“Statins can safely cut risk of heart attacks or strokes in healthy people aged over 70, world-first clinical trial shows.” (The Guardian)

“Cholesterol-lowering medication reduces major cardiovascular events by 30 per cent in older people without known cardiovascular disease.” (European Society of Cardiology)

“This common medication cuts heart-attack risk in older people.” (Sydney Morning Herald)

“The life-saving medication used to prevent heart attacks and strokes.” (9 News Australia)

“Atorvastatin Cuts Major CV Events by 30% in Older Adults, Driven Mainly by Nonfatal Events.” (Medscape)

Anyone reading these headlines could reasonably come away with a simple message: statins produce a large cardiovascular benefit in healthy people over 70. And in one sense, that message is correct. The newly reported STAREE trial found a statistically convincing reduction in major cardiovascular events with atorvastatin [1].

But “30% reduction” is a relative measure. For an older person deciding whether to take a tablet every day, potentially for the rest of life, there is another question that may be at least as important:

How much longer can I expect to remain alive and free from a major cardiovascular event if I take the drug?

The answer turns out to be considerably more modest than the headlines might suggest.

What did STAREE actually find?

STAREE was a large double-blind randomised trial conducted in Australian general practice. A total of 9,971 people aged 70 years or older, with no previous clinical cardiovascular disease, diabetes or dementia, were randomly assigned to atorvastatin 40 mg daily or placebo. Their mean age was 74.7 years and approximately 52% were women.

The trial had two primary outcomes. The first was major cardiovascular events, comprising cardiovascular death, nonfatal myocardial infarction, stroke or coronary revascularisation. The second was disability-free survival, defined as remaining alive without dementia or persistent physical disability.

After a median of 5.9 years, major cardiovascular events had occurred in 6.0% of participants receiving atorvastatin compared with 8.3% receiving placebo (Figure). The corresponding rates were 10.9 and 15.5 events per 1,000 person-years.

The hazard ratio was:

HR = 0.70

with a 95% confidence interval ranging from 0.61 to 0.82.

Hence the widely reported headline:

“Statins reduce major cardiovascular events by 30%.”

That statement is statistically correct. But another way of describing exactly the same result is that the observed event proportions differed by approximately:

8.3% - 6.0% = 2.3%

Over approximately six years, therefore, atorvastatin was associated with about 2 to 3 fewer first major cardiovascular events for every 100 people treated.

The same trial told a rather different story for its other primary endpoint. Disability-free survival was similar between groups: 12.8% versus 13.6% experienced death, dementia or persistent disability, corresponding to an HR of 0.94 (95% CI 0.84 to 1.05; P = 0.25).

So we have an interesting juxtaposition:

30% relative reduction in major cardiovascular events, but little detectable improvement in disability-free survival.

That deserves more consideration than a headline can provide.

What is the weakness of the hazard ratio?

The hazard ratio is extraordinarily useful for statistical analysis, but it is less useful for communicating treatment benefit to patients.

An HR of 0.70 describes the ratio of instantaneous event hazards between the two treatment groups, assuming the proportional-hazards framework is appropriate. It does not mean that every individual has a 30% lower probability of experiencing an event. More importantly, it tells us neither how many additional days a patient remains event-free nor how much extra life the treatment provides.

The difference matters especially in preventive medicine. Imagine a healthy 75-year-old asking:

“I understand that my risk is reduced by 30%, but what does that mean for me? Am I likely to gain a week, six months, or five years without a heart attack or stroke?”

The HR cannot answer that question directly.

Interestingly, the STAREE investigators had anticipated this problem. The investigators clearly understood the potential value of restricted mean survival time (RMST) and restricted mean time lost (RMTL). Their prespecified statistical analysis plan [2] stated that differences in RMST, or RMTL where competing causes of death were relevant, would be calculated and reported as treatment effects alongside the hazard ratio, or used in preference to an overall HR if proportional hazards were inappropriate.

Thus, RMST and RMTL are not alternative analyses suggested after seeing the STAREE findings. They were part of the investigators’ planned analytical framework before the primary results were known.

Yet numerical estimates of RMST or RMTL do not appear in the main presentation of the trial results.

The reason is unclear, and there is no need to speculate about it. But their inclusion would have been particularly valuable because these measures translate an abstract relative treatment effect into something most people readily understand:

time.

How much event-free time was gained?

RMST represents the average survival or event-free time accumulated up to a specified time horizon. RMTL expresses its counterpart as the amount of time lost because an event occurs.

Using the published STAREE cumulative-incidence curves and restricting follow-up to six years, close to the observed median follow-up of 5.9 years, I can obtain an approximate estimate.

Digitising and integrating the major cardiovascular-event curves gives an RMTL difference of approximately:

0.076 years

or about:

28 days.

This provides a very different way of expressing the STAREE result:

During approximately six years of treatment, the average participant receiving atorvastatin gained about one additional month free from a first major cardiovascular event compared with placebo.

The estimate should be regarded as approximate because it has been reconstructed from the published figure. The investigators could calculate the exact value, together with its confidence interval, directly from the participant-level data.

For disability-free survival, the corresponding difference from the curves appears to be only around one week over six years, although this estimate is necessarily more uncertain because the two curves lie so close together.

The two descriptions of the cardiovascular finding are therefore both correct:

“a 30% reduction in cardiovascular hazard”

and

“approximately 28 additional cardiovascular-event-free days over six years.”

But they communicate the magnitude of treatment benefit very differently.

What about the remainder of a person’s life?

For someone aged 70, six years is only part of the relevant time horizon.

According to US life tables, a 70-year-old man has approximately 14 years of remaining life expectancy, while a 70-year-old woman has approximately 16 years. At age 80, the corresponding residual life expectancies are approximately eight and nine-and-a-half years.

This leads to a more clinically interesting question:

How much longer might an older person remain alive and free from a first major cardiovascular event if statin treatment were continued for the remainder of life?

I explored this question by combining US sex-specific survival probabilities with the age-specific cardiovascular-event rates anticipated in the STAREE protocol and applying the observed treatment effect of HR = 0.70.

The resulting estimates are illustrative rather than direct trial observations:

These numbers represent additional time alive and free from a first major cardiovascular event, not additional life expectancy.

That distinction is crucial.

A prevented coronary revascularisation does not necessarily extend life. Neither does every prevented nonfatal myocardial infarction or stroke. STAREE therefore does not establish that a 70-year-old taking atorvastatin will live five or six months longer.

Rather, under the assumption that the cardiovascular treatment effect persists, our modelling suggests that a healthy 70-year-old might expect to spend approximately 5 to 6 additional months of the remainder of life without experiencing a first major cardiovascular event.

As treatment is started at progressively older ages, the expected gain becomes smaller because the remaining lifespan becomes shorter and competing causes of death increasingly limit the opportunity for cardiovascular prevention to exert its benefit.

I must emphasize that these are extrapolations (from the key figure in the paper), not STAREE results.

Caveat: The lifetime estimates require several assumptions. Most importantly, they assume that the HR of 0.70 observed during approximately six years persists throughout the person’s remaining lifetime. They also assume that the age-specific cardiovascular event rates used for STAREE planning reasonably approximate rates as people age, and that treatment effects are similar in men and women.

Furthermore, the US general-population life table does not precisely describe the STAREE population. STAREE recruited unusually healthy older people without established cardiovascular disease, diabetes or dementia.

The lifetime estimates should therefore be regarded as a thoughtful extrapolation of the trial evidence, rather than as findings demonstrated by the trial itself.

Nevertheless, they illustrate an important principle about the communication of preventive treatment effects.

A real benefit, but a modest absolute benefit

STAREE is an important trial. It provides convincing randomised evidence that atorvastatin reduces major cardiovascular events in healthy people aged 70 years and older.

But the magnitude of the benefit looks quite different depending on how it is expressed.

A 30% relative reduction in cardiovascular hazard sounds large.

An absolute difference of about 2.3 percentage points over six years sounds more modest.

And an estimated gain of approximately 28 additional days free from a first major cardiovascular event over six years puts the result directly onto a time scale that patients can understand.

If the cardiovascular benefit persists throughout the remainder of life, illustrative modelling suggests an additional two to six months alive without a first major cardiovascular event, depending on the age and sex at which treatment begins.

Those months may certainly matter. A myocardial infarction or stroke prevented is important to the person who would otherwise have experienced it. Statins are also inexpensive and, for most people, relatively easy to take.

But the results do not imply years of additional healthy life, nor has STAREE demonstrated a meaningful extension of disability-free survival.

The principal message from STAREE therefore deserves a little more nuance than some of the headlines suggest. Atorvastatin produces a genuine and statistically convincing cardiovascular benefit in healthy older adults, but the average absolute benefit to an individual appears modest.

For someone contemplating preventive treatment for the remainder of life, perhaps the most useful question is therefore not “By what percentage will this drug reduce my risk”, but rather “Given my age and how long I am likely to live, how much longer might I remain alive and free from a major cardiovascular event if I take this drug?”

For STAREE, the emerging answer appears to be: about 1 month during the 6 years actually studied, and perhaps a few months over the remainder of life if the treatment effect persists.

That is still a benefit. But it is a rather different message from “30% reduction” or “life-saving medication.”

____

[1] Zoungas S, Wolfe R, Moran C, Nicholls SJ, Cloud GC, Reid CM, Τonkin AM, Beilin L, Wierzbicki AS, Chong TTJ, Broder JC, Curtis AJ, Flanagan Z, Hopper I, Ryan J, Spark S, McNeil JJ, Nelson MR for the STAREE Investigators. Atorvastatin, Cardiovascular Events, and Disability-free Survival in Older Adults. N Engl J Med Published August 28, 2026.

[2] Wolfe R, Heritier S, Zomer E, et al. A randomised clinical trial of STAtin therapy for Reducing Events in the Elderly (STAREE): Statistical analysis plan. medRxiv. 2025. doi:10.1101/2025.02.24.25321974.

Here is a typical scene: a graduate student or postdoc presents a sophisticated analysis of gene-expression data. Colourful heatmaps, dendrograms, k neatly separated clusters. Cluster 1 maps onto one biological pathway, cluster 2 onto another, and from there a compelling biological story takes shape.

Then I ask something simple: “Why k clusters?”. The answer is almost always some version of “that’s what the software gave us.” I push a little further. What happens with 4 clusters, or 6 clusters? Is the distance metric Euclidean, correlation-based, something else? How sensitive is the result to the algorithm, the initialization, the preprocessing steps? Do the clusters hold up under resampling? Usually the conversation stalls there, because the presenter can’t answer any of it.

What worries me more is how often this happens among researchers with little or no formal training in biostatistics. They are applying methods that professional statisticians spend years learning to use correctly. Sampling variation, confounding, selection bias, measurement error, collider bias: these ideas may be poorly understood, or simply unfamiliar, and yet the results get presented with total confidence.

I’ve started calling this phenomenon as a kind of “software-driven research”. I think it’s becoming one of the defining methodological problems of the big-data, AI era.

When the tool starts to determine the question

“Software-driven research” isn’t a formally defined term. I use it loosely to describe research where the software’s capabilities begin to shape the analysis, and the analysis then shapes the scientific story, rather than the other way around.

The clearest sign is a presenter walking through a sequence of software outputs they don’t fully understand themselves.

Modern software has extraordinarily capability. R alone puts vast statistical and machine-learning ecosystems within reach. SAS, SPSS, Stata, bioinformatics platforms, cloud tools let researchers run sophisticated analyses with a handful of commands or a few clicks. These methods used to demand real specialist training.

Traditionally, research runs in a particular order: a question comes first. You form a hypothesis, define your target population, design the study, decide how variables will be measured, think through possible bias and confounding, define what you’re actually trying to estimate, then choose a model.

Software-driven research often runs backwards. You start with a dataset, maybe several public ones, you know roughly what a given package can do, you run a bunch of analyses, something interesting pops out, and a scientific narrative gets built around it after the fact.

By the time both of these end up as a published paper, they can look nearly identical. Underneath, they’re not the same thing at all.

Cargo-cult statistics

More than fifty years ago, Richard Feynman coined the term “cargo cult science” [1]. The term describes activities that have the outward appearance of science while lacking its deeper intellectual discipline. This is particularly true when when researchers fail to question their own assumptions or actively seek evidence that might show their conclusions to be wrong.

Philip Stark and Andrea Saltelli later extended this to “cargo-cult statistics” [2]: statistical practice that has become ritual rather than reasoning. Someone runs a cluster analysis simply because the data can be clustered. Someone throws every available variable into a regression because the software lets them. P < 0.05 gets treated as a verdict on truth. Random forests get chosen over logistic regression partly because they sound more impressive.

Underneath all of these is the same missing question: why?

Why is this model appropriate here? Why treat these observations as independent? Why assume a linear relationship? Is this variable really a confounder, or could it be a collider? Why was cross-validation done this particular way? And the question that actually matters is what do the study design and the data justify concluding?

In 2016 the American Statistical Association issued a formal statement on P-values [3], largely because mechanical readings of statistical significance had become so entrenched in practice. Significance is not the same as importance, and it’s not proof that a hypothesis is true.

That’s really the line between doing statistics and thinking statistically.

Take cluster analysis again. It makes the problem unusually visible, because the output is so visually persuasive — distinct, coloured groups practically beg to be read as real, natural categories. But the result can shift depending on the distance metric, the scaling, the algorithm, the initialization, the number of clusters specified, even quirks of the particular sample. Cluster validation exists as its own subfield precisely because apparent clusters often aren’t stable or meaningful. Kerr and Churchill proposed bootstrap methods for exactly this reason more than twenty years ago, to check how reliable clustering results in microarray experiments actually were [4].

So when someone announces “we’ve identified five molecular subtypes,” that’s really where the work should start, not end. Would an independent sample reproduce them? Would a different preprocessing pipeline make some of them vanish? Are they real biology, or an artifact of batch effects? Does the data actually support five discrete groups, or is that just what the algorithm was told to find?

Software will almost always produce clusters if you ask it to. The scientific question is whether those clusters tell you something about nature, or only about the algorithm.

This is also why the same dataset can produce wildly different conclusions in different hands. Silberzahn and colleagues showed this directly: 29 independent analytics teams, same dataset, same question — were football players with darker skin more likely to receive red cards? [5] The teams made different analytical choices along the way, and the results diverged sharply: odds ratios ranged from roughly 0.89 to 2.93, with twenty teams finding a significant association and nine finding none.

I think of this as “black-box science.” Data go in one end, the software runs, and a p-value, a hazard ratio, a cluster assignment, a feature-importance score, or an AUC comes out the other. The researcher understands the input and can read the output, but the reasoning connecting the two is a blind spot.

That blind spot is where the trouble starts. Every model carries assumptions. Every algorithm has tuning parameters. Every pipeline involves choices — about cleaning, transformation, variable selection, missing data, how the outcome is defined, how the data get split for training and testing, which metric counts as success. Any one of these choices can move the result.

Understanding the methodology matters more than knowing the command syntax.

AI takes it a step further

Traditional statistical software still forced researchers to make a lot of the decisions themselves. Fitting a Cox model meant at least recognizing you had time-to-event data, choosing covariates, specifying the model, writing the code, checking the assumptions.

Generative AI changes that. A student can now upload a dataset and simply ask:

“Analyze this and tell me which statistical method to use.” “Run the analysis.” “Interpret the results.” “Make publication-quality figures.” “Write the Results section.” “Write a Discussion suitable for a high-impact journal.”

Someone with no real grounding in study design, epidemiology, biostatistics, or the substantive field can now produce work that looks a lot like real research.

AI is very good at manufacturing the appearance of expertise.

Which is why it’s unsettling to see advertisements implying researchers no longer need to learn statistics or methodology at all. Research methodology isn’t a set of mechanical steps waiting to be automated away by a prompt.

A prompt can’t undo selection bias baked into a study design. A sophisticated regression can’t invent a counterfactual the design never provided. A machine-learning pipeline can’t rescue a badly measured outcome. And an AI explaining a model back to you is not the same as you understanding that model.

With conventional software, the model could become a black box. With generative AI, the entire research process can become a black box.

A generation with impressive CVs

My biggest worry, as an academic, is about the kind of scientists our universities are actually producing.

When I studied biostatistics 40 years ago, I learned it from the mathematical foundations up. I was asked to derived likelihoods, worked through classical theory, wrote Fortran codes. Not because there’s virtue in doing things the hard way, but because it instilled a particular discipline: understand how a method was built, how it works, and what it assumes, before you use it.

Just as important, the data were tied to a specific question. Researchers in my generation developed hypotheses, designed or helped design the studies, collected and checked the observations ourselves, ran the analyses, interpreted what came out. We had to live with the data, in a way that’s harder to replicate now.

