ACTIVE Trial: Speed Training and Dementia at 20 Years

In 1998, a team of investigators across six U.S. sites began enrolling healthy adults aged 65 and older into a randomized trial of cognitive training. The participants were told they would spend a few weeks in a classroom or at a computer. Nobody involved could have known that a short course of computer exercises would still be generating headlines in 2026, when a research group linked the training to Medicare claims covering two full decades. The result (DOI 10.1002/trc2.70197), published in Alzheimer’s & Dementia: Translational Research & Clinical Interventions, is among the longest follow-ups of any behavioral intervention aimed at dementia. It is also easy to misread.

Presented By Our Partners

This article walks through what the ACTIVE trial tested, what the 20-year analysis found, why one of the three training types appears to matter while the other two do not, and how much weight a careful reader should put on a result that rests on insurance claims rather than clinical examinations. The short version: the signal is real enough to take seriously and fragile enough to avoid overselling.

What the ACTIVE trial is

ACTIVE stands for Advanced Cognitive Training for Independent and Vital Elderly. It enrolled 2,802 community-dwelling adults aged 65 and older in 1998 and 1999. To qualify, participants needed a Mini-Mental State Examination score of 23 or higher and independence in all basic activities of daily living. In other words, these were older adults living on their own and functioning well, not patients with a memory complaint.

The design was four-arm, multisite and single-blind. Participants were randomized to one of three training programs or to a no-contact control group. The three programs targeted different cognitive abilities:

• Memory training taught explicit strategies, such as mnemonic devices, for remembering lists, text and other verbal material.
• Reasoning training taught strategies for finding patterns in series of letters, words or numbers, which is the kind of thinking used in everyday problem solving such as reading a medication schedule.
• Speed of processing training used an adaptive computer program that trained visual attention and divided attention. Participants had to identify and locate objects on a screen at progressively shorter display times, while tasks became harder as performance improved.

Each arm consisted of up to 10 sessions lasting 60 to 75 minutes, delivered over roughly five to six weeks. A random half of those assigned to training were also offered up to four booster sessions, delivered at about 11 months and 35 months after the initial course, according to the Johns Hopkins summary of the new paper. (Another outlet describes the timing as one and three years after enrollment, which is close but not identical, and the primary article is the authority on the exact schedule.)

The first major report, led by Karlene Ball and colleagues in JAMA in 2002 (PubMed 12425704), showed that each type of training improved the specific ability it targeted. That was the expected result. The harder question, which took two more decades to answer, was whether any of it would matter for the thing people actually fear: dementia.

What earlier follow-ups found

Before the 20-year analysis, ACTIVE had already produced a trail of findings. Training benefits for performance of everyday tasks persisted for up to five years in earlier work, and all three training arms were associated with better self-reported everyday function at 10 years. The speed training group also had fewer at-fault motor vehicle collisions in earlier analyses, an outcome that connects plausibly with faster visual processing.

At the 10-year mark, a dementia analysis reported that speed training was associated with a 29 percent lower incidence of dementia compared with control, and that each booster session was associated with further risk reduction. That finding generated interest and skepticism in equal measure. Ten years is a short window for dementia, because most people who will eventually develop the disease have not yet done so by then. A 20-year analysis can catch many more events, which is why the new result matters.

How the 20-year analysis was done

The new paper, led by Norma Coe of the University of Pennsylvania with senior investigators including Marilyn Albert of Johns Hopkins, George Rebok, Michael Marsiske, Karlene Ball and Sherry Willis, did not bring participants back for new cognitive exams. Instead, it linked ACTIVE participants to Medicare claims from January 1, 1999 through December 31, 2019. Dementia was identified using the Chronic Conditions Warehouse algorithm, which flags Alzheimer’s disease and related dementias from diagnosis codes appearing in claims.

Because the method depends on fee-for-service claims, the analysis was restricted to people enrolled in traditional Medicare at baseline. That cut the analytic sample to 2,021 people, or 72 percent of the original 2,802. Over the 20 years, about three-quarters of the participants died, at an average age of about 84. That level of attrition by death is expected in a cohort that started at 65 and older, but it shapes how the results should be read, because people who die without a dementia diagnosis never contribute a dementia event. The authors addressed this with competing-risk models, which gave similar results according to coverage of the paper.

The headline numbers

The central comparison was between each training arm and the control group, with and without booster sessions. The reported results, as summarized in secondary coverage of the paper, were as follows:

• Speed training with boosters versus control: adjusted hazard ratio 0.75, 95 percent confidence interval 0.59 to 0.95. In raw terms, Johns Hopkins reported that 105 of 264 participants (40 percent) in this group were diagnosed with dementia, compared with 239 of 491 (49 percent) in the control group.
• Speed training without boosters versus control: hazard ratio 1.01, confidence interval 0.81 to 1.27. No difference.
• Speed training, boosters assigned versus eligible but not assigned: hazard ratio 0.81, confidence interval 0.66 to 1.00, which is borderline.
• Memory and reasoning training: no significant reduction in dementia hazard, with or without boosters.
• Initial training alone, any arm: point estimates suggested roughly 12 to 15 percent lower risk, but these were not statistically significant.

A hazard ratio of 0.75 means that at any given time during follow-up, people in the speed-plus-booster group were diagnosed at about three-quarters the rate of controls. The Hopkins release describes this as a 25 percent reduction in incidence. The confidence interval runs from 0.59 to 0.95, so the upper end sits close to no effect, and a reader should hold the point estimate loosely.

