Does Living Longer Mean Ageing More Slowly?

Substantive correction, 20 September 2026: We withdraw claims that reduced lethal lesions chiefly explain male acarbose longevity, that NDGA establishes equivalence between lifespan extension and slower ageing, and that nonsignificant comparisons demonstrate no benefit or restoration to youth. We correct the necropsy design and measurement conventions, add effect intervals, direct sex comparisons and missing-data sensitivity, and rerun survival from original files. Published late-life functional studies of 17α-estradiol exist; the absence of a file in this directory was incorrectly presented as an evidence gap. Both editions, six figures and the reproducibility package are updated; former approvals remain historical.

01Four questions need different measurements

Does an animal live longer? Does it perform better at a particular age? Does it have fewer lesions? Does the rate at which changes accumulate with age become slower? These outcomes can be related without being interchangeable.

A minimal mechanistic model explains why: the overall mortality hazard is the sum of cause-specific hazards. Reducing one cause can improve survival without changing the ageing trajectory of other tissues. Conversely, improving an impairment that seldom causes death may have little effect on lifespan. This relationship explains how outcomes can separate; it does not identify the actual pathway of a particular drug.

Selection adds another complication. Late-life tests observe animals still alive and measured at that time. If treatment changes who survives to testing, it also changes the composition of the measured groups. Comparing survivors does not automatically identify functional effects in all originally enrolled animals, or the proportion of a survival benefit mediated by better function.

02What we actually obtained

We checked eight available phenotype/pathology workbooks and obtained the C2010, C2011 and C2013 lifespan files. The former cover rotarod, grip, body composition, fat pads, glucose, uterus weight and four selected lesions. The latter retain deaths separately from removals; survival estimates no longer depend on the previous article's old output. The official link to the NDGA rotarod workbook still returned 404 during this review. We can cite its published findings, but cannot claim to have independently reanalysed that dataset. Official data, pages and dictionaries, lifespan originals

Recalculation cannot resolve these discrepancies. Phenotype and lifespan files should not be assumed to contain a perfectly matched set of animals, doses, starting ages and sites. Their public identifiers do not support reliable individual linkage. We estimate group differences within each dataset separately. Harrison 2019, Strong 2016

03Survival: recomputed without global rankings

We estimated the relevant treatment-control comparisons from the three original files, separately by cohort and sex. The primary analysis begins at the recorded treatment-start month, explicitly converting a month to 30.4375 days. Sensitivity analyses use 30-day months and all recorded animals; deaths and removals before entry are counted separately. These are survival comparisons conditional on remaining observed at entry, not a claim to reconstruct every aspect of initial random allocation.

For acarbose, the site-stratified Cox hazard ratio is approximately 0.55 in males, with a 95% interval of 0.45–0.67; in females it is 0.73, with an interval of 0.59–0.89. Male acarbose and medium-dose NDGA show statistical evidence of age-varying hazard ratios. Their summaries should not be interpreted as a constant percentage slowing at every age.

We also calculate restricted mean survival time (RMST) from entry to a common attained age of 1,100 days, weighting the three sites equally. That horizon lies within observed support for every comparison arm; it was selected for support rather than for the largest treatment effect.

These are differences in survival-curve area over a specified window, not increases of that size in unrestricted mean lifespan. The male 5,000-ppm NDGA arm is added for a same-dose comparison between sexes; it is not the male grip regimen. Acarbose's female-minus-male RMST treatment contrast is about −75 days, with an interval of −125 to −25. Its Cox hazard-ratio interaction is less clear, P≈0.053. These estimands differ; choosing only the more significant one would not establish a general sex-specific mechanism. Global compound rankings have been removed.