These days, public datasets have changed that ecosystem. A student can pull a genomic dataset with tens of thousands of variables, grab a ready-made pipeline off GitHub, run some machine-learning models, generate attractive figures, and draft a manuscript, with AI assisting at nearly every step.

That’s a real expansion of access. It’s also produced something new: careers that are publication-rich and knowledge-poor.

A researcher can have 30 papers in high-profile journals, great citation counts, international collaborators, and genuine fluency with terms like LASSO, random forests, XGBoost, SHAP values, Mendelian randomization, propensity scores, clustering, causal inference, AI -- the whole vocabulary.

But real competence tends to show up in real time, under questioning.

What’s the actual research question? What population are you drawing inferences about? What’s your estimand? How were subjects selected, and where might that introduce bias? Which variable is the confounder, and how do you know? Could adjusting for it introduce a collider instead? What does the model assume, and what breaks if that assumption fails? Why five clusters and not four? Why should I believe this association is causal?

That’s usually where the gap between a polished publication record and actual methodological depth becomes visible.

AI can widen that gap considerably. In the past, running a sophisticated analysis generally required enough statistical grounding to do it. Now, producing something that looks sophisticated mostly requires knowing how to prompt well.

That’s a warning about our training, not just a fact about technology.

A matter of national importance

Universities chase rankings; academic systems reward publishable, interesting results. Modern software offers hundreds of routes to such results, and AI accelerates the search considerably. Methodology risks becoming the servant of publication instead of the servant of the question.

The consequence I worry about most is a country’s scientific and technological competitiveness.

A country can post rapidly rising publication counts, papers in prestigious journals, thousands of researchers with impressive titles. Look from a distance, all of that like national scientific strength. But publication output and technological capability are not the same thing.

Software-driven research tends to produce highly capable users of tools built by others. But these people have much less capacity to build the next method, algorithm, instrument, or platform themselves -- things I call ‘core technologies’.

By ‘core technologies’ I mean the underlying scientific, mathematical, methodological, and computational capabilities. Understanding these principles well enough can help invent something new.

Statistical theory is one of those core capabilities. So is experimental design, causal reasoning, algorithm development, measurement, instrumentation, mathematics, computation, and deep substantive knowledge of the field itself.

Training built primarily around software use carries a real national risk: lots of people who can operate the technology, far fewer who could build it. A nation full of sophisticated technology users isn’t automatically a nation of innovators.

Way out: rebuilding methodological competence

I don’t think the fix is recreating the biostatistics training of 40 years ago. A biology or medical student doesn’t need to prove the Lehmann–Scheffé theorem or write Fortran codes to analyse data. What they need is enough conceptual depth to understand what a method is actually doing, why it’s appropriate, when it fails, and what the study design can and can’t support.

A useful sequence for a modern quantitative researcher to internalize:

research question →
target population →
study design →
measurement →
data-generating process →
bias and confounding →
probability and uncertainty →
statistical model →
computation →
interpretation.

The order isn’t incidental. Before fitting a model, understand the process that generated the data — censoring, risk sets, what a proportional-hazards assumption — actually requires. Before clustering genomic data, understand distance metrics, scaling, dimensionality, batch effects, cluster stability. Before reaching for machine learning, understand overfitting, data leakage, calibration, external validation, and who exactly the “target population” is.

Training should start from real scientific problems, not software menus: how were the data generated, how were variables measured, where might bias creep in, what quantity are you actually trying to estimate, which model fits that question, how should it be evaluated, and what do the results actually mean.

For any method, a researcher should be able to answer three things: what problem it solves, why it works, and when it can mislead you.

Simple models are worth keeping around as benchmarks. Try logistic regression before XGBoost. Establish a conventional baseline before a deep neural net. Look at the basic structure of the data before running elaborate clustering. Draw the causal diagram before fitting a complicated causal model. Complexity should have to earn its place.

AI, in this picture, should be taught as an assistant whose output always needs auditing. Understand study design before asking AI to design a study. Understand bias and confounding before asking which variables to adjust for. Understand a statistical model before asking AI to write the code. Understand inference before asking AI to interpret your confidence intervals and P-values.

In an AI-saturated field, the more valuable skill may not be prompt engineering, but AI auditing. Anyone can ask an AI to develop a model or perform an analysis, but true expertise lies in critically evaluating the output and determining whether the results are valid, reliable, and worthy of trust.

Publication counts shouldn’t stand in for methodological competence. A quantitatively heavy paper or thesis should probably include something like a methodological viva.

Remove the laptop and the AI assistant. Put the candidate in front of a whiteboard. Ask them to explain their model and write down the relevant equations. Ask them to draw the causal structure and identify where bias could arise. Ask why they trust the result, and what evidence would cause them to change their conclusion.

I strongly believe that theory remains fundamentally important. Teach students enough theory to understand how a method works, enough methodology to question and scrutinize the analysis, and enough substantive science to judge whether the results make sense. Once those foundations are in place, students can make effective and responsible use of even the most powerful AI tools.

Every generation of scientists inherits more powerful tools than the last. This generation has AI that can write code, run analyses, search the literature, and draft papers with extraordinary speed. That should make science better. Whether it does depends on whether we still teach students to ask why.

References

[1] Feynman RP. Cargo Cult Science. Caltech commencement address, 1974.

[2] Stark PB, Saltelli A. Cargo-cult statistics and scientific crisis. Significance, 26/7/2018. https://doi.org/10.1111/j.1740-9713.2018.01174.x

[3] Wasserstein RL, Lazar NA. The ASA’s Statement on p-Values: Context, Process, and Purpose. The American Statistician. 2016;70:129–133.

[4] Kerr MK, Churchill GA. Bootstrapping cluster analysis: assessing the reliability of conclusions from microarray experiments. Proceedings of the National Academy of Sciences. 2001;98:8961–8965.

[5] Silberzahn R, et al. Many analysts, one data set: Making transparent how variations in analytic choices affect results. Advances in Methods and Practices in Psychological Science. 2018;1:337–356.

PS: After drafting this article, I asked two AI tools to assess whether the text was AI-generated or human-written. Interestingly, the first tool confidently concluded that the article was 100% AI-generated, whereas the second assessed that it was written by someone with a strong understanding of research methodology. This contrasting assessment highlights an important limitation of AI-detection tools.

That conclusion is correct if the question is simply which country contributes the largest total Share to the publications covered by the Nature Index. It becomes less certain, however, once we ask a different and arguably more informative question: which country produces more high-quality scientific output than would ordinarily be expected for a nation of its size?

The Nature Index publishes raw, unnormalised data, and explicitly acknowledges that its figures do not account for the size of a country, its scientific workforce, its research expenditure or the composition of its research fields. Once national size is incorporated into the analysis, the ranking changes substantially, and a model that allows scientific output to scale flexibly with population places the United States, rather than China, in first position.

This does not mean that the official Nature Index ranking is statistically incorrect. It means that raw output and performance relative to national size are different quantities, and that each answers a different question.

What is the Nature Index?

The Nature Index is a database that tracks the institutional and national affiliations of authors who publish in a selected group of highly regarded scientific publications. The journals and conference proceedings included in the Index are chosen through surveys and panels of active researchers, who are asked where they would most like to publish their best work, rather than being selected solely through citation measures such as the journal impact factor. The current Index covers 177 journals and 1 conference proceeding across the natural, health, applied and social sciences.

The Nature Index reports research output using two related measures, known as Count and Share, and the distinction between them is important because they describe different aspects of scientific participation.

Count

Count is the number of Nature Index articles in which a country has at least one affiliated author. If a paper includes authors from the United States, Australia and Vietnam, each of those countries receives a Count of 1, regardless of whether the country contributes one author or twenty authors.

The same internationally collaborative paper can therefore contribute to the Count of several countries at the same time. Count is easy to understand because it measures participation, but it cannot be added across countries to obtain the total number of unique papers.

Share

Share is a fractional measure that attempts to divide the contribution from each paper among the participating authors, institutions and countries. Every paper has a maximum total Share of 1, which is allocated according to the authorship and institutional affiliations represented in that paper, under the simplifying assumption that all authors contributed equally.

Suppose that a paper has ten authors, seven from the United States and three from Australia, with no multiple affiliations. The United States would receive a Share of approximately 0.7 and Australia a Share of approximately 0.3, although both countries would receive a Count of 1.

Share is therefore a more refined measure of national contribution than Count, because it prevents one internationally collaborative paper from being counted in full for every participating country. It remains, nevertheless, a measure of publication output within a selected group of journals, rather than a complete measure of scientific quality, discovery, innovation or social impact.

The current Nature Index ranking

The 2026 Research Leaders table is based on publications from 1 January to 31 December 2025, even though the research year represented in the columns is 2025. The official country ranking is ordered by raw Share, without adjustment for population or any other measure of national size.

The ten leading countries are:

By this measure, China is unquestionably number one, because its total fractional contribution to Nature Index publications is approximately twice that of the United States and about nine times that of Germany.

The result is impressive and deserves attention, particularly because China’s Share increased by more than 22% between 2024 and 2025, whereas the corresponding increase for the United States was approximately 4%. Nevertheless, the official table answers only the question of total contribution, and a country with 1.4 billion inhabitants would ordinarily be expected to produce more research than a country with 10 million inhabitants, just as a university with 10,000 researchers would ordinarily produce more papers than a university with 500 researchers.

Nature Index itself makes this limitation clear. Its data are deliberately presented in an unnormalised form, and the organisation advises users to combine them with other information when assessing research performance.

Ranking countries by Share per population

The simplest way to account for national size is to divide Share by population:

Share per million pop = Share / (Population / 1000000)

Using 2025 population estimates from the United Nations World Population Prospects 2024, the ranking changes dramatically.

Small but research-intensive countries rise rapidly, while the largest countries fall:

China, which ranks first on raw Share, falls to approximately twentieth place when Share is divided directly by population. The United States moves from second to seventh, while Switzerland, Singapore and Denmark become the three leading countries.

This per-capita ranking reveals something real and important, because it shows that several relatively small nations produce an exceptionally large amount of high-quality research for their population size. Switzerland and Singapore are not statistical accidents; both have highly developed research universities, strong international networks and concentrated investment in science.

At the same time, the per-capita calculation imposes a demanding assumption that is rarely stated explicitly: a country with twice as many people is expected to produce exactly twice as much scientific output. In mathematical terms, the calculation assumes that Share grows in direct proportion to population.

That assumption is unlikely to describe how national research systems actually operate. Scientific output depends on the number of active researchers, the amount of research expenditure, the presence of major universities and institutes, the structure of the economy, historical investment, field composition and international collaboration, none of which necessarily increases in exact proportion to the total population.

Dividing Share directly by population may therefore over-reward small countries and over-penalise large ones, even though it remains a useful descriptive measure of scientific output per head of population.

A negative binomial model provides a more flexible adjustment

A more defensible method is to estimate from the data how scientific output changes as population increases, rather than assuming in advance that a doubling of population must produce a doubling of research output.

Because Share is a non-negative integer (I rounded it up) and shows substantially more variation between countries than a simple Poisson model can accommodate, I fitted a negative binomial regression of the form:

Y_i ~ NegativeBinomial(mu_i, theta_i)

log(mu_i) = alpha + beta*log(P_i)

In this model, Y_i is the observed Nature Index Share for country i, P_i is its population, and mu_i is the Share expected for a country of that size. Most importantly, the population coefficient beta is estimated from the data, rather than being fixed at 1 as it effectively is in a conventional per-capita analysis.

The estimated value of was approximately 0.52, meaning that a doubling of population was associated with an increase in expected Share of:

2^0.52 = 1.44

Scientific output therefore increased with population, but much less than proportionally. A country with twice the population was expected to produce approximately 44% more Nature Index articles, rather than 100% more.

For each country, I calculated an observed-to-expected ratio:

O/E = Observed Share / Expected Share

A ratio of 1 means that a country produces approximately the amount predicted for its population. A ratio of 2 means that it produces twice the expected amount, while a ratio of 0.5 means that it produces only half the expected amount.

The resulting top ten were:

Under this model, the United States moves to first place because its Count is approximately four times the level expected for a country of its population. China remains an extraordinary scientific performer, producing approximately three times the modelled expectation, but it ranks third behind the United States and Switzerland.

This model-based ranking produces a more balanced result than the per-capita calculation, because it recognises the outstanding performance of small countries without assuming that total scientific output must rise in exact proportion to population. It also recognises the enormous research scale of the United States and China, while asking whether their observed output is unusually high relative to countries of comparable size.

The full list of 50 countries is as follows:

There is one technical distinction worth making. Negative binomial regression is appropriate for Count because Count is an integer outcome, whereas Share is theoretically a positive continuous quantity. When I repeated the model-based analysis using Count as the outcome and a Gamma regression with a logarithmic link, the United States again ranked first, followed by Switzerland and China, which suggests that the main conclusion is not dependent on choosing Count rather than Share.

Which country is really number one?

The answer depends entirely on the question being asked.

If the question is: Which country contributes the largest total Share to Nature Index publications?

The answer is China.

If the question is: Which country produces the largest Share per person?

The answer is Switzerland.

If the question is: Which country produces the greatest Nature Index output relative to the level expected for a country of its population?

The answer, under the model described here, is the United States.

These findings are not mutually contradictory, because they describe three distinct dimensions of national scientific performance. Raw Share measures total scale, Share per population measures per-capita intensity, and the observed-to-expected ratio measures how strongly a country performs relative to a data-derived expectation for its national size.

The difficulty arises when the raw Share ranking is described as though it were a complete and size-adjusted measure of scientific strength. China’s first position in the official Nature Index table is valid, but it reflects the largest total contribution rather than the largest contribution relative to population.

Conclusion

China’s rise in the Nature Index is a major development in global science, and its raw Share confirms that it now contributes more to the publications covered by the Index than any other country. That achievement should neither be minimised nor explained away through statistical adjustment.

Yet raw totals inevitably favour countries with large scientific systems, and the Nature Index does not claim that its official table has been normalised for national size. Once population is incorporated through a model that estimates how research output actually scales across countries, the United States emerges as the leading performer, with an observed output substantially greater than the amount predicted for a country of its population.

The most accurate conclusion is therefore more nuanced than the headline that China has simply replaced the United States as the world’s scientific leader. China now leads in total Nature Index Share, Switzerland leads in Share per person, and the United States leads in model-adjusted performance relative to population.

Each ranking is legitimate within its own terms, but none should be interpreted beyond the question it was designed to answer.

A more comprehensive international comparison would ideally adjust not only for population, but also for the number of researchers, national expenditure on research and development, GDP, field composition, institutional capacity and the extent of international collaboration. Until such data are incorporated, population-adjusted models should be viewed as an improvement over raw ranking, rather than the final word on national scientific performance.

Even with that qualification, one conclusion is clear: China ranks first in absolute Share, but when scientific output is evaluated relative to modelled national size, the United States remains number one.

___

[1] https://www.nature.com/nature-index/research-leaders/2026/country/all/global

That was the response from my neighbour, a woman in her 70s, when she saw her estimated fracture risk.

On paper, she was right. Fifteen percent can look small. But for 100 people like her, it means about 15 may sustain a fracture within the next five years. And for some, that fracture could mean surgery, chronic pain, loss of independence, residential care, another fracture, or even premature death.

That moment stayed with me.

It helped shape the thinking behind my recent paper in Current Osteoporosis Report, Beyond Fracture Probability: Communicating the Full Consequences of Fracture and Contextualization.

For many years, osteoporosis medicine has become very good at calculating risk. But I believe we need to become much better at explaining what that risk means.

A percentage alone is not enough.

Patients need to know:

• What could a fracture mean for my life?

• Could it affect my mobility or independence?

• Can anything be done to reduce my risk?

• How would treatment change the picture?

In the paper, I discuss three gaps in current fracture risk communication:

• The consequence gap: we often communicate the probability of fracture, but not what a fracture may mean.

• The controllability gap: we tell people they are at risk, but not always how that risk can be reduced.

• The format gap: we rely heavily on percentages, when many people understand risk better through natural frequencies, visual formats, and more meaningful context.

The aim is to make risk information honest, clear, and useful, not to frighten patients.

A fracture risk estimate should not be a lonely number. It should be connected to consequences, prevention, and decisions that matter to the person sitting in front of us.

I’ve written more about the story behind the paper here:

https://communities.springernature.com/posts/the-number-isn-t-the-story-rethinking-how-we-talk-about-fracture-risk?channel_id=behind-the-paper

Although the Methods section is often considered the easiest to write (because it draws directly from your research protocol) it still demands careful attention. Your protocol likely details your study design, participant criteria, measurements, and analysis plan. However, what you write in your paper must reflect not just what you intended to do, but what you actually did. This may involve updating your protocol description to include changes in data collection or statistical techniques, especially if new elements were introduced during analysis.

A strong Methods section is grounded in four key principles: internal and external validity, reproducibility, evidentiary support, and clarity.

