The paper trail of Alzheimer's drug development is enormous and almost nobody reads it whole. Pentara Corporation, a biostatistics firm in Salt Lake City, tried. Its database searches turned up 6,044 records; after stripping duplicates, secondary reports and trials that did not meet criteria, 217 placebo or sham controlled randomised studies remained, covering more than 130 treatments in people with mild cognitive impairment due to Alzheimer's or mild to moderate Alzheimer's disease, spanning 1994 to 2025. That work, by Samuel Dickson, Suzanne Hendrix and colleagues, is preliminary: it was a podium presentation at the Alzheimer's Association International Conference, published as an abstract in the journal Alzheimer's & Dementia on 25 December 2025, not yet as a full peer-reviewed paper with a trial-by-trial table.
The method matters more than it sounds. Most trials are judged on one yardstick at a time: a cognitive scale, or a function scale, or a global rating. The authors instead required that at least two of the three domains be reported, then combined cognition, function and global measures into a single global statistical test. Their argument is that a real drug effect should show up in all three, while a fluke rarely does. Their stated finding is blunt in both directions: some studies previously determined to be failures show, in their reanalysis, some indication of positive effects, while some previously thought promising look more clearly like failures. They also write that success and failure has been "partly obscured" by reporting only one of the three domains. Applied earlier, they suggest, a composite approach might have spared the field much of the fight over the 2021 aducanumab approval.
One piece of vocabulary needs pinning down, because it is easy to misread. The "time savings" in the title is not an audit of research years burned on bad designs. It is a measure of patient benefit: the horizontal gap between a treated curve and a placebo curve, expressed in months of progression delayed rather than points on a scale. The method was set out in a peer-reviewed paper by the same group in the Journal of Prevention of Alzheimer's Disease in April 2024. An earlier conference abstract from the group, in December 2023, applied it to five treatments (aducanumab, donanemab, lecanemab, AD04 and Souvenaid) and reported time savings ranging from two to seven months per year of treatment. Those are estimates derived from published trial results, not fresh trial findings, and they are not a promise about any individual.
For the harder arithmetic of design, a separate and fully peer-reviewed study is the better witness. In JAMA Network Open on 29 August 2025, Etienne Aumont, Pedro Rosa-Neto and colleagues catalogued 203 phase 2 and phase 3 Alzheimer's trials published between 1992 and 2024, involving 79,589 participants. Mean phase 3 duration rose from 20 weeks to 71 weeks, and phase 2 from 16 to 46 weeks. Mean phase 2 enrolment grew from 42 people to 237. Biomarker confirmation as an entry requirement went from 1 study in 36 in the 1992 to 2006 period to 60 of 167 after 2007. Trials, in short, became longer, larger and more selective, and were increasingly built to detect smaller differences. That is the context in which a thirty-one-year recount is being pitched: if the outcome measure is noisy, a long expensive trial can end in a verdict nobody trusts.
What does this mean at a kitchen table? If a relative sat through a two-year trial in 2006 and was told the drug did not work, that verdict was made with the tools of the day, and a reanalysis may soften or harden it. It does not resurrect the drug, change what a clinic can prescribe next week, or mean anyone was given something that secretly worked. It also arrives with an obvious interest to declare: Pentara develops and sells the time-component and composite methods it is arguing for, and its analyses are frequently commissioned by trial sponsors. The abstract names no drug as rehabilitated and publishes no list of reversed verdicts, so nobody can yet check which failures moved and by how much.