 (521). doi:10.15620/cdc/170564.

2. Arias E, Xu J, Kochanek K. United States Life Tables, 2023. Natl Vital Stat Rep 2025;74(6):1.

3. Brauer CA, Coca-Perraillon M, Cutler DM, Rosen AB. Incidence and mortality of hip fractures in the United States. JAMA 2009;302:1573-1579.

4. Newgard CD, Lin A, Caughey AB, et al. Falls in older adults requiring emergency services: mortality, use of healthcare resources, and prognostication to one year. West J Emerg Med. 2022;23:375-385.

5. Action to Control Cardiovascular Risk in Diabetes Study Group; Gerstein HC, Miller ME, Byington RP, et al. Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med 2008;358:2545-2559.

6. Bonds DE, Miller ME, Bergenstal RM, et al. The association between symptomatic, severe hypoglycaemia and mortality in type 2 diabetes: Retrospective epidemiological analysis of the ACCORD study. BMJ 2010;340:b4909.

7. SPRINT Research Group; Wright JT Jr, Williamson JD, Whelton PK, et al. A randomized trial of intensive versus standard blood-pressure control. N Engl J Med 2015;373:2103-2116.

8. SPRINT MIND Investigators for the SPRINT Research Group; Williamson JD, Pajewski NM, Auchus AP, et al. Effect of intensive vs standard blood pressure control on probable dementia: a randomized clinical trial. JAMA 2019;321:553-561.

A 23-year-old man was referred for “familial hypercholesterolemia.” He had progressive exercise intolerance since adolescence and yellow nodules on his Achilles tendons, plantar surfaces of his feet, elbows, and buttocks, as below:

He reported multiple affected relatives with similar nodules. He had an aortic ejection murmur and bilateral carotid and femoral artery bruits.

His lipid panel showed total cholesterol of 393 mg/dL with LDL-C in the low 300s mg/dL range. A sibling had even higher levels.

This looks like severe familial hypercholesterolemia (FH). What would you do next?

(Think, then scroll down)

The pathogenic sequence variant commonly found in his Old Order Amish family,
APOB (G10580A), was not identified.

If this is not FH or familial defective apoB, what biological mechanism could produce this phenotype, and what test would you order next?

(Think, then scroll down)

Plasma sterol testing was ordered and showed striking elevations in sitosterol (217.7 mg/dL, normal <5) and campesterol (109.1 mg/dL, normal <7). Targeted genetic testing then revealed a homozygous mutation in ABCG8 (G1720A), leading to a diagnosis of sitosterolemia, an FH phenocopy.

Further evaluation of his family showed multiple affected relatives with widely variable clinical and subclinical vascular disease despite the same mutation. Table 1 in the reference below summarizes sterol levels, carotid ultrasound findings, and clinical features across the kindred.

How would you treat him?

(Think, then scroll down)

Treatment shifted from statins to dietary sterol restriction and ezetimibe. He had clinical and biochemical improvement. Bile acid–binding resins were added later to further optimize his sterol levels.

For more details and learning, please see the reference below.

With deep appreciation to Dr. Amy Peterson and Dr. James Deline for including me and my laboratory in the care of these patients. Special thanks to the patients for allowing publication of their kindred.

Reference:

Peterson AL, DeLine J, Korcarz CE, Dodge AM, Stein JH. Phenotypic Variability in Atherosclerosis Burden in an Old-Order Amish Family With Homozygous Sitosterolemia. JACC Case Rep 2020;2:646-650. doi: 10.1016/j.jaccas.2019.12.041.

HIV/AIDS

The medicine I trained in no longer exists, and I am not sure younger physicians know what they missed (or were spared). My medical school years in New Haven and my internal medicine residency in Chicago were defined by HIV - then called HTLV-III, or AIDS. Indeed, patients with AIDS represented more than half of the patients I saw in the hospital. I kept a log of my patients and reviewed it recently - they all died. And they died young, in their 20s and 30s, of diseases that sound almost medieval: Pneumocystis carinii pneumonia (now called Pneumocystis jirovecii pneumonia), cryptococcal meningitis, Kaposi’s sarcoma, and AIDS-wasting syndrome, often while afflicted by diarrhea from Isospora belli, Giardia, and Mycobacterium avium intracellulare (MAI, now called Mycobacterium avium complex). MAI was a disseminated atypical mycobacterium that a competent immune system suppresses without effort, but that in AIDS spread everywhere, causing drenching fevers, profound weight loss, and a deterioration that nothing could stop. I learned to recognize “the look”: pale, thin, exhausted, arriving in a wheelchair well before the age of 35 - and it still haunts me.