Validity

Validity is central to scientific writing. Internal validity refers to the accuracy and reliability of the study’s measurements, procedures, and analyses. Without internal validity, the results of a study become questionable.

External validity refers to whether the findings can be generalized to other populations, settings, or circumstances. For example, results from a single hospital may not apply directly to another country, healthcare system, or patient population. In medical research, internal validity should always come first. A study must first be trustworthy within its own design before its findings can be applied more broadly.

For this reason, the Methods section should provide enough detail for readers to judge both the reliability of the study and the relevance of its findings to other contexts.

Reproducibility

Reproducibility is one of the foundations of science. A good Methods section allows other researchers to understand exactly how the study was conducted and, where possible, to repeat it using the same procedures.

This does not mean every minor detail must be included. Rather, the Methods section should contain all information necessary to understand the study design, setting, participants, measurements, outcomes, and statistical analysis. Ambiguous descriptions can weaken confidence in the findings, even when the results appear impressive.

For example, simply writing:

“This was a cross-sectional study.”

does not provide enough information. A stronger description would be:

“The study was designed as a cross-sectional investigation, in which 210 women were randomly sampled by the cluster sampling scheme. The women aged between 20 and 85 years. The study setting was in Saigon, Vietnam.”

This version is stronger because it identifies the study design, sample size, sampling method, age range, and setting.

Evidentiary Support

Methods should not only be described; they should also be supported when necessary. For widely used techniques, a brief description may be sufficient. However, for less common procedures, new measurement tools, or advanced statistical methods, authors should provide appropriate references.

For example, if a study uses bootstrap methods, Bayesian Model Averaging, multiple imputation, or a newly developed prediction model, the Methods section should cite relevant sources. This allows readers to evaluate the appropriateness of the method and explore it further.

In some cases, transparency can be improved by providing supplementary materials, computer code, algorithms, or detailed protocols. These additions are especially useful when the analysis is complex or when reproducibility is a major concern.

Clarity

A strong Methods section is both concise and comprehensive. The goal is not to include every possible detail, but to include the details that matter.

Authors should clearly specify the study design, participants, setting, procedures, measurements, outcomes, predictors, confounders, sample size, statistical methods, and software used. If missing data were handled in a particular way, variables were transformed, or analyses were stratified, these decisions should be stated clearly.

Consistency is also essential. If an outcome is called “30-day mortality” in the Methods section, the same term should be used in the Results and Discussion. Avoid renaming variables across sections, as this can confuse readers. Before writing, it is useful to list key terms, outcomes, and variables to ensure consistency throughout the manuscript.

What Should Be Included in the Methods Section?

The Methods section answers one central question: What did you do?

For clinical and epidemiological studies, the section is commonly organized under several subheadings:

• Study Design

• Study Setting

• Participants

• Study Procedures

• Data Collection

• Outcomes and Variables

• Sample Size

• Randomization and Blinding, if applicable

• Statistical Analysis

• Ethical Approval

Using clear subheadings helps readers find information quickly. It also helps reviewers assess whether the study was conducted properly. A useful rule is that everything reported in the Results section should be traceable back to the Methods section.

Let’s begin with Study Design. It’s not enough to write, “This was a cross-sectional study.” This tells the reader almost nothing. A stronger version adds specific context, such as:

“The study was designed as a cross-sectional investigation of 300 post-menopausal women aged 50 to 85 years in Can Tho, Vietnam. Participants were enrolled using a stratified random sampling method.”

For more complex settings, especially hospital-based studies, provide information about the facility itself. For example:

“This study took place at ABC Hospital, a leading tertiary teaching hospital in Ho Chi Minh City, Vietnam, serving southern Vietnam. With over 100 years of history and 1,500 beds, it handles high patient volumes. It has seven intensive care units (ICUs), including medical (20 beds, ~2,500 admissions/year), surgical (65 beds, ~19,000 admissions/year), and stroke (6 beds). Data were collected in ICUs from November 2014 to September 2015.”

If the setting is unfamiliar to most readers—like a regional hospital in Vietnam—additional background may be helpful:

“This study was conducted at Can Tho National Hospital (development cohort) and Can Tho General Hospital (validation cohort), Vietnam. The National Hospital, a tertiary teaching facility, serves 17 million people in the Mekong Delta, with its emergency department (ED) admitting ~75 non-surgical patients daily. The General Hospital, with 400 beds, serves Can Tho City, admitting ~50 non-surgical ED patients daily.”

This description is helpful because it tells readers not only where the study was conducted, but also the size of the population served, the type of hospitals involved, and the patient volume.

Next, turn to Participants. Clear eligibility criteria help readers assess the external validity of your study. A strong Participants subsection usually includes the age range, disease or condition of interest, diagnostic criteria, inclusion criteria, exclusion criteria, recruitment source, and consent process. For example, in a major osteoporosis trial (Cummings et al., NEJM, 2009), participant selection was described as follows:

“Women between the ages of 60 and 90 years with a bone mineral density T score of less than −2.5 at the lumbar spine or total hip were eligible for inclusion. Women were excluded if they had conditions that influence bone metabolism or had taken oral bisphosphonates for more than 3 years. If they had taken bisphosphonates for less than 3 years, they were eligible after 12 months without treatment. Women were also excluded if they had used intravenous bisphosphonates, fluoride, or strontium for osteoporosis within the past 5 years; or parathyroid hormone or its derivatives, corticosteroids, systemic hormone-replacement therapy, selective estrogen-receptor modulators, or tibolone, calcitonin, or calcitriol within 6 weeks before study enrollment.”

This example is effective because it defines the target population and explains the exclusion criteria in detail. Readers can easily understand which patients the findings apply to.

Another example comes from the Can Tho study, which clearly defines both inclusion and exclusion criteria (Ha et al Sci Reports 2017):

“We enrolled medical patients (non-trauma and non-surgical) from the two emergency departments between 13 March 2013 and 31 March 2014. The inclusion criteria were: all patients aged 16 years and older and who could give informed consent. Patients were excluded from the study if they had one of the following conditions: acute coronary syndrome, burns, cardiac arrest before admitting to the hospital or which occurred in the ED with failure of cardiopulmonary resuscitation, snakebite, insect bite or sting, poisoning (drugs, alcohol, intoxication, paraquat, insecticides, rodenticides, corrosive substances). We also excluded patients with burns, cardiac arrest with failure of cardiopulmonary resuscitation because these patients were deemed to be at high risk of mortality. Women in labor or patients dead on arrival were also excluded from the study. The study protocol and procedure were approved by the Can Tho National Hospital ethics committee.”

This example is useful because it specifies the recruitment period, clinical setting, inclusion criteria, exclusion criteria, and ethical approval.

Study Setting can also influence results, especially in clinical, epidemiological, and public health research. Geographic, environmental, cultural, and healthcare-system factors may all affect the interpretation of findings.

For example, in a vitamin D study, it is important to describe the geographical and climatic conditions because sunlight exposure is directly relevant to vitamin D status:

“The study was designed as a cross-sectional investigation, in which the setting was Ho Chi Minh City (formerly Saigon). The City is located at 10°45’N, 106°40’E in the southeastern region of Vietnam. The City is in the tropic and close to the sea; therefore it has a tropical climate, with an average humidity of 75%. There are only two distinct seasons: the rainy season, with an average rainfall of about 1,800 millimetres annually (about 150 rainy days per year), usually begins in May and ends in late November; the dry season lasts from December to April. The average temperature is 28°C (82°F), the highest temperature sometimes reaches 39°C (102°F) around noon in late April, while the lowest may fall below 16°C (61°F) in the early mornings of late December.”

This example shows how setting details can help readers understand the context of the research. In this case, climate and location are not background information only; they are relevant to the biological question being studied.

Study Procedures must be explained in a step-by-step manner. If patients were assigned to groups, received interventions, or completed questionnaires, describe how and when this occurred. For instance:

“Eligible patients provided written informed consent and completed a structured questionnaire administered by trained staff, covering demographics, medical history, physiological data, and lab tests (see Supplementary Table S1). At 30 days post-admission, staff contacted patients or relatives to confirm survival status. Patients hospitalized beyond 30 days were censored.”

For lab-based research, sample collection and processing should be documented carefully:

“Bedside specimens were collected and sent to the microbiology lab within 2 hours for culturing. Bronchoscopy used a 600 mm, 50 mm view-depth fiberscope (PortaView LF-TP, Olympus, Tokyo) with a video camera. The fiberscope was lubricated with sterile 2% xylocaine jelly (AstraZeneca, Sweden), and sedation used midazolam, fentanyl, and/or suxamethonium. Positioned 2 cm above the carina, 2–5 ml of bronchoalveolar lavage fluid was collected and analyzed within 30 minutes.”

This level of detail allows readers to understand exactly how the procedure was performed and whether the method was appropriate.

Measurement methods should be described precisely. If instruments or machines were used, authors should specify the model, manufacturer, and measurement conditions. This is especially important when measurement error could affect the results. For example:

“Blood pressure (diastolic phase 5) while patient was sitting and had rested for at least five minutes was measured by a trained nurse with a Copal UA-251 or a Takeda UA-751 electronic ausculatory blood pressure reading machine (Andrew Stephens, Brighouse, West Yorkshire) or with a Hawksley random zero sphygmomanometer (Hawksley, Lancing, Sussex) in patients with atrial fibrillation. The first reading was discarded and the mean of the next three consecutive readings with a coefficient of variation below 15% was used in the study, with additional readings if required.”

This example is strong because it describes the patient position, rest period, personnel, equipment, and method for selecting the final measurement. Such information helps readers judge the reliability of the data.

Clinical outcomes must be defined precisely. A vague outcome definition can make the results difficult to interpret.

For example, in one study, 30-day mortality was defined as follows:

“The primary outcome of the study was mortality within 30 days since the day of admission to the hospital. Mortality was defined as (1) death in hospital from any cause; (2) family-initiated discharge and death either on the way home or within 24 hrs after discharge; (3) doctor-initiated discharge and death at home. It should be noted that in Vietnamese culture, when a patient was in the end stage of disease, the patient or family often requests for discharge from hospital because they prefer to pass away at home.”

This is a good example because it defines the primary outcome in a way that reflects the local clinical and cultural context. Without this explanation, some deaths might be misclassified or misunderstood by readers unfamiliar with the setting.

Authors should also specify predictors, exposures, confounders, and effect modifiers when relevant. Only variables that appear in the Results section should be described in detail. For example, if bone mineral density is analyzed as an outcome, the Methods section should describe how bone mineral density was measured. Unrelated variables should not be included unless they are used in the analysis.

Sample size justification is important, particularly for randomized controlled trials and many analytical studies. The Methods section should explain how the sample size was determined. A good sample size statement usually includes the expected event rate or effect size, significance level, statistical power, allocation ratio, and adjustment for loss to follow-up or non-evaluable participants.

For example:

“We consider that the incidence of symptomatic deep venous thrombosis or pulmonary embolism or death would be 4% in the placebo group and 1.5% in the ardeparin sodium group. Based on 0.9 power to detect a significant difference (p 0.05, two-sided), 976 patients were required for each study group. To compensate for nonevaluable patients, we planned to enroll 1000 patients in each group.”

This example clearly explains the assumptions behind the sample size and how the final target number was determined.

For randomized controlled trials, randomization and blinding should be described clearly. These details help readers assess whether selection bias and performance bias were minimized.

A randomization statement should explain how the allocation sequence was generated and whether blocking or stratification was used.

For example:

“Women had an equal probability of assignment to the groups. The randomization code was developed using a computer random number generator to select random permuted blocks. The block lengths were 4, 8, and 10 varied randomly.”

Blinding should also be reported:

“All study personnel and participants were blinded to treatment assignment for the duration of the study. Only the study statisticians and the data monitoring committee saw unblinded data but none had any contact with study participants.”

Together, these descriptions allow readers to understand how treatment allocation was concealed and how bias was reduced.

The data analysis subsection should describe how the data were analyzed. It should be detailed enough for readers to understand the analytical approach, but not so technical that it becomes unreadable.

This subsection should usually include descriptive statistics, statistical tests, regression models, adjustment variables, handling of missing data, subgroup or sensitivity analyses, significance level, and software used.

For example:

“All data analysis was carried out according to a pre-established analysis plan. Proportions were compared by using Chi-squared tests with continuity correction or Fisher’s exact test when appropriate. Multivariate analyses were conducted with logistic regression. The durations of episodes and signs of disease were compared by using proportional hazards regression. Mean serum retinol concentrations were compared by t-test and analysis of covariance ... Two-sided significance tests were used throughout. The analysis was performed with the R statistical language.”

In the Cummings et al’s NEJM osteoporosis study, the authors provided a comprehensive statistical plan:

“Analyses of efficacy were based on the intention-to-treat principle. To adjust for multiplicity and maintain the overall significance level at 0.05, the primary end point of new vertebral fracture was required to achieve significance before the next end points in the sequence (nonvertebral fracture and hip fracture) could be tested. Analyses regarding vertebral fractures included all subjects who had at least one follow-up radiograph.

The effect of treatment on the risk of new vertebral fracture was analyzed with the use of a logistic-regression model with adjustment for age strata. An age-stratified Cox proportional-hazards model was used to compare the two study groups for the secondary end points. Score tests were used to calculate P values in each model.14,15 Subjects who were lost to follow-up or withdrew before having a fracture event had their last known fracture status carried forward. Radiographically defined vertebral fractures were analyzed by cumulative incidence and secondary end points by time-to-event analysis with the use of Kaplan–Meier methods. The absolute risk reduction between study groups was computed as the difference in incidence at 36 months for the primary end point and the difference in the Kaplan–Meier estimates at 36 months for the secondary end points with the use of a weighted average across the age strata. Analyses of changes in bone mineral density included all subjects who had at least one follow-up measurement at or before the time point under consideration. Missing values were imputed by carrying forward the last observation.”

This example shows how a complex statistical plan can be reported clearly. It explains the analysis population, adjustment for multiplicity, regression models, time-to-event methods, handling of missing data, and calculation of risk reduction.

For studies involving human participants, ethical approval should be reported. This usually includes the name of the approving ethics committee or institutional review board, the approval number if available, and the consent procedure.

For example:

“The study protocol and procedure were approved by the Can Tho National Hospital ethics committee.”

If written informed consent was obtained, this should be stated. If consent was waived, the reason should also be explained.

Common Mistakes

Several common mistakes can weaken a Methods section. The first is vagueness. Statements such as “patients were randomly selected” or “data were analyzed using standard methods” are not sufficient. Readers need to know how patients were selected and which methods were used.

The second is inconsistency. Variables and outcomes should be named consistently throughout the manuscript. A term introduced in the Methods section should not be renamed in the Results section.

The third is mixing results into the Methods section. The Methods section should describe what was done, not what was found.

The fourth is omitting important analytical decisions. Authors should explain how missing data were handled, which covariates were included in models, and whether analyses were planned in advance.

The fifth is including irrelevant details. A good Methods section is complete, but not cluttered. Details that are useful but too long for the main text can be placed in supplementary materials.

Conclusion

The Methods section may be the longest section of a scientific paper, often two to three times the length of the Introduction, but it is also one of the most scrutinized. Reviewers look carefully at the Methods section because it reveals whether the study was valid, reproducible, and transparent.

A strong Methods section explains the study design, setting, participants, procedures, measurements, outcomes, sample size, randomization, blinding, and statistical analysis with precision and consistency. It avoids vague descriptions, unnecessary detail, unsupported methods, and the temptation to report results prematurely.

Most importantly, the Methods section allows readers to judge the integrity of the study. A well-written Methods section does more than describe procedures; it shows that the findings rest on a sound scientific foundation.

Most scientific papers follow the IMRaD format—Introduction, Methods, Results, and Discussion. Within this structure, the introduction plays a crucial role: it sets the scene for your research and answers the foundational question, “Why did I do this study?” This is your opportunity to convince readers that your work fills a genuine need in the field. One widely adopted model for structuring an introduction is the “Create a Research Space” (CaRS) model, proposed by linguist John Swales [1]. The CaRS model breaks the introduction into three logical steps: setting the context, spotlighting a gap, and claiming that gap through a clear research objective.

The first step, setting the context, involves outlining the broader issue your study addresses. Here, you identify the problem, explain its relevance, and summarize what is already known. For example, if your research focuses on osteoporosis-related genes, you might begin by discussing the prevalence of osteoporosis and its impact on public health. Since a term like “fractures” may not resonate strongly with readers, it can help to underscore the condition’s broader implications—such as its effect on life expectancy or healthcare systems—perhaps even comparing it to more widely recognized illnesses like cancer. After establishing the significance of the topic, you would then provide a concise summary of existing research, identifying key findings or dominant theories to set the stage for your own study.

