BMI has been the default measure of body size in research for decades, mostly because it is cheap and easy to collect at scale. But a growing body of large cohort work suggests it is a fairly blunt instrument, and that where fat sits on the body matters more than the number on the scale.
At the same time, being very lean is also associated with poorer health outcomes, although the reasons are more complicated than they first appear. This piece works through what the evidence currently shows on both ends of that spectrum, how these measures hold up when used as mediators in causal models rather than just outcomes, and what that means for something as ordinary as deciding whether to reach for a tape measure or a set of scales.
BMI's blind spot
BMI cannot tell the difference between muscle and fat, and it says nothing about where fat is distributed. Waist-to-height ratio (WHtR) and waist circumference are increasingly used as more direct proxies for central adiposity, the visceral fat that sits around the organs and is more metabolically active than fat stored elsewhere.
The most rigorous recent test of this came from a cohort study of nearly half a million UK Biobank participants, with the findings replicated in the independent Whitehall II cohort. WHtR-defined obesity was generally the strongest predictor of the 78 health conditions examined, ahead of waist circumference and BMI, though the gap was not large across the board. Critically, the study found that swapping BMI for waist-based measures did not meaningfully improve risk prediction once someone already had a BMI of 30 or above; the real added value of WHtR showed up in the overweight range, roughly BMI 25 to 29.9, where BMI on its own is at its weakest as a signal of risk (Ho et al., 2025). That is a more accurate way to frame the argument than simply saying "WHtR beats BMI," because it outperforms BMI primarily where BMI is already weakest.
A separate systematic review of cardiovascular outcomes reached a similar conclusion, reporting that prospective cohort studies consistently showed WHtR outperforming BMI in predicting cardiovascular disease, and noting the appeal of a single, ethnicity- and sex-independent cutoff of 0.5, essentially keeping waist circumference under half of height (Tewari et al., 2023). WHtR has since been incorporated into the European Association for the Study of Obesity's revised criteria for central adiposity partly for this reason.
A note on using it as a mediator
WHtR and BMI are not just used as outcomes or exposures in this literature, they are also routinely used as mediators, the variable that sits in the middle of a causal chain and is thought to explain how an exposure produces an outcome. In nutrition epidemiology, a typical version of this design looks like: diet exposure leads to a change in adiposity, and that change in adiposity is what actually drives the downstream disease risk (Fairchild & McDaniel, 2017). A recent example used exactly this structure, testing whether adiposity, defined either as BMI or WHtR, mediated the relationship between ultra-processed food intake and a panel of inflammatory biomarkers. It did, explaining somewhere between roughly 13 and 70 percent of the total effect depending on the biomarker and which adiposity measure was used (Millar et al., 2025).
This is a reasonable design, but it inherits the same limitation already discussed above, just in a different position in the model. WHtR is a proxy for how much central fat someone is carrying, not a proxy for diet quality itself. Two people can arrive at a similar waist-to-height ratio through very different routes, one through a diet high in ultra-processed food, another through alcohol intake, disrupted sleep, a sedentary job, or simple genetic predisposition to central fat storage. When WHtR sits as the mediator, the model is really testing "does diet affect adiposity, and does adiposity affect the outcome," not "does diet affect the outcome through a specific, identifiable biological pathway." It is a proxy for the mechanism rather than the mechanism itself, and it will absorb variance from all the other things that shift waist size regardless of diet.
If the aim is to actually trace the biological route from diet to disease, a mediator that sits closer to the pathway does a better job. This is where circulating metabolomic and inflammatory markers come in, panels of amino acids, lipids, and lipoprotein subfractions from NMR metabolomics, inflammatory markers such as CRP, IL-6, or GlycA, gut-derived metabolites such as short-chain fatty acids or bile acids, or composite functional indices like HOMA-IR that reflect insulin resistance directly rather than inferring it from body size. A recent Mendelian randomisation mediation study took this approach, using gut microbiota-derived metabolites as mediators of obesity risk rather than a non-invasive body measurement, which is a mechanistically closer design than routing everything through BMI or WHtR (Li et al., 2025). The limitation is largely practical rather than methodological: these markers require blood samples and laboratory analysis rather than a tape measure, which is almost certainly why non-invasive body measurements remain the default in very large cohorts even though they are the blunter instrument.
