People born later in the 20th century reach old age with better memory and lose it more slowly than earlier generations, and the combined gain is large enough to plausibly account for the fall in dementia rates, a study in Nature Medicine reports.

Researchers led by Anders Fjell and Kristine Walhovd at the University of Oslo analysed about 783,000 memory test results from about 219,000 people in five long-running studies in Europe and the United States, with follow-up of up to 30 years. The paper was published on 17 November 2026.

What the researchers measured

Dementia incidence at a given age has been falling in Europe and the United States for decades. A 2020 meta-analysis of seven cohorts in Neurology put the drop at 13 per cent per decade between 1988 and 2015. It was unclear whether that reflects better memory on entering old age, slower decline, or both.

The Oslo team used word-list recall, a standard test of episodic memory. According to an earlier open-access version of the paper, the five studies were the English Longitudinal Study of Ageing, the US Health and Retirement Study, the Survey of Health, Ageing and Retirement in Europe, the Irish Longitudinal Study on Ageing and the Longitudinal Aging Study Amsterdam. Scores were put on a common 0 to 20 scale and corrected for practice effects, the boost people get from repeating a test.

Participants were grouped into five-year birth cohorts from 1910–14 to 1960–64. The preprint says the main analysis covered 88,550 people who had at least four memory tests, over a median follow-up of 12.5 years.

A double gain

The authors call the result a “double dividend”. For each decade of later birth, people scored on average 0.54 more words (95 per cent confidence interval 0.49 to 0.59) and their yearly decline was 0.070 words shallower (0.063 to 0.077). The preprint describes these first estimates as an upper bound, because they do not separate generation effects from calendar-year trends.

The gains were not spread evenly. People whose memory fell fastest benefited most: the generational effect among the fastest decliners was about twice the size of the effect for the typical person. Later-born groups also spent longer in stable stretches and were less likely to enter periods of rapid loss.

The team checked whether the memory slopes mean anything biologically. In a separate sample, people classed as severe decliners lost hippocampal volume at 2.00 per cent a year, against 0.63 per cent a year for those with flat memory. The hippocampus is a brain region central to memory.

Linking memory to dementia

The published abstract says the combined advantage is large enough to plausibly account for the reported 13 per cent per-decade fall in dementia incidence. It also says later-born generations show lower rates of clinical impairment, and that calendar-time trends affecting all generations at once are unlikely to explain the differences.

The authors say memory level is built mainly in early life, while the rate of decline is shaped by conditions in adulthood. They conclude that keeping dementia rates falling will need prevention that protects brain health across the whole lifespan, beginning in childhood. The preprint estimates that early-life conditions account for about half of the total generational advantage up to age 80, and lists better management of vascular risk factors, less smoking and lower pollution as plausible adult influences. The study did not test any of those causes.

What the study cannot show

The study is observational. It cannot show why later generations have better memory or slower decline, and the possible causes named in the preprint are suggestions, not findings.

Separating birth-cohort effects from age and calendar-year effects cannot be done from the data alone. The researchers used a bounding analysis that rests on assumptions, such as that memory does not improve with age and that later-born cohorts are not disadvantaged. The preprint says the slope advantage is positive under most plausible assumptions, but a near-zero effect cannot be fully excluded under the most conservative ones.

Restricting the main analysis to people with at least four tests also leaves a gap. The preprint says people whose memory declined fastest may have been less likely to complete four waves, and that reweighting the sample cannot fully correct for this.

The 13 per cent figure was not measured in these cohorts. It comes from separate studies, and the paper tests whether the memory gains are big enough to account for it. A word-recall score is also not a dementia diagnosis. The participants come from Europe and the United States, so the study says nothing about other regions.

The figures above come from the abstract of the published paper and the full text of the earlier open-access version. The published full text sits behind a paywall and was not read, so some numbers may differ in the final version.