Once the context is established, the next step is to identify the knowledge gap your research addresses. This is where you point out what remains unresolved, underexplored, or contested in the current literature. For instance, in the case of osteoporosis, you might explain that while several genes have been linked to the disease, they account for only a small percentage of the variation seen among patients—suggesting that other, undiscovered genetic factors may be involved. Similarly, if your research involves a meta-analysis on beta-blockers and fracture risk, you might highlight the inconsistency of past results, often due to small sample sizes, and argue that a more comprehensive analysis is needed. This is arguably the most important part of your introduction, as it justifies the relevance and necessity of your work.

With the gap clearly defined, the final step is to claim that gap by presenting your research objective or hypothesis. This might take the form of a single guiding question, or several related goals. In some disciplines, particularly in the social sciences, it’s also common to briefly outline the paper’s structure or highlight key findings, though this is less typical in biomedical research. Regardless of the format, your introduction should always conclude with a clear, specific statement of what your study aims to do. Without this, the paper can appear unfocused, and readers may struggle to understand the purpose behind your work.

Writing an effective introduction is not only about structure, but also about tone and approach. One helpful method is the “3C” strategy: Citation, Critique, and Constructive contribution.

• Citation: Back up key claims or data with references from trusted, peer-reviewed journals. Focus on studies from the last 5–10 years, and avoid predatory journals or non-peer-reviewed sources (like some local publications in Vietnam).

• Critique: When highlighting gaps, be diplomatic. Never say others are “wrong.” Instead, suggest alternative interpretations or note that later studies couldn’t replicate earlier results. Keep critiques constructive, showing how your work advances the field.

• Constructive: Frame your study as a positive contribution, not a jab at others. Stay professional and avoid sounding preachy.

A well-written introduction in a medical research paper is usually about two double-spaced pages long. Writing too much can come off as unfocused or excessive; writing too little may signal a lack of preparation or engagement with the literature. Many experienced researchers find it helpful to write the introduction after completing the results and discussion sections. By then, the main contribution of the paper is clearer, and it becomes easier to frame the context and rationale in a compelling way.

To illustrate these principles, consider this sample introduction for a study examining obesity diagnostic criteria for Asian populations (Ho-Pham et al. Obesity 2012) [2]:

Although obesity is recognized as a global public health problem, the extent of obesity is a matter of contention, due largely to a lack of consensus regarding definition. Clinically, obesity is defined as a condition characterized by excessive body fat to the extent that it is harmful to well being and health (1). Currently, the operational definition of obesity is based on body massindex (BMI). According to the World Health Organization criteria, any individual whose BMI is greater than or equal to 30 kg/m2 is considered obese (2). Although BMI is widely used in the diagnosis of obesity, it has been criticized, because it does not distinguish between fat mass, muscle mass, bone and vital organs (3-8).

It has been argued that a better classification of obesity should be based on percent body fat (PBF), in which any woman whose PBF >35% and any man whose PBF >25% is considered obese (9). Using the relationship between BMI and PBF, it has been suggested that in Asian populations, a BMI greater or equal to 25 should be classified as obese (10), because a BMI of 25 kg/m2 is assumed to correspond to about 25% and 35% body fat for Asian men and women, respectively (9). This classification is based on the assumption that for a given BMI, Asians have greater PBF than Caucasians (11, 12). However, a close examination of the data on which this assumption is based on (12) reveals little difference in PBF between Chinese in New York and Caucasian women. In this paper, we examine the validity of this assumption by comparing PBF between White American women of European ancestry and Vietnamese women living in Vietnam.

In this example, the introduction starts with the broader issue—obesity and its definition—then narrows to highlight a specific gap in how BMI cutoffs are applied to Asian populations. It concludes by presenting a clear objective: to test a widely held assumption through cross-cultural data.

Ultimately, your introduction must answer the question, “Why does this study matter?” The CaRS model offers a helpful roadmap: begin with context, move to the gap, and finish with your research aims. Along the way, use the 3C approach—cite reliable sources, critique thoughtfully, and offer a constructive perspective—to ensure your introduction is both informative and engaging. A well-crafted introduction sets the tone for the rest of your paper and positions your research as a meaningful contribution to the scientific conversation.

In the end, writing a strong introduction is an exercise in clarity of thought. It requires you to step back, understand the broader landscape of your field, identify what is missing, and articulate precisely how your study addresses that gap. By following a logical structure such as the CaRS model and adopting the 3C approach—citing carefully, critiquing diplomatically, and contributing constructively—you transform the introduction from a routine formality into a persuasive argument for your research.

A well-crafted introduction does not merely inform; it positions your study within an ongoing scientific dialogue. When readers finish your introduction, they should understand not only what you intend to investigate, but why it matters and why it deserves attention. Mastering this skill takes practice, but once achieved, it strengthens every manuscript you write and lays a solid foundation for meaningful scientific contribution.

[1] Swales JM. Genre Analysis: English in Academic and Research Settings. Cambridge: Cambridge UP 1990.

[2] Ho-Pham LT, Lai TQ, Nguyen ND, Barrett-Connor E, Nguyen TV. Similarity in Percent Body Fat Between White and Vietnamese Women: Implication for a Universal Definition of Obesity. Obesity 2012;18:1242-1246.

The Australian government has declared this a terrorist attack.

This was not just a terrorist act, but a deliberate massacre targeting the Jewish community.

The two perpetrators were father and son, Naveed Akram (24 years old) and Sajid Akram (50 years old), of Pakistani origin, living in Bonnyrigg, Sydney [2].

They fired from a pedestrian overpass into the crowd joyfully lighting the menorah candles. One of them was killed by police at the scene; the other was seriously injured and taken into custody.

Amid that hell, the image of a hero emerged: Ahmed Al Ahmed. This ordinary man courageously rushed in to disarm one of the gunmen and was shot and seriously injured. His actions deserve the highest praise, reminding us that courage can shine even in the darkest hatred.

Bondi is Sydney’s most famous beach. Anyone who visits Sydney without going to Bondi hasn’t truly experienced the city. It is also a gathering place for the Jewish community. The terrorists deliberately chose a spot crowded with Jewish people.

With only about 117,000 members (0.46% of Australia’s population), the Jewish community has made enormous contributions to the country: in commerce, science, medicine, politics, and the military. Names like Monash, Isaacs, Cowen, Dreyfus, Frydenberg, Lowy, Triguboff, and others are iconic symbols of Jewish leadership.

Despite settling here for hundreds of years, Jews in Australia still feel a spiritual connection to Israel. After the 7/10/2023 attack, they have increasingly become targets of hatred. Anti-Jewish protests, flag burnings, threats, and more have occurred relentlessly, yet the left-wing (Labor) government seemed to turn a blind eye, offering only soothing statements instead of decisive action.

■ The danger had been warned about for a long time

For many months, the Australian government received countless warnings about the rising wave of antisemitism. From weekly protests to minor attacks in Sydney and Melbourne. Yet the soothing words from officials and ministers in this left-wing government were like a breeze before the raging fire of antisemitism.

This weakness has allowed an atmosphere of hatred to spread unchecked. The tragic outcome at Bondi is the culmination, a deep wound for peaceful Australia.

We can only hope – just hope – that this moment will force the government to acknowledge their failures, strengthen protection for the Jewish community, and face reality.

■ “Asterisk Immigrants” and “Hotel Australia”

Australia, like America, is a nation of immigrants. But immigrants with an “asterisk.” What kind, you might ask? Those who immigrate legally and are determined to make this country prosperous and strong.

We Vietnamese came here carrying our value and history, but we are Australian citizens first and foremost. Our primary mission upon setting foot in Australia is to become better Australians than those born here. And we have achieved that. We join “Team Australia” (as former Prime Minister Tony Abbott put it).

That is the ideal of asterisk immigrants.

In recent years, Australia (like America, Britain, Germany, France, and Northern Europe) has experienced a new wave of immigration. Not as massive as in Western Europe, but still “mass immigration.” In this wave, many lack the ideals of the asterisk immigrant generation.

Many new immigrants do not join Team Australia. They are nurtured by Australia, enjoy a comfortable life, use Australian passports for travel – in short, benefit from Australia. But they see Australia only as a hotel – Hotel Australia. They do not contribute; on the contrary, some harbor hatred toward Australia and her people. A small number even return to their countries of origin to take up arms against Australia or its interests; and when facing obstacles, they demand to return to Australia. How ironic!

The terrorist Naveed Akram belongs to this new immigrant group: benefiting from Australia but turning against it.

And, it is precisely this mass immigration policy, combined with the behavior of new immigrants like Akram (who view Australia as a hotel), that alarms Australians and fuels the rise of extreme right-wing forces.

Australians have seen what happened in London, Paris, Berlin, and elsewhere now happening in Sydney and Melbourne. Communities are torn apart by parallel societies, closed religious communities, and government weakness. Australians like me have the right to ask: “Do we really want this to happen in Australia?”

These are the questions that must be asked:

· Who have we admitted into this country?

· How were they screened?

· Was there any serious vetting process?

· Or did we simply let people in out of compassion and politics?

And most importantly, how many more people like Naveed Akram are in Australia now?

■ The light will shine again

Peaceful Australia rarely witnesses tragedies like this. This event is a spiritual wound and a profound pain for the Jewish community and the entire nation.

I have many Jewish friends whom I mentioned in my memoirs Kangaroo Dreams. All three of my doctorate supervisors were Jewish. The closest mentor, whom I see as an older brother (and who sees me as a younger brother), is Jewish. My dearest lifelong friends and the most renowned figures in the world of osteoporosis are Jewish. The American friends who supported me in the American osteoporosis community and in my career are also Jewish.

One could say that Jews have brought light into my life. Therefore, their pain is my pain. I say this sincerely. I want to send this message to them: We will overcome the darkness, and the light of Hanukkah will shine again.

____

[1] https://www.abc.net.au/news/2025-12-15/nsw-sydney-bondi-beach-shooting-hanukkah-jewish-community/106141648

[2] https://www.theindiansun.com.au/2025/12/15/single-most-violent-and-deadly-act-of-antisemitism-in-australias-history-professor-slucki

Consider the following example of an abstract:

“This paper presents and assesses a framework for an engineering capstone design program. We explain how student preparation, project selection, and instructor mentorship are the three key elements that must be addressed before the capstone experience is ready for the students. Next, we describe a way to administer and execute the capstone design experience including design workshops and lead engineers. We describe the importance in assessing the capstone design experience and report recent assessment results of our framework. We comment specifically on what students thought were the most important aspects of their experience in engineering capstone design and provide quantitative insight into what parts of the framework are most important.”

While this abstract is structured and grammatically sound, it ultimately falls flat for several reasons. First, it lacks a clear point. It opens with a general goal but quickly drifts into vague procedural details, leaving readers unsure of the paper’s actual takeaway. Second, it’s overly self-referential, repeating phrases like “we explain,” “we describe,” and “we comment,” which makes the writing feel redundant and potentially irritating. Third—and perhaps most critically—it fails to present actual results. It describes what the paper does, but not what it finds. As the saying goes, it’s “all broth, no beef.” There’s no insight, no data, and no punchline.

The word “abstract” comes from Latin—ab meaning “away” and trahere meaning “to draw.” At its root, the abstract is a distillation. It should draw away the nonessential and leave the reader with a concentrated snapshot of your paper’s essence. A well-crafted abstract can communicate the purpose, methods, key findings, and conclusions of your study in just a few hundred words. This is vital: out of hundreds of people who may read an abstract, only a small fraction will actually read the full paper. As a former JAMA editor once noted, abstracts are often poorly done, despite being the most visible and widely read part of a paper. With journal limits typically ranging between 250 and 300 words, writing a strong abstract is no easy feat.

Principles

To help guide this process, a set of five core principles—originally inspired by Drs. Petrache and Diette from Thoracic—can serve as a framework for building a high-quality abstract. These principles are: novelty, impact, evidence, brevity, and clarity.

The first principle, novelty, emphasizes the need to show readers that your study offers something new. A sense of discovery should come through in the very first lines, ideally within the background or rationale. For example:

“Trabecular bone score has emerged as an important predictor of fragility fracture, but factors underlying the individual differences in TBS have not been explored. In this study, we sought to determine the genetic contribution to the variation of TBS in the general population.”

Or:

“It has been widely assumed that for a given BMI, Asians have higher percent body fat (PBF) than whites, and that the BMI threshold for defining obesity in Asians should be lowerthan the threshold for whites. Thisstudy sought to test this assumption by comparing the PBF between US white and Vietnamese women.”

Both examples set up a clear gap in knowledge and introduce the aim of the study, which immediately engages the reader.

The second principle is impact. Especially in medical or applied research, readers want to know why a study matters—what difference it could make. The study’s potential to change thinking or practice should be evident. This can be accomplished through brief but powerful statements, such as:

“These findings shed light on the genetic architecture and pathophysiological mechanisms underlying BMD variation and fracture susceptibility.”

Or:

“These data indicated that critically ill patients on ventilator in Vietnam were at disturbingly high risk of antimicrobial resistance”

Statements like these highlight the broader importance of the findings without straying into overstatement.

Next is evidence. A strong abstract includes specific data and conclusions that are directly supported by that data. It’s crucial not to misrepresent findings or make claims the study cannot support. For instance, if a study shows a 30% cancer risk reduction from calcium and vitamin D, it would be misleading to say there’s “no clinical impact,” as highlighted in the JAMA study by Lappe et al. (2017): “supplementation with vitamin D3 and calcium compared with placebo did not result in a significantly lower risk of all-type cancer at 4 years.“ Similarly, finding an association between smoking and cancer doesn’t justify claiming that banning smoking will eliminate cancer—unless the study includes an intervention proving that point. Good abstracts stay anchored to the data.

The fourth principle is brevity. Abstracts are compact summaries, often only 100 to 300 words long. That means every word must count. Think of the abstract like a news headline: concise, informative, and free of filler. No room exists for digressions or vague generalities.

The final principle, clarity, may be the most important. Scientific writing demands precision. Short, simple sentences and clear numbers help prevent misinterpretation. Consider this example:

“The study involved 1765 patients in the development cohort and 1728 in the validation cohort. The main outcome was mortality up to 30 days after admission.”

This version is better than a vaguer alternative like:

“The study involved two cohorts of patients from two independent hospitals. The main outcome was death after ICU admission”.

The first is specific and direct. The second is blurry and harder to interpret.

So, what makes a great abstract in practice?

It should function as a standalone piece—something that readers (and journals) can access independently from the full paper. In many submission portals, the abstract is uploaded separately. Think of it as a mini-paper or even a poem: highly compressed, intensely focused, and complete in itself. It should present a complete story snapshot, hitting all the major parts of your study—introduction, methods, results, and conclusion—in a cohesive arc. Results typically occupy the most space, but even they should be selective, featuring only key findings.

Importantly, results and conclusions must be included, not vague promises about what will be discussed later. Abstracts that lack hard data fail to communicate value. Include numbers—relative risks, odds ratios, sample sizes—along with a firm conclusion and, if space allows, a comment on significance or impact. Also, be mindful of word limits. Most journals cap abstracts at 100 to 300 words, including spaces. Some even ask for a “mini-abstract” or highlight, restricted to just 200 characters. Boiling down a multi-year research project into such a short form is difficult, but doing so can sharpen your thinking and writing.

Abstracts come in two basic formats: unstructured and structured. Unstructured abstracts are one continuous paragraph that summarizes the entire study. Structured abstracts divide the text into labeled sections—Background, Aims, Methods, Results, and Conclusion—making it easier to scan. Regardless of the format, the same core elements must be included.

Start with the research question and aim. This usually takes two sentences: one framing the problem in light of current knowledge, and one stating the specific objective of your study. For example:

“Antimicrobialresistance has emerged as a major concern in developing countries. The present study sought to define the pattern of antimicrobial resistance in 50 ICU patients with ventilator-associated pneumonia.”

Next comes the methods section, typically in four to five sentences. Describe the study design, participants, setting, measurements, risk factors, and outcomes. For instance:

“Between November 2014 and September 2015, we enrolled 220 patients (average age ~71 yr) who were admitted to ICU in a majortertiary hospital in Ho Chi Minh City, Vietnam. Data concerning demographic characteristics and clinical history were collected from each patient. The Bauer–Kirby disk diffusion method was used to detect the antimicrobial susceptibility.”

Or:

“The study involved 1765 patients in the development cohort and 1728 in the validation cohort. The main outcome was mortality up to 30 days after admission. Potential risk factors included clinical characteristics, vital signs, and routine haematological and biochemistry tests. The Bayesian Model Averaging method within the Cox’s regression model was used to identify independent risk factors for mortality.”