A more recent NHANES-based study offers a useful complement to this critique rather than simply another instance of it. Yuan et al. (2023) tested whether adiposity mediated the well-established link between adherence to the DASH diet and hypertension risk in 8,224 US adults, and found that WHtR was a substantially stronger mediator than BMI, accounting for roughly 36 percent of the total indirect effect compared with about 9 percent for BMI, together explaining 45 percent of the overall relationship. Rather than stopping at the body measurement mediator, the authors then used random forest modelling across 60 measured nutrients to identify which specific dietary components were driving that pathway, narrowing down to sodium, potassium, and an omega-3 fatty acid (octadecatrienoic acid) as the nutrients most consistently linked to both DASH adherence and WHtR. This is a more satisfying design than using WHtR as the endpoint of the analysis, because it uses the body measurement as a first-pass filter and then decomposes it into the specific nutrients responsible, pairing the coarse mediator with a nutrient-level pathway analysis rather than treating one as a substitute for the other. It is also a useful demonstration that WHtR, while a blunter instrument than a metabolomic panel, is not too blunt to be worth using as a starting point, provided the analysis does not stop there.
The other end of the curve
Here is where it gets more complicated. Across large cohort studies, the relationship between BMI and mortality is not a straight line, it is typically J-shaped or U-shaped, with elevated risk showing up at both low and high BMI. A UK cohort study of 3.6 million adults found this same J-shaped pattern for most causes of death, with the lowest risk sitting around 21 to 25 kg/m² (Bhaskaran et al., 2018).
The question is how much of the risk at the low end is real, and how much is reverse causation, meaning people have low body weight because they are already unwell, rather than unwell because of their low body weight. This is a well recognised problem in the field. Undiagnosed cancer, inflammatory bowel disease, chronic infection, and other conditions can all cause unintentional weight loss years before diagnosis, which then shows up in cohort data as "low BMI predicts death," when the more accurate story is "illness predicts both low BMI and death."
One of the more elegant attempts to strip this confounding out used a Norwegian cohort and a genetic instrumental variable approach, using the BMI of a person's adult children as a proxy for the parent's own BMI, since offspring BMI is not affected by the parent's current illness. When the analysis was done this way, the excess mortality risk at low BMI largely disappeared, while the risk at high BMI remained and even strengthened. The authors concluded that elevated mortality at high BMI looks genuinely causal, while the elevated mortality at low BMI is best explained by confounding from concurrent ill health rather than leanness itself being dangerous (Carslake et al., 2018). This distinction matters because low body weight is not inherently harmful. Instead, the observed association often reflects the illness that caused the weight loss rather than the weight loss itself.
This same pattern turns up in the dementia literature too. A UK Biobank study of over 155,000 older adults found that a lower weight-adjusted waist index, essentially less central adiposity relative to body weight, was associated with a higher risk of incident dementia over 13 years of follow-up, while higher values were protective (Suo et al., 2025). The authors proposed a biological explanation involving adipose-derived signalling molecules, but a study like this cannot rule out that some of the leanness in the higher-risk group reflects early, pre-diagnostic frailty or weight loss related to the neurodegenerative process itself rather than a truly protective effect of carrying more central fat. It is a good example of why single cohort studies, however large, need to be read carefully rather than taken as the final word.
Why the tape measure holds up as a practical measurement
There is also a practical case for WHtR that sits alongside the statistical one, and it has to do with how easy the measurement is to take reliably outside a clinical setting.
Waist measurement is a genuinely feasible thing to do at home. A validated community study found that self-measured waist and hip circumference agreed closely with measurements taken by a trained assistant, with a concordance of 0.96 for waist alone, and correlations with BMI, blood pressure, and blood glucose that were similar whether the measurement was self- or assistant-taken (Reidpath et al., 2013). It requires nothing more than a tape measure and a flat wall to stand against, no calibrated scale, no clinic visit, and the accuracy holds up well enough for the measurement to be useful for tracking change over time rather than only in a research setting.
The takeaway
Waist-to-height ratio does appear to add real predictive value over BMI, but mainly in the overweight range rather than uniformly across the board, and it is not a magic replacement metric. Being underweight is genuinely associated with worse outcomes in cohort data, but a meaningful part of that signal looks like it reflects illness causing low body weight rather than low body weight causing illness. Used as a mediator, WHtR is a coarser instrument than a metabolomic panel, but the DASH diet example shows it is not too coarse to be a useful starting point, provided the analysis does not stop there.
Taken together, the evidence suggests that waist-to-height ratio is one of the simplest and most informative health measurements most people can take at home. It is inexpensive, reproducible, and often more revealing than body weight alone, particularly for people in the overweight range. BMI still has an important role in clinical practice and research, but between medical appointments, a tape measure may prove to be the more useful tool. Perhaps it's time to stop asking, "What do I weigh?" and start asking, "What does my waist measure?"