In a 20-year follow-up of a randomized trial, older adults who completed booster processing-speed training had a lower hazard of Medicare-recorded dementia diagnoses. The finding keeps a trainable cognitive skill in the prevention conversation.
Processing speed—the ability to identify and respond to information quickly—can improve with practice in older adults. The more consequential question is whether training that skill changes dementia risk. A 20-year analysis of the ACTIVE trial offers an encouraging but statistically unsettled answer: a lower hazard of diagnosed Alzheimer disease and related dementias appeared among speed-trained participants who completed booster sessions, but not among those who received no booster training.
A Longevity.Technology report on the long-term findings highlights the potentially actionable idea behind the research. Processing speed is modifiable, unlike many established dementia risk factors. Yet demonstrating that a cognitive skill can be trained is different from demonstrating that its training prevents a clinical disorder.
The signal was concentrated among booster completers
ACTIVE randomized 2,802 older adults to training in memory, reasoning or processing speed, or to a control group. The speed intervention used repeated, adaptive exercises rather than medication or a general lifestyle programme. Some participants also received opportunities for booster training after the initial course.
In the 20-year Medicare-linked analysis, speed-trained participants who completed at least one booster session had a hazard ratio of 0.75 for a recorded Alzheimer disease or related dementia diagnosis compared with controls. The 95% confidence interval was 0.59 to 0.95. A hazard ratio of 0.75 represents a 25% lower relative hazard over the follow-up period; it does not mean that every participant’s absolute probability of dementia fell by 25%.
There was no corresponding signal among speed-trained participants without booster exposure. Their hazard ratio was 1.01, with a 95% confidence interval of 0.81 to 1.27—an estimate closely aligned with no difference from control.
That contrast makes continued participation central to the result. It may mean that a larger or maintained training dose is needed. But the reported group was also defined by completing a booster, which is a participant behaviour rather than simply the original randomized assignment. People able and willing to keep attending may differ from non-completers in ways that also affect long-term diagnosis risk. The trial’s randomized foundation therefore does not remove all uncertainty from this particular comparison.
Statistical handling adds another layer. A subsequent reanalysis applying a Bonferroni–Dunn correction found that all corrected confidence intervals included a hazard ratio of 1. This adjustment raises the threshold for declaring a result significant when several comparisons are being tested. It does not establish that training has no effect, but it means a specific protective effect was not robust to that correction.
What processing-speed training involved
The intervention was more specific than the broad category of “brain games” might suggest. Participants practised visual searches in which they identified and located information while the available inspection time became progressively shorter. Task difficulty adjusted as performance improved.
The programme comprised ten initial sessions, with as many as eight subsequent booster sessions. Among participants who completed all ten initial and eight booster sessions, targeted processing-speed performance improved by approximately 2.5 standard deviations compared with randomized participants who attended no sessions, according to an analysis of the training dose and cognitive response.
That is a substantial difference on the skill the intervention was designed to train. It establishes that later-life processing speed is responsive to structured practice. It does not, by itself, show that the improvement changes the biological processes that cause dementia or that any commercially available cognitive-training product would produce the same result. The ACTIVE exercises, schedule and adaptive progression form part of the intervention being evaluated.
How the dementia result changed over time
ACTIVE’s dementia findings have evolved as follow-up lengthened and investigators used different ways to identify cases. The original five-year analysis did not find a reduction in incident dementia. At ten years, however, investigators identified 260 cases and reported a 29% lower hazard in the speed-training group than in controls. The hazard ratio was 0.71, with a 95% confidence interval of 0.50 to 0.998, meaning the interval only narrowly excluded no difference in the ten-year analysis.
The ten-year outcome was based on research diagnostic criteria rather than a clinician’s diagnosis. The 20-year analysis instead relied on diagnoses recorded in Medicare claims. Claims extend follow-up without requiring every participant to return for an in-person assessment, but they reflect healthcare encounters and coding practices. The two approaches are therefore related measures of dementia occurrence, not interchangeable ones.
Investigators also reported that each additional speed-training session was associated with a 10% lower unadjusted dementia hazard at ten years. This apparent dose-response pattern is intriguing, but the session-by-session association was not itself randomized or adjusted. Greater attendance may be part of the explanation, or it may identify participants with other favourable characteristics.
The ten- and 20-year results should also not be treated as independent replications: they arise from the same original trial population. Their value lies in showing what happened to the signal over different periods and with different dementia definitions.
A trainable skill, not yet a prevention prescription
Two conclusions can reasonably coexist. First, structured processing-speed training can produce large improvements on the cognitive ability it targets. That is the trial’s clearest causal finding and gives older adults and practitioners a realistic basis for discussing what this form of training can achieve.
Second, dementia prevention remains a hypothesis rather than an established benefit. The lower hazard appeared in booster completers, the estimates changed with follow-up and outcome definition, and the 20-year association did not remain statistically specific after correction for multiple comparisons.
For people considering cognitive training, the evidence supports expecting improvement in the practised processing-speed skill—particularly with sustained participation—rather than assuming protection from dementia. For practitioners and developers, it underlines the need to distinguish gains on a trained task from a clinically diagnosed outcome.
Long-term analyses that preserve randomized comparisons, specify how multiple testing will be handled and use consistent dementia ascertainment could determine whether the signal reflects genuine prevention. For now, ACTIVE places processing speed among the more interesting modifiable research targets in cognitive health, without establishing its training as a dementia-prevention method.