Then, present the results, emphasizing key numerical findings that directly answer your research question. For the resistance study, this might be:

“Antimicrobial resistance was commonly found in ceftriaxone (88%), ceftazidime (80%), ciprofloxacin (77%), cefepime (75%), levofloxacin (72%). Overall, the rate of antimicrobial resistance to any drug was 93% (n = 153/164), with the majority (87%) being resistant to at least 2 drugs. The three commonly isolated microorganisms were Acinetobacter (n = 75), Kebshiella (n = 39), and Pseudomonas aeruginosa (n = 29). Acinetobacter baumannii were virtually resistant to ceftazidime, ceftriaxone, piperacilin, imipenem, meropenem, ertapenem, ciprofloxacin and levofloxacin. High rates (>70%) of ceftriaxone and ceftazidime-resistant Klebsiella were also observed.”

For a genetic study, a different kind of data might be summarized:

“We identified 56 loci (32 novel)associated with BMD atgenome-wide significant level (P <5x10^-8). Several of these factors cluster within the RANK-RANKL-OPG, mesenchymal-stem-cell differentiation, endochondral ossification and the Wnt signalling pathways. However, we also discovered loci containing genes not known to play a role in bone biology.”

The abstract should end with a conclusion, one or two sentences that interpret your results and state their importance. For instance:

“We conclude that the risk of mortality among ED patients could be accurately predicted by using common clinical signs and biochemical tests.”

To see these principles in action, consider this structured abstract from a study on body composition in Vietnamese women (BMC Musculoskeletal Disorders, 2010, Ho-Pham et al.):

“Background. The relative contribution of lean and fat to the determination of bone mineral density (BMD) in postmenopausal women is a contentious issue. The present study was undertaken to test the hypothesis that lean mass is a better determinant of BMD than fat mass.

Methods. This cross-sectional study involved 210 postmenopausal women of Vietnamese background, aged between 50 and 85 years, who were randomly sampled from various districtsin Ho Chi Minh City (Vietnam). Whole body scans, femoral neck, and lumbar spine BMD were measured by DXA (QDR 4500, Hologic Inc., Waltham, MA). Lean mass (LM) and fat mass (FM) were derived from the whole body scan. Furthermore, lean mass index (LMi) and fat massindex (FMi) were calculated asratio of LM or FM to body height in metre squared (m2).

Results. In multiple linear regression analysis, both LM and FM were independent and significant predictors of BMD at the spine and femoral neck. Age, lean mass and fat mass collectively explained 33% variance of lumbarspine and 38% variance of femoral neck BMD. Replacing LM and FM by LMi and LMi did not alter the result. In both analyses, the influence of LM or LMi was greater than FM and FMi. Simulation analysis suggested that a study with 1000 individuals has a 78% chance of finding the significant effects of both LM and FM, and a 22% chance of finding LM alone significant, and zero chance of finding the effect of fat mass alone.

Conclusions. These data suggest that both lean mass and fat mass are important determinants of BMD. For a given body size -- measured either by lean mass or height -- women with greater fat mass have greater BMD.”

Or this unstructured abstract, from a study comparing body fat in white and Vietnamese women (Ho-Pham et al. Obesity, 2010):

“[Background] It has been widely assumed that for a given BMI, Asians have higher percent body fat (PBF) than whites, and that the BMI threshold for defining obesity in Asians should be lower than the threshold for whites. [Aim] This study sought to test this assumption by comparing the PBF between US white and Vietnamese women. [Methods] The study was designed as a comparative cross-sectional investigation. In the first study, 210 Vietnamese women ages between 50 and 85 were randomly selected from various districts in Ho Chi Minh City (Vietnam). In the second study, 419 women of the same age range were randomly selected from the Rancho Bernardo Study (San Diego, CA). In both studies, lean mass (LM) and fat mass (FM) were measured by dual-energy X-ray absorptiometry (DXA) (QDR 4500; Hologic). PBF was derived as FM over body weight. [Results] Compared with Vietnamese women, white women had much more FM (24.8 ± 8.1 kg vs. 18.8 ± 4.9 kg; P < 0.0001) and greater PBF (36.4 ± 6.5% vs. 35.0 ± 6.2%; P = 0.012). However, there was no significant difference in PBF between the two groups after matching for BMI (35.1 ± 6.2% vs. 35.0 ± 5.7%; P = 0.87) or for age and BMI (35.6 ± 5.1% vs. 35.8 ± 5.9%; P = 0.79). Using the criteria of BMI ≥30, 19% of US white women and 5% of Vietnamese women were classified as obese. Approximately 54% of US white women and 53% of Vietnamese women had their PBF >35% (P = 0.80). [Conclusion] Although white women had greater BMI, body weight, and FM than Vietnamese women, their PBF was virtually identical. Further research is required to derive a more appropriate BMI threshold for defining obesity for Asian women.”

Throughout, strong, specific verbs—evaluate, assess, compare, determine, investigate—help clarify the study’s purpose:

“To evaluate interferon effects in experimental…”

“To assess botulinum toxin’s impact on cerebral palsy…”

“To compare image acquisition times for digital…”

“To determine the origin of oligodendrocytes in…”

“To investigate p53’s role in induction…”

Final tips: avoid obscure jargon or unfamiliar abbreviations. Write the abstract last, after the full paper is finished. And be prepared to spend significant time refining those 200–300 words. Like poetry, a good abstract condenses complexity into clarity. Think of it as your paper’s handshake—make it firm, clear, and unforgettable.

That man was me.

Today that same man (now with a lot more wrinkles, a lot less hair, and many publications) is holding the English edition of his memoir Kangaroo Dreams, and I still get goosebumps thinking about it.

This book is the story I never imagined I’d be able to tell in English.

It begins in the green pasture of the Vietnam’s Mekong Delta: barefoot childhood, water buffaloes, rice fields, and bomb craters. Then came April 1975, the fall of the Republic of Vietnam, years of hardship, my family scattered, and finally the night in 1981 when we made the daring decision: escape or slowly disappear.

Twenty-five of us crowded onto a small fishing boat built for half that number. We were robbed by pirates, almost died of thirst, and somehow drifted to Thailand. The refugee camps were harsh, but one day an Australian delegation came through. When they asked where I wanted to go, I said, “Australia… because I’ve never seen a kangaroo.”

They laughed wildly. A few months later, I was on that plane, wearing donated clothes and carrying nothing of my own.

What happened next still feels like a fairy tale I’m not allowed to be the main character of: washing dishes in St Vincent’s Hospital kitchen, studying English at night, scholarships, fellowships, doctorates, and decades later walking back into that same hospital, this time as a professor leading a research team that helps millions of people with osteoporosis.

Fast-forward to 26/1/2022: exactly 40 years after landing in Australia with literally nothing, I received the Order of Australia! The man who just wanted to see a kangaroo got the country’s highest civilian honour instead.

Sometimes the universe really does have a sense of poetry.

Kangaroo Dreams isn’t a misery memoir. It’s a love letter, to the strangers who handed me a a second-hand shirt and trouser, to the country that said yes when so many others said no, and to that stubborn 20s-year-old who refused to let go of a ridiculous dream about a hopping animal.

Forty-five years ago you welcomed a stranger with zero dollars and one impossible hope; today I am proud to call you home. Australia, from the bottom of my heart – thank you for my life.

If you’ve ever started over with nothing…

If you’ve ever been told your dreams are too big…

If you’ve ever felt like an outsider who somehow, against all odds, found a home…

Then I wrote this book for you.

After many rejections from traditional and big-name publishers here in Australia, I decided to bring this story to the world myself—because some dreams, apparently, are too stubborn even for gatekeepers.

The English edition is out on Amazon and Barnes and Noble (paperback, hardcover, and Kindle). I told the story exactly the way I would tell it to you over coffee: no embellishments, plenty of tears, and a fair bit of laughter. Here’s the link:

Amazon: https://www.amazon.com/Kangaroo-Dreams-Tuan-Van-Nguyen/dp/1965142613

Barnes and Noble: https://www.barnesandnoble.com/w/kangaroo-dreams-tuan-van-nguyen/1148735998

If it moves you, please consider leaving a review or passing it on to someone who needs reminding that the most impossible journeys can begin with one shirt and a crazy hope.

From the bottom of my very full heart, thank you for letting me share this with you. The 26-year-old who arrived with nothing finally saw his kangaroo… and he’s still hopping.

With endless gratitude.

Giấc Mơ Kangaroo là lời tự sự về hành trình 45 năm của tôi trên đất khách quê người. Từ những ngày lênh đênh giữa biển khơi, đến những tháng năm tạm dung trong các refugee camps ở Thái Lan, rồi may mắn được đặt chân đến vùng đất Down Under – nước Úc xa xôi. Cuốn sách ghi lại những bước khởi đầu gian nan, từ những ngày làm phụ bếp giữa mùi củ hành cay ra nước mắt, đến những ngày tháng làm lại cuộc đời mới.

Không chỉ là hành trình cá nhân, Giấc Mơ Kangaroo còn là những trang viết dành cho những người bạn đồng hành. Đó là những đồng nghiệp, những nghiên cứu sinh đã để lại dấu ấn trong cuộc đời tôi. Một chương sách được dành riêng để khắc họa những gương mặt và cơ duyên tôi gặp họ.

Tôi còn dành một chương khác kể về những bài học ẩn sau vinh quang trong khoa học. Tôi nghĩ đó là những trải nghiệm quí giá sẽ mang đến cho bạn đọc những góc nhìn mới mẻ và đầy cảm hứng.

Đặc biệt, cuốn sách có một chương dài, nơi tôi kể lại hành trình tình cờ về Việt Nam từ những năm đầu thập niên 2000. Đó là một quãng thời gian 25 năm dài miệt mài cống hiến, rong ruổi khắp dải đất hình chữ S, kết nối với hàng vạn người – từ bạn bè, sinh viên đến những đồng nghiệp. Những trải nghiệm đó không chỉ là kỉ niệm, mà còn là nhịp cầu nối tôi với quê nhà.

Giấc Mơ Kangaroo không chỉ là câu chuyện của tôi, mà còn là của hàng triệu thuyền nhân trên thế giới mà họ chưa có dịp kể lại. Đó là một chứng từ của những người con xa xứ, mang theo khát vọng và lòng biết ơn trên mỗi chặng đường đời. Đây có lẽ là một câu chuyện tiêu biểu của hàng triệu người lần đầu tiên (ở trong nước) được kể lại trên mặt sách.

Sáng 28/9/2025, PhanBook đã tổ chức một buổi ra mắt sách tại Nam Thi House (Sài Gòn). Tôi rất cảm kích được dịp chào đón hơn 120 bạn đọc, bạn bè và đồng nghiệp đến chung vui. Sự hiện diện của các bạn là niềm vinh hạnh lớn lao và nguồn động lực vô giá đối với tôi. Thật ấm áp khi được gặp lại những người bạn cũ, cùng ôn lại những câu chuyện về “đường lưỡi bò” thời xa xưa – những kỉ niệm mà tôi vô tình bỏ quên trong trang sách. Những khoảnh khắc đó, nhờ các bạn, đã trở nên sống động và ý nghĩa hơn bao giờ hết. Tôi ước mình có thể trò chuyện lâu hơn với từng người, nhưng thời gian có hạn, và tôi tin các bạn thấu hiểu điều đó. Dù vậy, mỗi nụ cười, mỗi lời chúc mừng của các bạn đều để lại trong tôi những ấn tượng đẹp.

Xin cám ơn các bạn báo chí đã đưa tin:

https://nguoidothi.net.vn/nguyen-van-tuan-50032.html

https://vietnamnet.vn/hoi-ky-chan-thuc-ve-nha-khoa-hoc-goc-viet-lam-vien-si-3-to-chuc-danh-tieng-2446918.html

https://baotintuc.vn/van-hoa/giac-mo-kangaroo-va-hanh-trinh-cong-hien-cho-khoa-hoc-toan-cau-20250928125907825.htm

Các bạn ở Việt Nam có thể mua Giấc Mơ Kangaroo tại các nhà sách lớn như FAHASA và Minh Khai. Các bạn trong và ngoài Việt Nam cũng có thể mua sách qua:

Tiki:

https://tiki.vn/product-p278686496.html?spid=278686497

NetaBooks:

https://www.netabooks.vn/giac-mo-kangaroo

PhanBook:

https://phanbook.vn/products/giac-mo-kangaroo-nguyen-van-tuan

Có thể liên lạc trực tiếp Hà Thảo (PhanBook) qua địa chỉ facebook:

https://www.facebook.com/HT.Kira

Xin chân thành cám ơn các bạn.

*****

Tựa

Cuộc sống hiếm khi nào đi theo một đường thẳng. Cuộc đời tôi là một hành trình đầy những khúc quanh bất ngờ, những mất mát đau lòng và những khoảnh khắc tuyệt vời. Cuốn hồi ức này ghi lại hành trình 45 năm của tôi, từ một thanh niên di tản đi tìm tương lai trên xứ người, qua hàng loạt cơ duyên có dịp cống hiến cho khoa học toàn cầu, kể cả Việt Nam.

Chuyến hải hành táo bạo và nguy hiểm vẫn còn sống động trong ký ức của tôi. Sự tuyệt vọng và tương lai bấp bênh trôi dạt trên biển. Tuy nhiên, giữa những khó khăn, một tia hy vọng đã lóe lên. Một câu nói đơn giản về con kangaroo − biểu tượng quốc gia của Úc − trong cuộc phỏng vấn đã đưa tôi đến một vùng đất tràn đầy cơ hội.

Nước Úc đã dang rộng vòng tay đón tôi − một người di tản mệt nhoài. Úc đã trở thành nơi cư trú an lành, nơi tôi có thể xây dựng lại cuộc sống và theo đuổi niềm đam mê học hành. Đây không chỉ là câu chuyện về hành trình cá nhân tôi mà còn về những cá nhân phi thường khác đã giúp tôi trên suốt chặng đường đời − những người thầy hướng dẫn tôi, những đồng nghiệp truyền cảm hứng cho tôi và một quốc gia đã coi tôi như một người con.

Cuốn hồi ức được viết và phát hành vào năm 2025, năm đánh dấu tròn nửa thế kỷ thuyền nhân Việt Nam định cư ở Úc và các nước phương Tây. Có thể xem những thông tin và dữ liệu trong hồi ức không chỉ là hành trang của cá nhân tôi mà còn là những chứng từ đóng góp vào kho tàng dữ liệu của thuyền nhân ở hải ngoại. Những thông tin này có thể giúp bạn đọc trong và ngoài nước hiểu tại sao ngày xưa chúng tôi rời quê hương, tại sao ngày nay chúng ta có một cộng đồng người Việt ở hải ngoại đã và đang có những đóng góp quan trọng cho Việt Nam.

Tuy rằng cuốn hồi ức mô tả hành trình của một cá nhân, nhưng tôi nghĩ đó cũng là hành trình chung của hàng triệu đồng hương di tản hay thuyền nhân Việt Nam trên khắp thế giới chưa có điều kiện kể lại câu chuyện của mình. Họ cũng từng trải qua những ngày tháng khó khăn trên xứ người, cũng phấn đấu để tồn tại trong xã hội mới, giúp quê nhà và tìm cách vượt qua nghịch cảnh. Câu chuyện của tôi cũng là câu chuyện của họ. Tôi muốn nghĩ rằng cuốn hồi ức này nói giùm cho các đồng hương đó.

Khi đọc những trang sách này, bạn sẽ không chỉ gặp những thành tựu mà còn cả những khó khăn và thất bại, những năm tháng mất mát vì chiến tranh, nỗi nhớ quê hương đã bỏ lại và sự đeo đuổi không ngừng nghỉ giấc mơ bắt nguồn từ một vùng đất xa lạ. Cuối cùng, đây là câu chuyện về vượt qua nghịch cảnh, tinh thần dấn thân và tác động sâu sắc mà một người có thể đạt được khi được trao cơ hội phát triển.

Nguyễn Văn Tuấn

*****

Hình ảnh buổi ra mắt sách tại Nam Thi House (Sài Gòn)

In addition to grabbing attention, titles serve a functional role in indexing and discovery. They act as concise summaries in academic databases, libraries, and institutional repositories. Because of this, a title must be both engaging and informative, balancing creativity with clarity. This section offers guidance on how to write titles that effectively communicate your study’s value while increasing its visibility and impact.

Scientific titles vary widely, but they generally fall into three categories based on their content: descriptive, declarative, and question-based. Descriptive titles are the most neutral; they simply state the topic of the research without drawing conclusions. For example, Genetics and Personalized Fracture Risk Assessment gives a clear sense of the subject without making claims about findings. Declarative titles, by contrast, state the study’s outcome outright. An example would be PRX Gene Mutation Boosts Mortality Risk, which directly communicates a result. Question-based titles take a different approach, using curiosity to draw the reader in. A title such as Does Estrogen Harm Bone Health? encourages readers to explore the paper for the answer.

