"For people who are less active, the key message is that small amounts can still matter," Edward Giovannucci told Everyday Health in June. "Building a routine gradually may be more important than trying to do a lot at once."
That is the co-author of the study talking. And it is the opposite of the number everyone extracted from his work.
The number is 90 to 120 minutes a week. It has propagated across health coverage as if it were a dosing chart—two hours, no more, filed under the same mental folder as "8 glasses of water" and "10,000 steps."
But two words are missing from nearly every version of the headline: "associated with." And behind those two words sits a study design that cannot deliver what the coverage promises.
Two words are missing from nearly every version of the headline: 'associated with.' Behind them sits a study design that cannot deliver what the coverage promises.
Last week this newsletter covered how resistance training appears to erase muscle-aging signatures at a modest dose—tissue-level evidence, mechanistically clean. This week the claim jumps a category: from what happens inside muscle to whether lifting weights makes you die later. That jump is where the reasoning gets slippery.
Start with what the Harvard study actually is. Published June 2 in the British Journal of Sports Medicine, it pooled three decades of data from nearly 150,000 adults across the Health Professionals Follow-Up Study, the Nurses' Health Study, and the Nurses' Health Study II.
Participants reported their weekly resistance training and aerobic exercise via periodic questionnaires. That is the machinery producing the number.
The findings are genuinely striking. Adults averaging 90 to 120 minutes a week of strength training had a 13% lower risk of death from any cause, a 19% lower risk from cardiovascular disease, and a 27% lower risk from neurological diseases like Alzheimer's, compared with those who did none.
Combine that with aerobic activity and the association climbs to a 45% lower mortality risk—among the lowest in the entire cohort. Evidence level here: E10, an observational cohort. Robust for association. Silent on causation.
That distinction is not pedantry. It is the whole ballgame.
An observational cohort tracks people who chose to lift weights and compares them to people who chose not to. The people who lift differ, on average, in a hundred unmeasured ways—more health-conscious, more able-bodied to begin with, better resourced, more likely to also sleep and eat well.
Questionnaire-based cohorts adjust for what they can measure. They cannot adjust for what they cannot. Reverse causation lurks too: people already sick, already frail, already sliding toward a neurological diagnosis tend to train less. Their deaths then pile up in the "no training" group and inflate the apparent benefit of training.
None of this makes the signal fake. The direction is almost certainly real. But the magnitude and the precision are exactly what this design is worst at pinning down.
Now the part being oversold hardest: the plateau. The Harvard study reported that training beyond 120 minutes a week "did not appear to provide any additional benefit."
That single clause has been rewritten into a ceiling—a sense that more lifting is pointless or even wasteful. One outlet framed it as the study challenging "the notion that more is always better," with benefits that "plateau after 120 minutes."
Be careful here. A flattening curve in the upper tail of a self-reported observational distribution is one of the least trustworthy things in all of epidemiology.
The people reporting the highest training volumes are the fewest. The confidence intervals are widest exactly where the plateau appears. Self-report error grows with volume—remembering "about two hours" is easier than accurately logging four.
A flat line in that region is compatible with a real plateau, with measurement noise swamping a small further benefit, or with the sickest high-volume trainers dragging the tail down. The data cannot distinguish these. Treating the plateau as a demonstrated biological ceiling reads precision into a place the method has none.
So separate the two piles cleanly.
What is well-supported by randomized evidence is that resistance training builds strength, size, and physical function at modest doses. The ACSM's 2026 Position Stand synthesized 137 systematic reviews covering more than 30,000 participants from randomized trials.
Its verdict: compared with no exercise, RT reliably improved muscle strength, hypertrophy, power, endurance, gait speed, and balance. Strength responds best to heavier loads through a full range of motion, two to three sets, at least twice a week. This is E8—strong, randomized, and consistent.
The companion network meta-analysis in the same journal, 178 studies for strength and 119 for hypertrophy, found every prescription beat no exercise, with the authors suggesting that "healthy adults can adopt a resistance training prescription of their choice and preference." A separate meta-analysis of 157 articles even located a genuine dose-response plateau—but for strength, at 773,000 to 887,000 arbitrary units of dosage, not for death.
Read that ACSM synthesis carefully and you notice what it does not claim. Its measured endpoints are muscular and functional. Strength. Size. Walking speed. Getting out of a chair.
It notes in passing that RT is associated with reduced mortality and lower cardiovascular, cancer, and diabetes risk—but those associations ride on the same observational literature, not on the randomized trials it actually pooled. The randomized evidence proves RT makes you stronger and more functional. It does not prove any dose makes you live longer.
That is the gap. And it is not a small one.
Nobody has run a randomized controlled trial with strength training as the intervention and all-cause mortality as the endpoint, powered to detect a difference. Such a trial would require thousands of participants, years of follow-up, and adherence discipline that is brutally hard to maintain. It is expensive and unglamorous. So it hasn't happened.
Which means the entire "90-120 minutes lowers your mortality" claim rests on association, not experiment. The strongest thing we can say with randomized backing is that RT builds the strength and function that plausibly underlie a longer, more independent life. The final causal link is inferred, not shown.
This does not mean the guidelines shouldn't change. The cardio-centric framing of most public health advice genuinely does under-weight resistance training, and reportedly a majority of adults do none at all. A signal this consistent, pointing this direction, across 150,000 people and three decades, deserves to move the needle.
But notice what the honest reading actually recommends. The gulf that matters is between zero and something—between the majority doing no strength work and a person doing even 30 or 60 minutes a week. That is where every line of evidence, randomized and observational, agrees the payoff is largest.
The "90-120 minutes" figure is a floor-versus-zero story wearing the costume of an optimum. It tells you that some beats none by a wide margin. It does not tell you that 105 minutes is a precisely calibrated dose or that minute 121 is wasted.
Giovannucci said as much himself. Small amounts still matter. Build gradually. That is the language of a floor, not a ceiling.
The precision prescription is a mirage assembled by headline writers out of a real but blurry association. The direction is trustworthy. The two-hour number is not a target. And the study that would turn any of this into causal fact—a mortality-endpoint RCT—remains unrun.
Until it exists, lift because the randomized evidence says it makes you stronger, more mobile, and more independent as you age. Those are the outcomes we can actually prove. The longer life is a reasonable bet layered on top—not a number you can dose to the minute.