Metabolic score for insulin resistance (METS-IR) predicts all-cause and cardiovascular mortality in the general populationevidence from NHANES 2001–2018

There is a number that predicts whether you will die of heart disease, and it requires no specialized blood draw, no fasting insulin test, and no endocrinologist. Just four values most people already have from a routine checkup. The number is called METS-IR, or the metabolic score for insulin resistance, and almost nobody has heard of it. That's what makes this study worth your attention. Here's the problem it's solving. Insulin resistance, the state where your cells stop responding properly to insulin, sits at the center of two trends that are moving in the wrong direction simultaneously. Obesity in the United States is projected to hit fifty percent of the population by 2030. And as obesity rises, so does insulin resistance, along with the risk of cardiovascular disease. Clinically, insulin resistance often precedes type 2 diabetes by years and quietly accelerates atherosclerosis, dysglycemia, and blood pressure abnormalities long before any diagnosis is made. So why not just measure insulin resistance directly? The gold standard, a procedure called the hyperinsulinemic-euglycemic clamp, holds blood glucose steady while infusing insulin and measures how much glucose the body uses. However, this method is expensive, procedurally complex, and completely impractical for large population studies. The next best thing, the homeostatic model assessment of insulin resistance, or HOMA-IR, requires a fasting blood draw to measure serum insulin. That sounds simple, but fasting insulin is often hard to obtain at scale, varies by race, and fails in patients on insulin therapy or with impaired beta-cell function. So researchers have developed surrogate indexes that skip the insulin measurement entirely. Four of them are the focus of this study: the TyG index, which uses triglycerides and fasting glucose; METS-IR, which adds body mass index and HDL cholesterol to that mix; the triglyceride to HDL cholesterol ratio, or TG/HDL-C; and HOMA-IR itself for comparison. The question Mingxuan Duan and colleagues set out to answer is which of these, if any, actually predicts who dies. The study drew from nine cycles of the National Health and Nutrition Examination Survey, or NHANES, running from 2001 to 2018. After applying strict inclusion and exclusion criteria, ruling out participants under eighteen or over eighty-five, those who were pregnant, or those missing key lab values, the final cohort numbered fourteen thousand six hundred fifty-three adults. Mortality status was tracked through December thirty-first, 2019, using the NHANES linked mortality file matched to the National Death Index. That gave the team one million seven hundred forty-three thousand six hundred seven person-months of follow-up, with a median of one hundred sixteen months, just under ten years. Over that period, two thousand eighty-five participants died of any cause, and five hundred forty-nine died specifically from cardiovascular disease. One of the study's methodological choices is worth pausing on. Instead of selecting adjustment variables the traditional way, by clinical habit or prior literature, the team used the Boruta algorithm, a machine-learning feature-selection method built on random forests. Here's what it does: it takes every candidate variable, creates shuffled decoy versions of each one, and then runs hundreds of random forest iterations, comparing how often the real variable outperforms its own decoy. After five hundred iterations, only the variables that consistently beat their shadows get flagged as truly important. For all-cause mortality, Boruta ranked age, preexisting cardiovascular disease, serum creatinine, systolic blood pressure, blood urea nitrogen, and hypertension at the top. Those Boruta-selected variables then went into Cox proportional hazards models as covariates. To detect non-linear relationships, the team added restricted cubic splines, a statistical tool that lets the data curve freely rather than forcing a straight-line assumption. Now the results. Among the four insulin-resistance indexes, only METS-IR was significantly associated with both all-cause and cardiovascular mortality after full adjustment. Treated as a continuous variable in the fully adjusted model, each one-unit increase in METS-IR corresponded to a one point five percent higher hazard for all-cause death and a one point eight percent higher hazard for cardiovascular death. The other indexes fell short in telling ways. The TyG index predicted all-cause mortality but not cardiovascular mortality. HOMA-IR predicted all-cause mortality but similarly failed on the cardiovascular outcome. TG/HDL-C was not significant for either. METS-IR was the only index that cleared both bars simultaneously. Then the spline analysis revealed something that changes how you interpret that finding entirely. The relationship between METS-IR and mortality is not a straight line going upward. It's a U-shape. There is an inflection point at a METS-IR value of forty-one point thirty-three. Below that threshold, higher METS-IR is actually associated with lower mortality. Each additional unit reduces the hazard for all-cause death by approximately two point eight percent. Above forty-one point thirty-three, the direction reverses sharply. Every additional unit above that threshold raises the adjusted hazard for all-cause mortality by one point nine percent and for cardiovascular mortality by two point eight percent. Let that sit for a moment, because it's counterintuitive. METS-IR is an index of metabolic dysfunction. You might expect that lower is always better. But the data say otherwise. People at the very low end of the METS-IR range face elevated mortality risk too. The shape is consistent with the idea that extremely low values might reflect frailty, malnutrition, or other conditions that raise mortality risk through entirely different pathways than metabolic syndrome. The sweet spot is somewhere in the middle, and straying too far in either direction is dangerous. Age reshapes this picture further. Stratified analyses showed that the METS-IR signal is essentially concentrated in people under sixty-five. In the eleven thousand three hundred forty-two participants younger than sixty-five, METS-IR was significantly associated with all-cause mortality, with a hazard ratio of one point zero fourteen and a p-value of zero point zero zero eight. In the three thousand four hundred eighty-two participants aged sixty-five and older, the association disappeared, with a hazard ratio of zero point nine ninety-four, confidence interval crossing one point zero, and a p-value of zero point two four seven. The interaction between age group and METS-IR was statistically significant for both all-cause and cardiovascular mortality. Duan and colleagues note that exposure to insulin resistance of the same severity may produce more severe downstream complications in younger patients than in older adults. This pattern is consistent with prior literature they cite. Whatever the mechanism, the practical implication is clear: METS-IR carries its strongest predictive signal precisely in the population that still has time to act on it. So what is METS-IR, exactly? The formula, described verbally: take the natural logarithm of two times fasting blood glucose plus triglycerides, multiply that by body mass index, and divide by the natural logarithm of HDL cholesterol. No insulin assay is required. Every value in that equation comes from a standard lipid panel and a BMI measurement. That accessibility is the point. In clinical settings or population surveys where fasting insulin simply isn't available, which is most of the world, METS-IR gives you a workable signal without the logistical burden. The study has real limitations worth naming. It is observational and retrospective, so causation cannot be established. The NHANES data includes self-reported information on lifestyle and medical history, which introduces measurement error. Insulin resistance was measured only at baseline, a single snapshot, so the analysis can't capture how metabolic status changed over the decade of follow-up. And the cohort is drawn from the United States, which limits how directly these findings generalize to other populations with different dietary patterns, genetic backgrounds, and healthcare contexts. What the study does establish is this: in nearly fourteen thousand seven hundred Americans followed for close to ten years, METS-IR predicted all-cause and cardiovascular mortality better than three competing indexes, including the widely cited TyG. The relationship is non-linear, with a meaningful threshold at forty-one point thirty-three. The effect is strongest in adults under sixty-five. And the score requires nothing more than a routine blood panel. For a metric that's barely known outside specialist literature, that's a result worth taking seriously. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.