While each style has its place, studies have shown that descriptive titles tend to receive more citations than declarative or question-based ones. Question titles may attract more clicks or downloads, but they are cited less often [2]. Declarative titles, while potentially impactful, run the risk of sounding overly confident—something to avoid in scientific writing, where findings are rarely absolute.

Structurally, titles typically take one of two forms: a single sentence or a two-part construction separated by a colon. Single-sentence titles are direct and to the point. For example, Network Analysis Links Alpha-Synuclein to Ovariectomy-Related Bone Loss immediately tells the reader what the study found and how. Two-part titles allow for more nuance, often combining a headline-like summary with details about the methodology. In Better Survival in Sickle Cell Disease: Insights from a Cohort Study, the first half presents the key finding, while the second half highlights the study design.

Writing an effective title involves more than choosing a format; it requires strategic decision-making. Research on title length and citation patterns has revealed some useful principles. First and foremost, a good title is concise. A study in The Lancet found that the most-cited papers had titles averaging 18 words, while less-cited papers averaged only 9. The sweet spot for title length appears to be between 10 and 18 words. Within that range, you should aim to include your study’s main focus, method, or novel contribution. Words like “new” or “innovative” can signal originality. For instance, A New Model for Predicting Diabetes Risk in Thais clearly communicates a novel approach and a specific population.

On the other hand, overly long titles can be overwhelming and unfocused. Consider this example: Effects of Felodipine on Blood Pressure, Heart Rate, Plasma Renin, Angiotensin II, Catecholamines, and Aldosterone in Essential Hypertension. While thorough, it is far too dense. A better option would be: Felodipine’s Impact on Essential Hypertension, which captures the essence without unnecessary detail.

Another key principle is to lead with what matters most. Readers often skim titles, especially in search results or table-of-contents pages. By putting the central message up front, you make it easier for readers to recognize the relevance of your work. If your paper focuses on smoking as a major risk factor, a title like Smoking is Associated with Post-Fracture Mortality is more effective than burying the main point at the end. Similarly, if your emphasis is on genetics, starting with Genetics and Personalized Fracture Risk Assessment puts the spotlight where it belongs.

It’s also essential to highlight your study’s methodological strengths. In medical research, study design often signals the reliability of the findings. If your study is a randomized controlled trial (RCT) or a meta-analysis, say so in the title. Compare three possible titles:

Zinc for Growth

Zinc for Growth in Preterm Infants

Zinc for Growth in Preterm Infants: A Randomized Controlled Trial

The first one is vague. The second title is a bit informative, but not enough. The third title is the most robust, immediately identifying the study population and method. When crafting a title, ask yourself what makes your study new, what the main result is, and what methodological detail adds credibility. For example, a twin study on bone mass might be titled Genetic Determinant of Bone Mass: A Twin Study—succinct, informative, and methodologically transparent.

In addition to clarity and structure, your title must also serve a practical function: discoverability. Online databases like PubMed rely heavily on keywords for indexing. Including relevant terms increases the chances of your paper being found. For instance, if your research involves using ultrasound to assess fracture risk, make sure both “ultrasound” and “fracture” appear in the title. A title like Fracture Risk Assessment: the Role of Ultrasound is both informative and searchable. Alternatively, Ultrasonography as a Novel Tool for Fracture Risk Assessment brings the method into focus.

Above all, a strong title is informative. It should clearly convey the main message of the paper. Compare Bone Loss is Associated with Blood Cell Counts with Bone Loss and Blood Cell Counts. The former suggests a specific relationship, while the latter is vague and unhelpful. Likewise, overly broad titles like Postmenopausal Osteoporosis fail to provide any insight into what the study actually contributes.

Even experienced researchers can struggle with titles, often due to habit or lack of training. Many titles in academic databases are unclear or cluttered, making them difficult for general readers to understand. Common mistakes include using technical jargon, relying on abbreviations, and including irrelevant details such as study location or time period. For example, the title A Study of Statin and Bone Loss in Women Aged 60–90 in District 1, Ho Chi Minh City, Vietnam is overly specific and clunky. A more effective version would be Statins and Bone Health in Older Women or Statin’s Protective Effect on Bone Loss: A Prospective Study. These alternatives are clearer, shorter, and more focused.

In conclusion, writing an effective scientific title requires the same level of care as the research itself. By following a few key principles—keeping it concise, starting with the main idea, emphasizing methodological strengths, using relevant keywords, and making the message clear—you increase the chances that your work will be noticed, read, and cited. Two-part and descriptive titles tend to perform best, while jargon, vague terms, abbreviations, and excess detail should be avoided. A well-crafted title does more than name your work—it opens the door to your research.

___

[1] Kerkut GA. Choosing a title for a paper. Comp Biochem Physiol A Physiol 1983;74(1):1. doi: 10.1016/0300-9629(83)90702-8.

[2] Jamali HR, Nikzad M. Article title type and its relation with the number of downloads and citations. Scientometrics 2011;88:653–661.

I just want to talk about a minor error that has led to a corrigendum on our conference abstract

As the head of a research group, I value openness, especially when reflecting on our own missteps. Today, I would like to share a small but instructive error that led us to issue a corrigendum, and the valuable lesson it reinforced.

To clarify terminology: a corrigendum is issued when authors bear responsibility for an error, while an erratum addresses journal-side mistakes. In our case, the responsibility was entirely ours.

The matter arose with a conference abstract—not a full peer-reviewed paper—submitted by one of my PhD students. With the best of intentions, she listed a colleague who had kindly helped with data analysis as a co-author. Unfortunately, he had not yet seen the abstract and asked for his name to be removed. That request was completely understandable and appropriate.

As principal investigator, I immediately took full responsibility. Because this was only a conference abstract (a brief summary rather than a formal publication), the correction was straightforward and the impact minimal. We contacted the organizers promptly and published a short corrigendum updating the author list.

The core lesson was simple yet profound: even for an abstract, authorship must never be assumed. The ICMJE criteria—widely respected across scientific fields—require not only significant intellectual contribution, but also final approval of the submitted version and willingness to be accountable. Our colleague had not been given the chance to provide that approval, so including his name, however well-meant, fell short of the standard.

This experience was a gentle but necessary wake-up call. Thankfully, because it involved only an abstract and was addressed quickly and transparently, no real harm was done. Still, it humbled us and prompted an immediate change: from now on, no one—on abstracts or full papers—will ever appear as an author without explicitly confirming their agreement in writing.

Small oversights can carry big lessons. I am grateful to the colleague who spoke up, to my student who grew from the experience, and to a system that lets us correct the record gracefully.

I have learned that putting your error into words can strangely feel liberating, and that is precisely the feeling washing over me right now. We move forward with lighter hearts and sharper habits.

The workshops typically feature eight sessions covering key topics: crafting effective titles, writing strong introductions, methods, results, and discussion sections, mastering scientific English, navigating peer reviews, and choosing the right journals. Over time, I have updated this guide after each workshop to better address the needs of medical researchers in Vietnam. The guide is organized into the following ten sections:

1. Use of AI in scientific writing

2. How to write a standout title

3. How to structure a compelling introduction

4. How to detail your methods

5. How to present your results effectively

6. How to craft a thoughtful discussion

7. How to write a concise abstract

8. How to respond to reviewers’ comments

9. Scientific English

10. Selection of journals

In this inaugural post, I would like to focus on a timely topic: the use of AI in scientific writing.

***

'Writing is thinking on paper' – that is my favorite quote from William Zinsser. Indeed, writing is a mental exercise.

Artificial Intelligence (AI) has become an integral part of academic life. From helping students check their grammar to supporting researchers in analyzing large data sets, AI tools are reshaping how we write, think, and publish. However, as powerful and convenient as AI can be, its growing presence in academic writing raises critical questions about authorship, originality, responsibility, and the future of human intellectual development.

Understanding AI’s Place in Writing

AI is often described as a tool that mimics human intelligence to perform tasks like reasoning, learning, or generating content. In the context of scientific writing, AI can play two very different roles: it can assist writers or attempt to replace them.

AI-assisted writing refers to scenarios where the human writer remains fully in control. The individual creates the ideas and writes the content and then uses AI to refine it—perhaps through grammar checks, sentence rephrasing, or stylistic suggestions. In this case, AI acts more like an advanced editing tool. Most academic institutions and publishers—including the Committee on Publication Ethics (COPE), the American Psychological Association (APA), and publishers like Sage—accept this use of AI, provided the writer maintains full authorship and responsibility. Typically, no formal disclosure is needed when AI is used in this limited, supportive role.

AI authorship or AI-generated writing, on the other hand, is fundamentally different. Here, AI takes on the role of the writer, producing large sections—or even entire articles—based on prompts provided by the user. While this may seem like an efficient shortcut, it introduces serious ethical concerns. First, the content may not be original because generative AI pulls from countless sources, often without clear attribution, and may inadvertently copy or misrepresent ideas. Second, AI is known to “hallucinate”—a term used when it fabricates information, references, or data that appear convincing but are entirely false. Finally, since AI tools cannot take responsibility for what they produce, the burden of accuracy and integrity still falls on the human user.

Because of these issues, academic guidelines emphasize the need for transparency. If any significant portion of content is created by AI, it must be clearly disclosed. This includes citing the AI tool used, the date of access, and even the prompt that generated the content. Failure to do so can be viewed as a breach of academic integrity and may lead to rejection by journals or disciplinary action in educational institutions.

Writing as Mental Exercise

One of the most overlooked dangers of overusing AI in writing is what it does to the human mind. Writing is not just a means of communication—it’s a form of mental training. The process of forming an argument, finding the right words, and revising your ideas exercises critical thinking, deepens understanding, and cultivates creativity. These cognitive benefits are not just academic; they shape how we see the world, make decisions, and solve problems.

When we rely too heavily on AI to do the intellectual heavy lifting, we risk weakening these vital skills. If AI handles the brainstorming, writing, editing, and summarizing, what’s left for the human brain to do? Over time, this dependency may lead to a dulling of mental sharpness. People may become less perceptive, less analytical, and ultimately less capable of producing meaningful, original work. In simple terms, over-reliance on AI can make us intellectually lazy—even, some argue, "dumber." Just as physical inactivity leads to weaker muscles, mental inactivity leads to weaker minds.

Ethical Use and Responsibility

Being a responsible user of AI doesn’t mean rejecting it altogether; instead, it means using it wisely, with full awareness of its strengths and limits. If you use AI to clean up grammar or suggest better phrasing, you’re still the author. But the moment AI starts shaping your arguments or generating original content, you must be transparent—and you must stay critically engaged.

Ethical writing also means verifying everything AI produces. Never assume that AI-generated text is accurate, unbiased, or original. You, the writer, are accountable for every word submitted under your name. Before including any AI-generated content in your work, check it for factual errors, plagiarism, bias, and logical consistency.

Some AI functions—such as code corrections, creating tables, or reducing word counts—are less visible in the final writing. Still, best practices suggest acknowledging such uses in your methodology or in an endnote, even if formal citation isn’t required.

What the Major Publishers Say

As AI becomes more common in academic workflows, leading publishers and journals have begun issuing clear policies to guide its ethical use. While these policies vary in emphasis, they share a commitment to human oversight, transparency, and intellectual accountability.

Elsevier permits the use of AI tools for language editing and improvement—as long as this use is disclosed. However, it strictly prohibits the use of AI in editorial decisions or peer review processes. Authors remain fully responsible for the content and must ensure that AI use does not compromise the integrity of the publication.

Science Journals (including the journals published by Science) take a stricter stance. They prohibit any use of AI-generated text or images without explicit editorial permission. Unauthorized use is considered a form of scientific misconduct.

The JAMA Network (Journal of the American Medical Association) strongly discourages the use of AI and requires full disclosure of any AI-generated content. Like Science, JAMA also prohibits the use of AI in peer review.

Springer Nature allows authors to use AI tools for language enhancement, again with disclosure, but forbids AI use in generating scientific images or making core scientific contributions. The focus is on maintaining human intellectual input at the center of the research process.

Taylor & Francis supports the responsible use of AI with the condition of clear human oversight. It prohibits AI in image creation and mandates disclosure of any AI assistance or generated content.

NEJM AI, a publication of the New England Journal of Medicine, aligns with ICMJE (International Committee of Medical Journal Editors) guidelines. It requires full disclosure of any AI use in submissions and prohibits listing AI tools as authors under any circumstance.

In summary, AI may support the writing process, but it cannot replace the human intellect behind scholarly work. Disclosure, transparency, and human accountability remain non-negotiable principles.

The Path Forward

AI is here to stay, and it has a place in academia. It will continue to evolve and become even more embedded in academic workflows. But that doesn't mean we should let it overtake the human role in research, writing, and learning. As students and scholars, we must lead the way in using AI responsibly—leveraging its strengths while maintaining the mental discipline, ethical standards, and intellectual integrity that define true scholarship.

By embracing AI as a supportive tool rather than a creative replacement, and by maintaining transparency in its use, we can uphold the standards of academic integrity while benefiting from technological advancement. Acknowledge it in your manuscript or in an author note. And always ask yourself: Is this helping me think better—or just think less?

I just want to emphasize: AI can be a powerful ally, but it should never become your voice. Your thinking, your creativity, and your ability to express ideas clearly—these are skills worth protecting and practicing. After all, the most valuable work in academia doesn’t come from machines: it comes from minds.

Remember the essence of scholarship: Write with tools but think with your own mind.

Following the Vietnam War's conclusion in April 1975, when Vietnam unified under communist rule, millions fled the country as refugees from 1975 to 1990, braving perilous seas in search of freedom. Small river boats, carrying fragile hopes amidst towering waves and pirates, symbolized resilience and the desire for a better life. This exodus introduced the term “boat people” to the English language.

But no one could have imagined that, more than three decades later, Vietnamese refugees continue to risk everything to escape, facing journeys often more perilous than those of the earlier boat people. Today’s refugees confront not only treacherous seas but also asylum claim rejections or being turned away by third countries, even when recognized as refugees. They rely on the empathy of predecessors who endured similar hardships and understand displacement’s pain. Shira Sebban’s Vietnam’s Modern-Day Boat People, which I discovered at its poignant launch at the Vietnamese Community Activity Center in Sydney, deeply resonates with me. As a boat person resettled in Australia in 1982, I see this book as a vital link between my past and the painful yet hopeful stories of today’s refugees.

Shira Sebban, an Australian of Jewish descent, begins her book with a poignant and inspiring tale. In 2015, 20 Vietnamese from Bình Thuận province, led by three mothers and their 12 children, fled religious persecution and government property seizures on a small, poorly equipped boat, seeking freedom in Australia. Their dreams were crushed when the Australian Navy intercepted them in Australian waters, and a controversial screening process led to their return to Vietnam. The journey’s organizers, including Hồ Trung Lợi, were arrested upon return and sentenced to two years in prison. His wife, Trần Thị Thanh Loan, narrowly avoided a three-year sentence due to a last-minute amnesty, leaving her to care for their four children, aged 4 to 16.

The ordeal continued for these families. In 2017, they attempted another escape, but their small boat capsized off Indonesia after striking a coral reef. Rescued but stranded in Indonesia for 5 years, they endured harsh conditions in an immigration detention center, where women and children were confined in windowless rooms, sleeping on floor mattresses and hanging clothes on ropes. Their children were denied schooling due to their lack of legal status. In 2022, through the persistent advocacy of Shira Sebban, the Vietnamese community in Queensland, Vietnamese-Canadian Senator Ngô Thanh Hải, and Canada’s Private Sponsorship of Refugees Program, the group was granted asylum in Canada, where they received housing, job opportunities, and education for the children.

Shira Sebban learned of the Vietnamese families’ plight through an Australian newspaper. Moved by their story, she contacted lawyer Võ An Đôn in Vietnam to offer support. Sebban began by raising funds to prevent Trần Thị Thanh Loan’s children from being sent to an orphanage, later expanding her efforts through three community fundraising campaigns to assist five families in similar situations. Her book chronicles the refugees’ struggles and highlights the solidarity of activists, including collaborations with UNHCR and VOICE Canada, as well as compassionate acts like purchasing a computer for a child or covering school fees for a girl in Indonesia. Sebban’s visits to the families in 2018 and 2022, and her presence at a supported young man’s graduation, reflect her deep commitment. Her gesture of gifting English-Vietnamese dictionaries to the families upon their gaining permanent residency in Canada poignantly symbolizes companionship and hope for their new future.

Bringing Shira Sebban’s book to readers was a challenging yet inspiring journey. Australian publishers rejected the manuscript, deeming refugee stories 'outdated' and unappealing. However, Sebban’s passion and the story’s profound humanity captivated McFarland Publishers in the US, who embraced its emotional depth. To reach Vietnamese readers, translator Hồ Trọng Hiệp in Australia meticulously translated the book into Vietnamese, published by Quill Hawk Publishing in the US The success was driven by Amy M. Le, Quill Hawk’s CEO and a former boat person from Trà Vinh who fled Vietnam as a child after 1975. Having left a high-tech career, Amy Le dedicated herself to literature as a writer, speaker, and publisher, amplifying Asian-American voices globally. Sebban fondly refers to Amy as her 'wonderful midwife,' who helped her work flourish.