Chow concentrations are nominal. Harrison also reports lower measured acarbose content than intended in some batches. Equal nominal ppm across experiments therefore need not imply equal actual exposure. Harrison 2019

Earlier cohorts also tested survival effects of these compounds. Different doses, starting ages and cohorts cannot automatically be collapsed into one universal drug effect. Harrison 2014

04Acarbose: clearer female rotarod benefits, uncertainty in males

From individual mouse values, the female day-6 treatment difference is about +54.7 seconds, with a 95% interval of 26.1–83.4. The male difference is +7.8 seconds, with an interval of −11.1 to 26.7. The female data support improvement; the male result remains uncertain rather than proving no benefit. Directly comparing treatment effects gives a sex contrast of 46.9 seconds, with an interval of 13.1–80.8 and Holm P≈0.014 across the three rotarod sex comparisons.

Six-day summaries must not silently mix observation counts. Primary means and slopes require all six days, excluding the one female control missing day 2 from those summaries but retaining her in the day-6 comparison. The female mean difference is about 28.7 seconds, with an interval of 7.7–49.7; the learning-slope difference is 10.3 seconds/day, with an interval of 5.9–14.7. Including her five observed days in sensitivity analyses gives approximately 26.6 seconds and 9.4 seconds/day, preserving the direction. A learning slope over six test days is not a rate of ageing across life.

Treated females were also lighter. Conditional on weight at testing, the day-6 difference is +34.0 seconds, with an approximate HC3 interval of 0.5–67.4; the six-day mean interval crosses zero. But treatment has already affected weight. It may be a mediator, a parallel outcome or associated with selection into testing, rather than an ordinary baseline confounder. The adjusted estimate is not automatically a pure anti-ageing effect independent of weight, and the change after adjustment is not an identified mediated fraction.

Acarbose grip results point in mixed directions. The male forelimb-force difference is about −14.2 gram-force, with a pointwise interval of −25.7 to −2.6; all-limb force differs by +15.6 gram-force, with an interval of −1.0 to 32.2. Across the eight comparisons of four readouts in two sexes, no Holm P is below 0.05. The original categorical benefit/harm language was too strong. These measurements need further testing and should not be averaged into a functional age.

05Body composition is not a complete health assessment

Among mice measured at 22 months, fat-percentage differences are about −18.6 percentage points in females and −10.3 in males, with approximate 95% intervals of −24.0 to −13.2 and −15.0 to −5.7. Male lean mass is also lower by 2.50 g, with an interval of −4.38 to −0.61. The female difference is −0.47 g, with an interval of −1.69 to 0.76; this does not establish preservation of lean mass.

At that time, female control/treatment counts are 28/28 and male counts 16/30. Trajectory plots use observations available at each age; paired changes separately require values at both 8 and 22 months. Survival and missingness may change the groups being observed. These curves do not recover trajectories for every animal, and lower fat alone is not proof of better health. Twelve-month fat pads provide small-sample mechanistic evidence from another site; their weight-conditional analyses do not identify direct causal shares either.

06Pathology shows lesion signals, without attributing lifespan gains

The public workbook contains four selected lesions, not the original study's complete pathology panel. Male lung tumours occur in 15/41 controls and 9/54 treated mice: a prevalence difference of approximately −19.9 points, with a pointwise 95% interval of −37.0 to −2.2. Male liver degeneration occurs in 12/41 versus 4/54: −21.9 points, with an interval of −37.7 to −6.4. Female glomerulosclerosis has different denominators, 28 and 57; overall counts must not be substituted.

After Holm correction across eight sex-specific Fisher comparisons, only male liver degeneration remains below 0.05, at about 0.049. This corrects comparisons within the four already selected lesions, not the full original screen. Other nonsignificant lesions listed in the paper are absent from this workbook and were not independently reanalysed.

More fundamentally, these are scheduled-age necropsies at approximately 22–25 months, pooling three sites and historical controls. They are neither causes of death in the individual survival cohort nor linked mediator data. Fewer lung tumours could relate to improved survival, but these data cannot show how many tumours would be lethal or quantify their contribution. We therefore withdraw the claim that reduced lethal lesions chiefly explain male lifespan extension.