Paul Sax has written about the arc of this history better than I can, but the numbers from those years are stark enough on their own. Before AZT, median survival after an AIDS diagnosis was 11-13 months. AZT monotherapy, approved in 1987, changed that substantially: among patients who received it, median survival approached two years in some cohorts (1). The combination nucleoside era of the early 1990s extended survival further, but in 1995 and 1996 protease inhibitors arrived, and HAART changed everything: AIDS mortality in the United States fell by more than 70% between 1995 and 2000. Today, a 40-year-old starting antiretroviral therapy with a good CD4 count can expect to live into their mid-to-late 70s, within a few years of the general population (2). From eleven months to near-normal life expectancy within roughly a decade is one of the most compressed therapeutic revolutions in the history of medicine. That hope has sustained me in medicine and influenced my research career on the cardiovascular effects of HIV and its treatments.

Heart Attacks

We literally used to run to the emergency room when someone was thought to have a heart attack, and back then I loved it. We’d look at the ECG, the CPK and LDH levels, and decide about infusing TPA, because “minutes equal myocardium.” I also remember watching patients, including some in their 50s and 60s, bleed into their brain from the treatment meant to save them. If they lived, we prescribed aspirin, metoprolol, and eventually ACE inhibitors. I chose cardiology over general internal medicine because more of those patients lived. I loved the acuity and the physiology, but ultimately what drew me in was the impact of prevention.

The arc of mortality for ST-elevation myocardial infarctions (STEMIs) tells a parallel story to that of HIV/AIDS, though less compressed. In the era I trained, in-hospital mortality from a heart attack was 15-18%, which already was a significant improvement over the ~30% mortality of the pre-CCU era, when continuous monitoring and rapid defibrillation did not yet exist. With thrombolytics, mortality fell to the upper single digits (3). With primary percutaneous intervention (“angioplasty” and “stents”) and modern secondary prevention such as statins and cardiac rehabilitation, 30-day mortality in contemporary STEMI patients is only 4-6%. As important, functional recovery of survivors and their ability to do things that really matter, like returning to work, living normal lives, and participating in activities without shortness of breath have improved comparably (4).

Cancer

I did eight months of oncology as an internal resident, and it was brutal: a revolving door of lung, liver, pancreatic, breast, and head and neck cancers, as well as melanoma and leukemia, many with certain death in less than a year, after repeated hospital admissions, intubations, paracenteses, thoracenteses, and nearly anything you can do to a person with a needle or a tube. Some of that has not changed. But much of it has changed so profoundly that it is hard to convey to anyone who did not see the before times.

Chronic myeloid leukemia (CML) had a five-year survival rate of 22% in the 1970s and 1980s. Today it exceeds 70%, with tyrosine kinase inhibitors providing most patients with near-normal life expectancy. Multiple myeloma has gone from a median survival of 2-2.5 years before 2000 to more than eight years now, with selected patients cured. Metastatic melanoma saw five-year survival more than double, which is one of the most rapid survival gains in any solid tumor. Liver cancer was uniformly fatal when I trained, but now carries a five-year survival approaching 25%. Pancreatic cancer remains dire: five-year survival still is around 13%, though for localized disease survival has improved several-fold (5). Lung cancer has seen survival improve meaningfully, driven by immunotherapy and targeted agents, with the largest absolute gains among solid tumors (6).

Other Disease States

Similar stories run through much of the rest of medicine. Dialysis went from a scarce, committee-rationed intervention in the 1960s to early 1980s, to a routine chronic therapy that sustains hundreds of thousands of patients. Solid organ transplantation moved from experimental to standard of care, and long-term survival after kidney transplantation now is common (7). Rheumatoid arthritis went from progressive joint destruction managed with gold salts and high-dose prednisone to remission as a realistic and common goal, with methotrexate optimization and biologic therapies. Crohn’s disease and ulcerative colitis, which I watched destroy abdomens and send patients to the operating table repeatedly, now are managed with therapies that can induce mucosal healing. None of these transformations is complete. But the medicine I practiced in training and the medicine practiced today are not the same discipline.

In parallel, US life expectancy has increased by ~9 years since 1960, from just under 70 to just over 79 years (8). That is a lot to celebrate. Admittedly, the progress has been uneven and punctuated, with drug overdoses, the COVID-19 pandemic, and structural failures in primary prevention and in the health system taking their tolls - and international comparisons are not flattering. But today I want to celebrate what we have actually accomplished, recognizing that our work is far from done.