As a former boat person who faced the terrors of the sea, I was profoundly touched by the struggles of these families. Trần Thị Thanh Loan’s anguish as her children were labeled 'traitors' during school flag ceremonies, the harassment at her fruit stall, and the stark image of children confined in a prison-like detention center evoke memories of a grim era. Yet, Sebban’s book powerfully underscores that even the faintest hope can illuminate the darkest paths. Loan’s defiant words, 'If you want to deport us, shoot us all,' embody not just resistance but an unbreakable spirit, a beacon of resilience guiding us through hardship.

Vietnam’s Modern-Day Boat People imparts profound lessons. It demonstrates that compassion can transform lives, as seen in Sebban’s and activists’ efforts—from preventing family separation to securing a new life in Canada—showing that individuals and small communities can achieve miracles. Acts like providing a computer or funding school fees are more than material aid; they ignite hope and faith. The book powerfully advocates for justice, critiquing Australia’s harsh immigration policies that overlook human rights, while urging reforms to protect vulnerable refugees. The contrast between Australia’s strict border controls and Canada’s humanitarian approach prompts reflection on the duties of civilized nations. Ultimately, the book celebrates perseverance and determination. Despite imprisonment, detention, and danger, these families clung to their dream of freedom, proving that hope fuels resilience and courage crafts stories with hopeful endings.

Vietnam’s Modern-Day Boat People is a must-read, not only for activists, refugee scholars, or those engaged with immigration policy, but for anyone who values human resilience amid hardship. For readers, it serves as more than a tale of daring change; it’s a mirror reflecting personal histories and a reminder that the pursuit of freedom requires courage and solidarity. Sebban’s work transcends storytelling, acting as a call to action to amplify silenced voices and advocate for a just future. This book is not merely to be read but to be felt and acted upon, inviting each of us to contribute to the narrative of hope and freedom.

[1] Vietnam’s Modern-Day Boat People: Bridging Borders for Freedom by Shira Sebban, translated by Hồ Trọng Hiệp from the original English edition published by McFarland in 2024 (236 pages). The Vietnamese edition was published by Quill Hawk in 2025.

According to the news article, in 2024, the University of Technology Sydney (UTS) recorded a revenue of $1.3 billion compared to an expenditure of $1.4 billion, resulting in a $100 million deficit. A UTS spokesperson stated, "We cannot sustain these financial losses, so we must reduce expenditure to safeguard our core teaching and research activities." To address this, the university plans to cut 400 staff positions to achieve cost savings. It's all about money.

The uncertainty surrounding job security at UTS has left staff feeling anxious, as they await clarity on whether they will be affected by the redundancies. It’s understandable that a third of UTS employees are experiencing psychological distress, given the significant emotional toll of this situation. This is a serious concern that cannot be overlooked.

The ABC article highlights the case of my coleague, Dr. Hossai Gul, who risks potential repercussions for speaking out against job cuts due to restrictive institutional policies on public commentary. Her situation reflects a broader conflict between academic freedom and institutional priorities across the higher education sector. As a scientist and academic with over 30 years of experience, I have witnessed a profound transformation in the university system. Once celebrated as centers of independent thought and rigorous scholarship, universities are now increasingly shaped by market-driven imperatives, reshaping their mission and the role of academics within them.

Historically, universities were guided by the Humboldtian ideal, a 19th-century vision championed by Wilhelm von Humboldt that positioned higher education as a pursuit of knowledge for its own sake. Universities were sanctuaries of intellectual freedom, where professors and students collaborated to create and disseminate new ideas through high quality research. Admission was highly selective—during my student days in Vietnam, fewer than 5% of school-leavers attended university—a focus on academic excellence. This environment fostered critical thinking and philosophical inquiry, with universities serving as guardians of intellectual culture, where debate and dissent were not just tolerated but encouraged.

These days, the university sector operates within a markedly different framework, often resembling businesses more than scholarly communities. This shift has fundamentally altered the role of professors, moving them from independent scholars to multifaceted service providers navigating economic and administrative pressures. To understand this transformation, it is worth exploring the traditional and modern roles of professors in greater depth.

In the Humboldtian model, professors were primarily scholars, deeply engaged in original research that advanced their disciplines. They were intellectual leaders, shaping their fields through pioneering work and mentoring students to engage critically with complex ideas. A professor of literature, for instance, might guide students through intricate texts, encouraging them to challenge assumptions and develop original interpretations. Similarly, a scientist might train students in experimental design, fostering contributions to cutting-edge discoveries. Teaching was an extension of their research, designed to inspire intellectual curiosity rather than deliver standardized content. Academic freedom was central, allowing professors to pursue controversial or unconventional research without fear of reprisal. As public intellectuals, they contributed to societal debates through writing, lectures, and media, often challenging political or cultural norms.

In contrast, modern university has redefined the professor’s role, prioritizing efficiency and financial metrics over scholarly depth. With a significant increase in university attendance—nearly 40% of young Australians now pursue higher education—the focus has shifted toward accessibility and throughput, often at the expense of intellectual rigor. Professors are increasingly expected to deliver pre-packaged courses designed for mass consumption, functioning as content deliverers rather than intellectual leaders. Administrative burdens have also grown, with professors dedicating significant time to tasks like grant applications, compliance reporting, and course management, leaving less room for research or mentorship.

This shift is compounded by a growing emphasis on financial contributions as a measure of academic success. Increasingly, professors are evaluated not on the quality or rigor of their research but on the amount of funding they secure through grants or the revenue generated through teaching large student cohorts. This focus on financial metrics can sideline important but less immediately profitable research, discouraging innovation and critical inquiry. In one meeting about my own institution’s mission, I argued that universities should lead by producing graduates who advance industry through innovation and critical thinking. However, the majority favored producing graduates who meet existing industry requirements, positioning universities as followers rather than leaders! This experience reflects a broader trend across the sector: universities are aligning with market demands, prioritizing employability over transformative scholarship.

The casualization of academic labor further exacerbates these challenges. Redundancies and a global trend toward replacing tenured positions with precarious, fixed-term contracts have become commonplace. Contingent staff, such as adjuncts or early-career academics, face limited job security and reduced opportunities for independent research. A contract-based lecturer, for instance, may avoid pursuing innovative or risky projects, fearing it could jeopardize their employment. This precariousness also discourages public critique, as restrictive institutional policies often limit staff from commenting on administrative decisions. The result is a chilling effect on academic freedom, undermining the ability of professors to serve as public intellectuals or challenge institutional priorities.

This transformation reflects a broader redefinition of the university sector’s purpose. Once spaces for fostering intellectual independence, universities now often function as credentialing institutions, driven by revenue from student fees and corporate partnerships. The humanities, which cultivate critical thinking, are increasingly deprioritized as 'unprofitable,' while vocational programs aligned with industry needs are expanded. Having spent three decades in academia, I recall a time when a professorship was a respected profession, offering the time and resources to pursue meaningful research and build lasting relationships with students. Today, many academics are stretched thin, balancing teaching, administration, and the pursuit of funding, with little opportunity for the reflective, transformative work that once defined their role.

Addressing these challenges requires thoughtful reform. Increasing public funding could reduce reliance on student fees and corporate influence, allowing universities to prioritize scholarship. Strengthening protections for academic freedom, such as revising restrictive policies on public commentary, would empower professors to engage in open dialogue. Fostering a return to a more selective admissions and rigorous standards could shift the focus back to intellectual excellence, ensuring universities produce graduates who lead rather than follow industry trends. Finally, prioritizing research quality and rigor over financial metrics would encourage professors to pursue innovative and impactful scholarship.

The challenges facing the university sector, as evidenced by widespread redundancies and restrictive policies, are not isolated but symptomatic of a broader shift in higher education. The courage of academics such as Dr. Hossai Gul who speak out, despite the risks, serves as a reminder of the importance of preserving the university’s role as a space for free thought and scholarship. By reflecting on the changing role of professors and advocating for reforms that restore the sector’s original mission, we can work toward a future where higher education values knowledge, critique, and intellectual independence over economic imperatives. The challenge is significant, but the stakes—preserving the soul of the university—are even greater.

The article by Aryee et al. [1] examined the prevalence of obesity using a multidimensional definition that included body mass index (BMI), waist circumference (WC), and percent body fat. The authors defined elevated body fat as ≥25% for men and ≥35% for women, citing the World Health Organization (WHO) report [2] as the source for these thresholds.

However, the cited WHO report does not provide any recommendation for these percent body fat thresholds. The primary anthropometric indicators endorsed by WHO for defining obesity are BMI and, to a lesser extent, WC. We previously examined this issue [3] and found that the commonly cited PBF thresholds originate from a misquotation of the WHO Technical Report. This misattribution has been perpetuated in the obesity research literature for over 25 years, despite the lack of validated PBF thresholds for defining obesity.

While Aryee et al. employed PBF alongside other criteria, referencing the WHO as the source lends undue authority to these specific cut-offs. This may influence the interpretation and comparison of obesity prevalence across different definitions and risks further entrenching an inaccurate citation.

We suggest that the authors consider clarifying the origin of these thresholds in future work, particularly when citing WHO guidance. This will contribute to methodological transparency and help ensure accurate interpretation in obesity research.

Tuan V. Nguyen and Lan T. Ho-Pham

References

[1] Aryee EK et al. Prevalence of obesity with and without confirmation of excess adiposity among US adults. JAMA 2025 Apr 17; [e-pub]. (https://doi.org/10.1001/jama.2025.2704)

[2] World Health Organization (WHO) Physical status: the use and interpretation of anthropometry: report of a WHO Expert Committee. Published 1995. Geneva, Switzerland: WHO Technical Report Series 854; p 378 http://whqlibdoc.who.int/trs/WHO_TRS_854.pdf

[3] Ho-Pham LT, Campbell LV, Nguyen TV. More on body fat cutoff points. Mayo Clin Proc 2011;86:584.

These publications represent the culmination of hard work and collaboration, and I'm thankful to all my co-authors. I especially thank my coauthors Thach Tran, Hoa Nguyen, Tam Do, Lan Ho-Pham, Duy Hoang, Kristel de Dios, Ngoc Huynh, Huy Nguyen, Tommy Nguyen, and Johns Hopkins University colleagues Nickolas Papadopoulos and Bert Vogelstein.

A special highlight was the successful PhD defense and graduation of Dr. Hoa Nguyen – a well-deserved achievement!

I'm also deeply grateful for the opportunity to have conducted 5 workshops in Vietnam, empowering over 500 researchers with skills in research methodology and scientific publication.

I believe that sharing knowledge is crucial for advancing science, and I'm excited to continue this work in 2025.

Wishing everyone a healthy, happy, and fulfilling 2025. May the new year bring new opportunities for growth and discovery in all our endeavors.

The most straightforward method of ranking countries is by the number of gold medals won, as gold is the pinnacle of Olympic achievement. According to this measure, the US and China are tied at the top, each securing 40 golds. Japan, with 20 golds, impressively took third place, while Australia, my home country, landed in fourth with 18 golds, ahead of France's 16.

However, focusing solely on gold medals doesn't give the full picture, as it overlooks the value of silver and bronze medals. Surely these achievements matter too? Indeed, they do. So, a more comprehensive approach is to consider the total medal count. By this metric, the US still leads with 126 medals, followed by China with 91, Great Britain with 65, France with 64, and Australia with 53.

But even the above method isn’t entirely fair. Gold medals should be valued more than silver, and silver more than bronze. To account for this, I propose a weighted system where bronze medals are worth 1 point, silver medals are worth 2 points, and gold medals are worth 3 points. Using this formula:

T = 3Gold + 2Silver + 1*Bronze.

When ranked by T, the US still holds the top spot with 250 weighted medals, followed by China, France, Great Britain, and Australia.

Yet, this ranking might still be skewed in favor of larger countries, which have a greater pool of talent to draw from. A fairer method might be to normalize the weighted medal count by the population of each country:

R = T / population (in millions).

Using this metric, New Zealand emerges as the top performer of the Olympics Games 2024. With a population of just 5.1 million and a weighted total of 47 medals, New Zealand achieves an outstanding 9.22 medals per million people—the highest in the world.

My Australia comes in second place. With a population of 26.7 million and 108 weighted medals, Australia boasts 4.04 medals per million people. We will surely be #1 next time.

An interesting side note: another way to analyze performance is to consider how many athletes it takes for a country to win a single medal. This is similar to the concept of 'Number Needed to Treat' in medicine (i.e., how many patients must be treated to prevent one adverse outcome). The ratio can be calculated as:

Q = number of athletes / number of medals.

According to this measure, China excels, needing only 2 athletes to win 1 medal. South Korea follows closely with a Q of 2.2, and the US ranks third with a Q of 2.5. On the other hand, Spain requires 12.5 athletes to win a single medal!

These numbers are, of course, meant to be taken lightly. Some suggest that wealthier nations win more medals because they can invest more in sports, but this isn’t always true. For instance, North Korea, which is poorer than Vietnam, has 6 medals while Vietnam has none. Conversely, wealthy Singapore has only 1 medal. In fact, the correlation between the number of medals and a country’s per capita income is quite low.

In my view, the number of medals a country wins depends primarily on two factors: luck and skill. Luck plays a role when an athlete performs exceptionally well at just the right moment, while skill is cultivated through training and a supportive environment. If we use the Q factor as a rough measure of skill, then China and South Korea certainly stand out.

In conclusion, based on this analysis, New Zealand can truly be considered the top-performing country in the 2024 Olympics, with Australia not far behind.

I reviewed the list of publications that ScholarGPS included in my profile and noticed that several papers were missing. This exclusion is due to their policy of intentionally omitting papers with more than 20 authors -- that is interesting. After examining their methodology, I found it to be quite robust. It considers the productivity, impact, and quality of the work, resulting in the 'ScholarGPS Ranks.' This approach seems less biased compared to those based solely on citation counts.

Numerous platforms utilize bibliometric data to rank scientists, each having its own strengths and weaknesses. The introduction of ScholarGPS, with its reliable and transparent methodology, is a valuable addition for the scientific community.

___

[1] https://scholargps.com/scholars/60950131216086/tuan-v-nguyen

[2] What is Highly Ranked Scholars™? Well, ScholarGPS explains as follows: “Highly Ranked Scholars™ - Lifetime are eminent authors (active, retired, and deceased) whose Top Percentage Ranks places them in the top 0.05 % of all scholars due to their lifetime scholarly contributions in the following four categories: Overall (All Fields), with respect to their specific Field, with respect to their specific Discipline, and with respect to all Specialties with which they are associated.”

Two catalysts prompted me to create a novel model for evaluating fracture risk: an insightful commentary by Dr. Richard Wasnich and our observation on the relationship between bone mineral density and fracture risk.

Dr Wasnich's commentary

At the time, the diagnosis of osteoporosis was (and still) relied on a measurement of bone mineral density expressed in terms of a T-score. A consensus conference convened by the WHO established the definition that an individual with a T-score below -2.5 is classified as having 'osteoporosis.' This definition appears sensible because, through a series of studies, we know that those individuals are at high risk of fracture.

After that consensus, Dr. Richard Wasnich, a prominent figure in bone research, wrote a dissenting editorial (Wasnich R. Consensus and the T-score fallacy. Clin Rheumatol 1997;16(4):337-9). I was captivated by this commentary where he asserted eloquently:

“What are the issues surrounding the use of T-scores, as recommended by the WHO panel? On the one side, they seemingly offer simplicity, which is sorely needed. However they are not readily translated into interventional guidelines. The opposing viewpoint is that T-scores are a major step backwards into the realm of 'fracture thresholds.'

Fundamental to this debate is the fact that bone density is a risk factor, and not a diagnostic test. So making a 'diagnosis' of osteoporosis based on the presence of a single risk factor, at a single point in time, is already a tenuous concept."

And, the subsequent passage truly inspires and drives my motivation.

“We need is an estimate of absolute fracture rate […] There is no need to obscure this useful information by inventing a new statistic, e.g. the "T-score.”

Absolute risk. That is what we need!

Our observation: most fracture cases are not osteoporotic

I thought the dichotomous classification of T-score was absurd. Consider having a T-score of -2.48, which is not deemed osteoporotic, yet a T-score of -2.51 is categorized as such. In our previous studies we discovered that the relationship  between T-score and fracture risk was continuous, lacking a distinct 'break point' suitable for straightforward dichotomization.