07Glucose: total area differs from the postmeal increment

The refeeding experiment at 5.5 months has ten mice per sex and chow group, eighty in total. High-dose males have approximately 22% lower six-hour total glucose area than controls, but fasting glucose starts 21.3 mg/dL lower. For area relative to each mouse's starting concentration, the difference is about −3,806 mg·min/dL, with a 95% interval of −8,157 to 546. The female high-dose incremental difference is approximately 23, with an interval of −3,952 to 3,997. Neither nonsignificance establishes absence of a female drug effect.

Subtracting the starting level defines another descriptive quantity; it does not remove baseline changes from the drug's pharmacology as an error. Total and incremental area answer different questions. A young-mouse experiment after six weeks of exposure does not identify mediation of a multiyear survival effect. The original chain from “more complete male pharmacology” to greater male longevity exceeded the data.

08NDGA and 17α-estradiol: retain benefits and restore missing evidence

The NDGA paper uses each mouse's mean over five grip trials. On that scale, the male duration difference is approximately +71 ms, with a 95% interval of 24–118 and Holm P≈0.027 across four sex-specific force/duration comparisons. Taking only the best trial gives approximately +93 ms, a different estimand now treated as sensitivity analysis. There are only nine treated males and twenty-one controls. A mean similar to young references is not equivalence or restoration of biological age. Grip duration is time until release during pulling on a grid, rather than general hanging endurance.

The female duration difference is −7.7 ms, with an interval of −32.7 to 17.3, and remains uncertain. The paper also reports rotarod improvement in both sexes. Although the individual workbook was unavailable, that published functional evidence is not erased by a missing directory file. The conflicting start-age labels further limit matching function to survival. Strong 2016

The directory's 17α-estradiol file is a uterine-response assay in young ovariectomized mice: four same-age OVX controls, seven at each dose and ten intact references tested at 6.25 months. Increased uterus weight in the higher group demonstrates an estrogen-like tissue response in this experiment. Different ages and surgical states prevent treating the intact group as a restoration target, and this is not a late-life benefit.

Published late-life functional evidence does exist. Garratt and colleagues reported improvements in selected muscle and grip/rotarod outcomes in older male UM-HET3 mice in separate cohorts, with tests of gonadal status and a later treatment start. Design, ages, sample sizes and weight relationships impose their own limits, and these are not paired functional observations in our lifespan animals. Nevertheless, the study directly contradicts the broad claim that no functional research is public. Garratt 2019

09What would strengthen a claim of slower ageing?

Evidence should span organs, function and pathology, with enough ages and time points, young treated controls, attention to survival and attrition, and designs connecting intervention, function and disease or death. Even improvement across several measures needs to be distinguished from an earlier shift in level or an age-independent pharmacological effect.

The rapamycin debate illustrates this. Wilkinson observed attenuation of several nonlethal age-related changes while also recording adverse phenotypes. In Neff's broad study of male C57BL/6J mice, some improvements also occurred in young animals and could not all be attributed to slower ageing. The strains, tests and coverage differ. Our ITP phenotype table cannot settle the disagreement. Wilkinson 2012, Neff 2013

Scheduled-age canagliflozin pathology provides examples of improvement in several lesions that seldom cause death. A separate end-of-life study of later-start canagliflozin and 16α-hydroxyestradiol did not find a clear pathological pattern explaining opposite lifespan directions between sexes. The latter estrogen compound must not be confused with 17α-estradiol. Scheduled-age pathology, end-of-life pathology

These interventions can have real survival and specific functional effects, whose magnitude, coverage and uncertainty differ. The available data cannot give each treatment a validated label of “only suppresses lesions” or “broadly slows ageing.” All calculations concern mice; they do not establish human healthspan benefits or support a treatment regimen.

This is a correction with results already known, not a new preregistration. The reproducibility package contains pinned input records, code, aggregate outputs, validation and six figures; original workbooks are fetched from official sources. Successful replay does not remove missingness, conflicting metadata, absent individual linkage or uncertainty associated with unavailable cage identifiers.