I remind myself of that when I am frustrated by the sometimes crushing burden of patients to be seen, inbox messages to answer, and the relentless pressure to make money off the health and well-being of our population. Like many of you, I adjust and find ways to be helpful and to contribute to medical care and research, all in service of our patients.

References

1. Moore RD, Hidalgo J, Sugland BW, Chaisson RE. Zidovudine and the natural history of the acquired immunodeficiency syndrome. N Engl J Med. 1991;324:1412–1416.

2. Trickey A, Sabin CA, Burkholder G, et al. Life expectancy after 2015 of adults with HIV on long-term antiretroviral therapy in Europe and North America: a collaborative analysis of cohort studies. Lancet HIV. 2023;10(5):e295–e307.

3. Puymirat E, Simon T, Steg PG, et al. Association of changes in clinical characteristics and management with improvement in survival among patients with ST-elevation myocardial infarction. JAMA. 2012;308(10):998–1006.
4. Thrane PG, Olesen KKW, Thim T, et al. Mortality trends after primary percutaneous coronary intervention for ST-segment elevation myocardial infarction. J Am Coll Cardiol. 2023;82(10):999–1010.

5. Siegel RL, Kratzer TB, Giaquinto AN, Sung H, Jemal A. Cancer statistics, 2025. CA Cancer J Clin. 2025;75(1):10–45.

6. Howlader N, Forjaz G, Mooradian MJ, et al. The effect of advances in lung-cancer treatment on population mortality. N Engl J Med. 2020;383(7):640–649.

7. Hart A, Lentine KL, Smith JM, et al. OPTN/SRTR 2019 annual data report: kidney. Am J Transplant. 2021;21(Suppl 2):21–137.

8. Woolf SH, Schoomaker H. Life expectancy and mortality rates in the United States, 1959–2017. JAMA. 2019;322(20):1996–2016.

Like most people, I learned to ride a bike when I was a child, but my relationship to it always was practical: finding friends to play with and getting from one place to another. As will be a surprise to no one that knows me and my natural (lack of) balance, I never jumped curbs, popped wheelies, or did other tricks, so I never developed the instinctive handling skills or confidence that seemed to come naturally to other kids. That part never really changed. Even now, 23 years after becoming an avid road cyclist, I can’t ride no-hands, and part of me feels like I should be putting on a seatbelt when I clip in.

My first ten-speed was a blue Sebring Viscount, advertised as being made from “aerospace” material. What I remember most is how light it was.

I started biking again in 2001. A friend and I started riding on Sunday mornings just to talk. He showed up on a black steel bike that I thought was made out of iron and looked like it might survive a collision with a tank. I showed up with a Trek hybrid that reflected my usual instincts, with a bell, a giant bright headlight, handlebar extenders with rearview mirrors that could turn into mountain bike handle extensions, and a hard-wired cycle computer that told me not just how fast I was going, but how light or dark it was, my altitude, and the road temperature. Over time the rides got longer. My seat went up. The handlebars rotated forward. The kickstand disappeared. I started paying attention to distance and speed and the bike morphed from being something I used to get somewhere to something I was trying to improve.

Then one day I fell in love. I was riding alone just after sunrise on a Sunday morning, and I went down Old Sauk Pass for the first time ever, into a valley. There was no traffic, no conversation, no devices asking for anything. Just the wind, the gorgeous sun shining through moist leaves, and the soundtrack of red-winged blackbirds, cardinals, and robins, interspersed with blue jays, black-capped chickadees, and mourning doves taking off - as well as goldfinches, red-tailed hawks, and turkey vultures flying. I fell in love with that feeling in that moment, though I didn’t realize I was trying to recreate it for the next 20+ years.

I ultimately bought my first road bike in 2003 and I loved it immediately. It was a titanium Litespeed Firenze, a much better bike than its rider. I also was a little afraid of it. The tires looked impossibly thin and jumped forward with the slightest pedal stroke. Clipping in made no sense to me. I practiced on my neighborhood circle and fell often enough that neighborhood kids would stop, watch, and laugh, but I eventually learned. I bought multicolored cycling jerseys and a Livestrong yellow band and read Lance Armstrong’s book, “It’s Not About The Bike.” I looked the part and loved cycling.