What led to the choice of the -2.5 threshold? On reviewing the literature, I learned that this threshold was selected to align the prevalence of osteoporosis with the lifetime risk of fracture among Caucasian women, which is approximately 30% (Kanis et al. The Diagnosis of Osteoporosis. JBMR 1994;9:1137-1141). However, it's worth noting that this threshold faced criticism from Regina C. Elandt-Johnson and Gayle E. Lester (JBMR 1996;8:1198-1200).

Crucially, we noted that over 50% of women and 70% of men who experienced fractures did not exhibit 'osteoporosis' (meaning that their BMD T-scores were above the -2.5 threshold). In other words, if we treat those with T-score < -2.5, we miss a lot of high risk people.

Dr. Nguyen D. Nguyen, my Ph.D. student at the time, and I were fascinated by the observation. I tasked Nguyen with conducting a sophisticated analysis known as "Bayesian Model Averaging" to identify factors beyond the T-score that were linked to fracture risk. He found that apart from old age and low BMD, the number of falls and the number of prior fractures were very important risk factors for fracture. When he presented the result in a lab meeting, I said to myself: this makes sense!

We then wrote a series of papers to describe our predictive models and how they could be used for individualized fracture risk assessment. I decided to send the papers to Osteoporosis International because the journal was (and still is) a highly clinically oriented venue. In the papers, we made a point that:

“The ultimate aim of developing a prognostic model is to provide clinicians and each individual with their risk estimate to guide clinical decisions. At present, individuals with low bone mineral density (i.e., T-scores being less than -2.5) or with a history of prior low trauma fracture are recommended for therapeutic intervention. This recommendation is logical and appropriate, since these individuals – as shown in this study and previous studies – have higher risk of fracture, and treatment can reduce their risk of fracture. However, because fracture is a multifactorial event, there is more than one way that an individual can attain the risk conferred by either low BMD or a prior fracture. Indeed, virtually all women aged 70 years with BMD T-scores less than -1.5 and all 80-year-old men with BMD T-scores less than -1.0 can be considered ‘high risk’. On the other hand, no 60 year old men or women without a prior fracture and a fall are considered high risk, even when their BMD T-scores are below -2.5. This demonstrates the informativeness of a multivariable prognostic model, and the limitation of a risk stratification-based approach for risk assessment for an individual.”

We also made another point re the uniqueness of fracture risk:

“Each individual is important and unique. […] Prognosis is about imparting information of fracture risk to an individual and each individual is a unique case, because there exists no ‘average individual’ in the population. The more risk factors are considered, the greater likelihood of uniqueness of an individual’s profile being defined. Therefore, by modeling risk factors in their continuous scale the present models can be uniquely tailored to an individual.”

Actually, my idea of ‘individualization‘ was not new; I learned it from colleagues in the cancer research field. At the time, cancer researchers were busily developing nomograms for predicting the risk of having cancer, and it appeared that these nomograms worked well for many cases. Why do these probabilistic tools work well? Now, we know that highly experienced clinicians can also make good prognoses, and that is a fact. Unfortunately, their predictions are highly variable and less consistent, or in scientific language, clinicians’ predictions are irreproducible. But reproducibility is a bedrock of science. So, scientifically, we cannot rely on a clinician’s judgment. Statistical prognostic models have been shown to outperform clinical judgment because these models can objectively incorporate many risk data. Moreover, any prognosis from a statistical model is unbiased, consistent, and completely reproducible.

One year after the publication of our papers, the FRAX model — developed under the sponsorship of the World Health Organization — was published. So, doctors and patients in the world now have at least two tools to assess their own risk of fracture in their convenience. The two models, Garvan and FRAX, have helped transform the management of osteoporosis worldwide.

BONEcheck

After more than 10 years of experience with FRC and recent advances in research, I have identified several features that could improve the utility and relevance of the tool:

• Prediction timeframe: FRC and FRAX primarily offer a 10-year forecast for fracture risk. I believe that managing a 10-year risk is more challenging for elderly individuals than a 5-year risk. Therefore, a shift in the prediction timeframe is warranted.

• Risk presentation: all current fracture risk assessment tools present numerical probabilities, which can pose a challenge for the general population to comprehend. There is a need for a more user-friendly presentation.

• Treatment context: existing risk estimates produced by risk assessment models don't provide the benefit in terms of fracture reduction and increased survival (and potential risk) if a high-risk patient opts for treatment; limiting the communication of risk and clinically useful discussions between patients and their physicians.

• Refracture: an existing fracture significantly elevates the risk of subsequent fractures, yet existing fracture risk prediction models do not estimate the risk of refracture.  

• Mortality: most fractures, especially hip fractures, are linked to an increased risk of mortality. However, existing tools do not incorporate mortality into their predictions.

To address the aforementioned issues, our team at the University of Technology Sydney (UTS), supported by an NHMRC grant, has overhauled the original Garvan Fracture Risk Calculator, introducing a novel and advanced version named BONEcheck^TM. This updated version of BONEcheck incorporates features absent in existing tools, including:

• Five-year frame prediction.

• Treatment contextualization: BONEcheck incorporates data from randomized controlled trials (RCTs) to inform patients about the specific reduction in fracture risk associated with medication use, tailored to their age and risk profile.

• Risk of refracture: BONEcheck features a dedicated module for predicting the probability of refracture.

• Mortality: BONEcheck draws from previous research to provide users with the risk of mortality following a fracture.

• Skeletal Age: BONEcheck introduces the concept of Skeletal Age, representing an individual's skeleton age due to a fracture or exposure to risk factors that heighten fracture risk.

• Remeasurement of bone mineral density (BMD): BONEcheck incorporates a module predicting the time required to reach osteoporosis in individuals not currently classified as osteoporotic. This feature assists clinicians in advising patients on the timing for repeat BMD measurements.

• Predicting osteoporosis: for individuals without BMD data, BONEcheck includes a module to predict the risk of osteoporosis (T-score less than -2.5).  

• Preventive information: BONEcheck is designed with a patient-centric approach, aiming to empower individuals with information that enables them to actively reduce their risk of fractures.

Several validation studies have conclusively shown that the rebranded BONEcheck, formerly the FRC, demonstrates fracture risk prediction accuracy equal to or surpassing that of FRAX. The predicted probability of fracture derived from BONEcheck/FRC also aligns closely with clinical decisions. Notably, FRC/BONEcheck comes recommended by the Royal Australian College of General Practitioners, Healthy Bone Australia, Osteoporosis New Zealand, and the Asia Pacific Consortium for Osteoporosis for application in clinical practice. Our FRC tool is a key component of “Know Your Bones” that helps people self-assess their bone health. (You can click on the above link to have your test now!)

Lastly, I am pleased to announce that BONEcheck is entirely free of charge. Since its launch in May 2023, BONEcheck has garnered usage from over 15,000 users across 165 countries worldwide.

The development and implementation of personalised fracture risk assessment has been considered a revolution in the management of osteoporosis (Saag and Geusens, Arthritis Res & Ther 2009). I am delighted to have played a pioneering role in this transformative journey.

_____

Post Script: BONEcheck is now accessible to users through multiple platforms. Users can access it directly from our website or download the app from the Apple Store or Google Play. Please click on the links below to start utilizing the BONEcheck tool:

Website: https://bonecheck.org

Apple Store: https://apps.apple.com/app/bonecheck/id6447424513.

Google Play: https://play.google.com/store/apps/details?id=org.saigonmec.bonecheck.

The development of BONEcheck and its features can be found in the following article: https://www.sciencedirect.com/science/article/pii/S2405525523000481?via%3Dihub

_____

There have been hundreds of articles on our fracture risk assessment model in the literature, including ours. Here are some of the recent reviews that I have written for journals and books:

• Nguyen TV. Personalized fracture risk assessment: where are we at? Expert Review in Endocrinology and Metabolism 2021;16(4):191-200.

• Nguyen TV (2020). Toward the era of precision fracture risk assessment. J Clin Endocrinol Metab. pii: dgaa222.

• Nguyen TV, Eisman JA. Post-GWAS Polygenic Risk Score: Utility and Challenges. JBMR Plus 2020;4: e10411.

• Nguyen TV. Personalised assessment of fracture risk: which tool? Aust J Gen Pract 2022;51:189-190.

• Nguyen TV. Individualized Fracture Risk Assessment: State-of-the-Art and Room for Improvement. Osteoporosis and Sarcopenia 2018;4(1):2-10.

• Nguyen TV. Individualized Assessment of Fracture Risk: Contribution of “Osteogenomic Profile”. J Clin Densitom 2017;20:353-359.

• Nguyen TV, Eisman JA. Assessment of fracture risk: population association vs individual prediction. J Bone Miner Res 2017 Dec 27.

• Nguyen TV, Eisman JA. Genetic profiling and individualized assessment of fracture risk. Nature Review Endocrinolology 2013 Mar;9(3):153-61.

• Nguyen TV, Center JR, Eisman JA. Individualized fracture risk assessment: progresses and challenges. Curr Opin Rheumatol. 2013 Jul;25(4):532-41.

• Nguyen TV, Eisman JA. Genetics and the individualized prediction of fracture. Curr Osteoporos Rep 2012 Sep;10(3):236-44.

• Nguyen TV. Mapping translational research into individualized prognosis of fracture risk. International Journal of Rheumatic Diseases 2008; 11:347-358.

• Tran B, Center JR, Nguyen TV. Translational genetics of osteoporosis: from population association to individualized risk assessment. In Primer on the Metabolic Bone Diseases and Disorders of Mineral Metabolism, Seventh Edition, Ed: Clifford Rosen. ASBMR 2017 Edition.

• Nguyen TV, Eisman JA. Pharmacogenetics and pharmacogenomics of osteoporosis: personalized medicine outlook. In Genetics of Bone Biology and Skeletal Disease, Edited by RJ Thakker, MP Whyte, JA Eisman, T Igarashi. Academic Press Amsterdam 2017 Edition.

• Nguyen TV. Individualized Progress of Fractures in Men. In Osteoporosis in Men – the effect of gender on skeletal health, 2nd Ed, Edited by ES Orwoll, JP Biezikian, and D Vanderschueren. Academic Press, 2011.

BONEcheck stands as an innovative iteration, often referred to as Mark II, of the original Garvan Fracture Risk Calculator (FRC) introduced in 2008. Positioned as an alternative to FRAX, another tool for fracture risk assessment, BONEcheck distinguishes itself with a range of features that cater to the needs of both medical professionals and patients alike.

• Five-year frame prediction. Current fracture risk assessment tools typically project a 10-year risk, which may not be practical for individuals in their 70s or 80s. Managing a 5-year risk is more feasible and user-friendly.  

• Treatment contextualization. Existing antiresorptive and bone-forming therapies effectively reduce fracture risk in patients with fractures or osteoporosis. However, this treatment benefit is often overlooked in current fracture risk assessment tools. BONEcheck incorporates data from randomized controlled trials (RCTs) to inform patients about the specific reduction in fracture risk associated with medication use, tailored to their age and risk profile.

• Risk of refracture. Our research has revealed that individuals with a prior fracture face an elevated risk of experiencing another fracture, a factor not accounted for in existing tools. BONEcheck features a dedicated module for predicting the probability of refracture. 

• Mortality. Fractures, particularly hip fractures, correlate with a significant increase in mortality risk, information often lacking in current tools. BONEcheck draws from previous research to provide users with the risk of mortality following a fracture.

• Skeletal Age. BONEcheck introduces the concept of Skeletal Age, representing an individual's skeleton age due to a fracture or exposure to risk factors that heighten fracture risk. This metric aids the public in understanding the impact of fractures on mortality.

• Remeasurement of bone mineral density (BMD). BONEcheck incorporates a module predicting the time required to reach osteoporosis in individuals not currently classified as osteoporotic. This feature assists clinicians in advising patients on the timing for repeat BMD measurements.

• Predicting osteoporosis. For individuals without BMD data, BONEcheck includes a module to predict the risk of osteoporosis (T-score less than -2.5). This information serves as a valuable screening tool for identifying individuals at high risk of osteoporosis, warranting further BMD assessment.

• Preventive information. BONEcheck is designed with a patient-centric approach, aiming to empower individuals with information that enables them to actively reduce their risk of fractures.

The development of fracture risk assessment tools has been marred by controversies involving conflicts of interest [2] and a dearth of transparency [3]. In contrast, the development of BONEcheck / FRC prioritizes transparency, as the predictive equations are openly accessible [4-5]. The detailed procedure and methodology for BONEcheck have been published [1, 4-5], and I encourage you to explore them for a comprehensive understanding.

Numerous validation studies have demonstrated that the FRC, now known as BONEcheck, predicts fracture risk as effectively as or more accurately than FRAX [6-11]. The predicted probability of fracture derived from BONEcheck/FRC also exhibits a high level of concordance with clinical decisions [12-13]. FRC / BONEcheck is recommended by the Royal Australian College of General Practitioners, Healthy Bone Australia, Osteoporosis New Zealand, Asia Pacific Consortium for Osteoporosis for use in clinical practice. Finally, I am pleased to say that BONEcheck is entirely free of charge.

To sum up, my team and I have created and launched a digital tool named 'BONEcheck' for global fracture risk assessment, and it is now accessible to users at no cost. This tool serves to facilitate discussions between doctors and patients regarding fracture risk, clinical implications, and treatment benefits, enabling informed decision-making.

PS: BONEcheck is now accessible to users through multiple platforms. Users can access it directly from our website or download the app from the Apple Store or Google Play. Please click on the links below to start utilizing the BONEcheck tool:

Website: https://bonecheck.org

Apple Store: https://apps.apple.com/app/bonecheck/id6447424513.

Google Play: https://play.google.com/store/apps/details?id=org.saigonmec.bonecheck.

References

[1] Nguyen ND, Frost SA, Center JR, Eisman JA, Nguyen TV. Development of prognostic nomograms for individualizing 5-year and 10-year fracture risks. Osteoporos Int 2008 Oct;19(10):1431-44.

[2] Järvinen TL, et al. Conflicts at the heart of the FRAX tool. CMAJ 2014 Feb 18;186(3):165-7.

[3] Collins GS, Michaëlsson K. Fracture risk assessment: state of the art, methodologically unsound, or poorly reported? Curr Osteoporos Rep 2012 Sep;10(3):199-207.

[4] Nguyen ND, Frost SA, Center JR, Eisman JA, Nguyen TV. Development of a nomogram for individualizing hip fracture risk in men and women. Osteoporosis International. 2007;18:1109–1117. doi: 10.1007/s00198-007-0362-8. 

[5] Nguyen DT, Ho-Le TP, Pham L, Ho-Van VP, Hoang TD, Tran TS, Frost S, Nguyen TV. BONEcheck: A digital tool for personalized bone health assessment. Osteoporos Sarcopenia. 2023 Sep;9(3):79-87.

[6] Holloway-Kew KL, Zhang Y, Betson AG, Anderson KB, Hans D, Hyde NK, Nicholson GC, Pocock NA, Kotowicz MA, Pasco JA. How well do the FRAX (Australia) and Garvan calculators predict incident fractures? Data from the Geelong Osteoporosis Study. Osteoporos Int. 2019 Oct;30(10):2129-2139.

[7] Langsetmo L, Nguyen TV, Nguyen ND, Kovacs CS, Prior JC, Center JR, et al. Independent external validation of nomograms for predicting risk of low-trauma fracture and hip fracture. CMAJ. 2011;183(2):E107–E14.

[8] Pluskiewicz W, Adamczyk P, Franek E, Leszczynski P, Sewerynek E, Wichrowska H, et al. Ten-year probability of osteoporotic fracture in 2012 Polish women assessed by FRAX and nomogram by Nguyen et al.-Conformity between methods and their clinical utility. Bone. 2010;46(6):1661–67.

[9] Sandhu SK, Nguyen ND, Center JR, Pocock NA, Eisman JA, Nguyen TV. Prognosis of fracture: evaluation of predictive accuracy of the FRAX algorithm and Garvan nomogram. Osteoporos Int. 2010;21(5):863–71.

[10] van Geel TA, Nguyen ND, Geusens PP, Center JR, Nguyen TV, Dinant GJ, et al. Development of a simple prognostic nomogram for individualising 5-year and 10-year absolute risks of fracture: a population-based prospective study among postmenopausal women. Ann Rheum Dis. 2011;70(1):97–7.

[11] Bolland MJ, Siu AT, Mason BH, et al. Evaluation of the FRAX and Garvan fracture risk calculators in older women. J Bone Miner Res 2011;26(2):420–27. doi: 10.1002/jbmr.215.

[12] Inderjeeth CA, Raymond WD. Case finding with GARVAN fracture risk calculator in primary prevention of fragility fractures in older people. Arch Gerontol Geriatr 2020;86:103940. doi: 10.1016/j.archger.2019.103940.

[13] Pluskiewicz W, Adamczyk P, Franek E, et al. FRAX calculator and Garvan nomogram in male osteoporotic population. Aging Male 2014;17(3):174–82.