But then I started doing what I tend to do: I added things and started optimizing. A wireless Bluetooth cycle computer to monitor not just speed and distance, but also cadence and my heart rate (using a chest strap). Then a newer and better version of it. A mirror on the handlebars, which later migrated to my glasses. Brighter lights. An extra chainring for climbing hills. I started paying attention to grams, the weight of components, the weight of my helmet, and where the bottle sat on the frame. All of it seemed to matter at the time and had a rationale, as well as the expectation that all of it should work, all of the time.

When it did, it was satisfying. When it did not, it was hard to ignore. I would stop mid-ride to reconnect sensors, adjust things, try to get everything to sync the way it was supposed to work. There were times I was more focused on whether my heart rate signal had dropped than on the road in front of me.

One Sunday morning, riding through the UW-Arboretum with a friend, I stopped to do exactly that. “I can’t get my heart rate or speed to work,” I said. After stopping the third time, he looked at me and said, “I guess you’re not bicycling then. Whatever will you do?” It was a throwaway line, but it landed. At some point, I had replaced the ride with the numbers.

The next decade made the question of safety more real. My wife died suddenly, and with that came a fear of being out of contact or unavailable. A friend had a concussion, not on a bike, but it changed how we thought about risk. We chose different routes and joined an industry geared towards us with better helmets, more and brighter lights, and then another layer of connectivity, including smartwatches with notifications, helmets with crash detectors that notified your friends, and rear-facing cameras. What I gradually noticed was that the long conversations I relished when cycling with friends increasingly became interrupted by notifications, the same interruptions that exist everywhere else, now were appearing on rides that used to be one of the few places they did not.

The barrage of advertising and messaging charged on: “Return home safely.” “Increase awareness.” “Stay in control.” Well, safety is my love language, so they got my attention. I remember being told, “Jim, you need a rear-facing camera, it’s amazing for safety.” But I did not understand it. My primary concern is not getting hit by a car and if I did (G-d forbid) it’s surviving with my brain intact, not whether I have them on video. “But Jim, drivers behave differently because they are being recorded - it’s safer.” Me: “That assumes they know they are being recorded, which I don’t think they can tell at 55 mph, especially if you think they don’t notice you.”

Last summer came radar: “Jim, it’s amazing. You can see cars behind you from 150 meters away.” And I found myself asking, what exactly am I going to do differently because I know there is a car 492 feet behind me, even if it is accelerating? Radar didn’t seem to affect my limited options of (1) waving my hands to be more visible, (2) staying on the right side of the road where I cycled anyway, and (3) bailing off the road. I like my rearview mirror that is attached to my sunglasses, so much so that it is second nature to use. But radar highlighted something larger for me: the cumulative effect of gadgets on attention, and not just on the road ahead. Like fancy cycling computers that included text messaging and GPS routing, they increasingly pulled riders toward electronics and troubleshooting rather than the ride itself.

“I am not sure if my radar is working.”
“Can you ride behind me so I can see if it picks you up or just the cars?”
“Car back! Do you see it? Oh, it turned off.”

At the same time, I found myself moving the other way. Not as a statement, but in service of that amazing moment 23 years ago when I descended down Old Sauk Pass, listening to the wind and the birds, fully absorbed in the moment I was living in. I never got a smart bike computer or radar. I even took the cadence magnet off my crank and kept my phone safely away in my back pocket in silent mode. I kept the blinking lights in the front and back, especially since there is some (weak) evidence to support the use of biovisibility measures. I still wear a heart rate monitor, but only when I am heart rate training (like now, since I am recovering from stacked illnesses).

Last Sunday morning, I rode alone. It was early, and the birding soundtrack was back. I even saw a turkey, some does, and a killdeer on the side of the road, as well as two sandhill cranes that recently came back to Wisconsin for spring breeding and nesting. When I got home, I had nine missed messages, but the world went on. I also had ridden faster than the week before, though I only learned that later. I felt so relaxed and realized that I finally had optimized my cycling, not by adding more inputs, but by removing many of them. I was riding faster, my fitness was improving, and yet the experience itself had become simpler and calmer.

We live in an attention economy, a world increasingly built around connectivity, metrics, and optimization. Sometimes those things genuinely are useful. But often they slowly displace the experience they were meant to support. The key questions are “optimized for who” and “optimized for what.”

For me, cycling is about fitness, camaraderie, and being outside in the bright light, the crisp air, and the sounds around me. If I want metrics and connectivity, I can stay indoors on Zwift. But out on the road, I rediscovered something increasingly rare: the ability to pay attention.