The pattern across arms is more informative than any single number. Only one of three interventions shows an effect, and only when paired with boosters. A result like that is harder to attribute to random noise than a uniform small benefit across all groups, but it is also the kind of result that invites suspicion about multiple comparisons. With three training types, with and without boosters, there were many possible comparisons, and one of them crossed the significance line.

Featured Partner

Invest in the Infrastructure Behind Modern Medicine

As healthcare expands beyond hospital walls, the buildings and campuses supporting that shift are generating compelling returns for investors who move early. The Healthcare Real Estate Fund offers qualified investors direct access to a curated portfolio of medical office, outpatient, and specialty care facilities.

Learn More →

Why speed training and not memory or reasoning?

This is the question the paper cannot fully answer, and the authors say so. Marilyn Albert has said the work “conveys a clear message but also leads us to ask many new questions.”

The leading hypothesis, offered in the Hopkins and PsyPost coverage, concerns the type of learning involved. Memory and reasoning training are strategy-based and explicit: a participant is taught a technique and consciously applies it. Speed training is adaptive and repetitive, and the learner improves by practice without necessarily being able to describe a strategy. That pattern fits what cognitive scientists call implicit or procedural learning. One reading is that this kind of learning may drive more durable changes in the brain systems that support attention and processing speed. This is a hypothesis, not a finding, and the trial was not designed to test it.

A related line of reasoning comes from the adaptive nature of the program. Because the software increases difficulty as the participant improves, each person stays near the edge of their ability throughout. Strategy training, by contrast, teaches a fixed method, and the benefit may plateau once the method is learned. Whether that difference explains the long-term result is unknown.

A third possibility is that processing speed and divided attention sit closer to the early cognitive changes that precede dementia. Slowed processing is one of the earliest measurable shifts in many people who later develop Alzheimer’s disease and vascular cognitive impairment. Training that targets those functions might act on something nearer to the disease process than training that targets verbal recall. That too is a story, not a demonstration.

The booster problem

The booster finding is the part most likely to be misunderstood. Speed training without boosters showed a hazard ratio of 1.01, which is no effect at all. The benefit appears only with the additional sessions. It is tempting to read that as a dose-response relationship: more training, more protection. That could be right. But the coverage of the paper flags a methodological wrinkle that matters.

Boosters were randomized, but the analysis of booster effects was limited to people who completed at least 8 of the 10 initial sessions. Completing most of the initial course is something that happens after randomization, and people who finish are not a random subset. They may be healthier, more motivated, more cognitively intact at baseline or better supported at home. Restricting to completers can therefore introduce selection bias that the original randomization was meant to prevent. The authors and the coverage are explicit that the findings should not be read as causal. This is why the borderline 0.81 result, which compares those assigned boosters with those eligible but not assigned, deserves attention. It sits closer to the intention-to-treat logic, and its confidence interval touches 1.00.

The limits of claims-based dementia

The outcome measure deserves its own section, because it is the point on which a skeptical reader should spend the most time.

Medicare claims record diagnoses that clinicians enter for billing. They are not the product of a standardized cognitive assessment. Dementia is widely underdiagnosed in routine care, and when it is diagnosed, the code may be applied late in the course of disease. Differences in diagnosis rates between groups can reflect differences in healthcare access, education and how often a person visits a physician, not only differences in underlying disease. PsyPost’s coverage notes this concern explicitly, and News-Medical lists claims-based identification and under-recording as limitations.

There is a subtler issue too. If cognitive training leaves people better at performing on cognitive tests or at managing daily life, it could change the likelihood that a clinician notices and codes dementia, without changing the biology. Better compensated people may be diagnosed later even if the pathology is the same. The study cannot distinguish a delay in the disease from a delay in its recognition. For the person living through it, a delayed diagnosis that reflects preserved function may still be worth having. For the science, the distinction is important.

The sample also excludes Medicare Advantage enrollees, a growing share of older Americans, and the original cohort was drawn from volunteers who agreed to take part in a trial. That limits how far the results travel to other populations.

What the result does not mean

Several tempting readings are not supported by the data:

• It does not show that brain games prevent dementia. The tested intervention was a specific, adaptive, clinically designed speed-of-processing program with a defined schedule of sessions and boosters. Commercial brain-training apps differ in design, difficulty and evidence base, and the ACTIVE result does not transfer to them automatically.
• It does not show that memory or reasoning training is useless. Those arms improved the abilities they targeted in earlier reports. They simply did not show a significant dementia effect in this analysis.
• It does not show that five weeks is enough. Without boosters, the speed arm showed no effect.
• It does not replace lifestyle and medical prevention. Blood pressure control, hearing correction, physical activity, and avoiding smoking have a larger and better-established evidence base for dementia risk. George Rebok framed the training as something that could be added to lifestyle interventions, not substituted for them.

How this fits the wider evidence

The ACTIVE result lands in a field that has been cautious about cognitive training. Large reviews have repeatedly found that people get better at the tasks they practice, with weaker evidence that the gains generalize to untrained abilities or to clinical outcomes. Dementia prevention trials of multidomain lifestyle programs, which combine exercise, diet, vascular risk management and cognitive activity, have produced modest and sometimes mixed results on cognitive scores. Against that backdrop, a randomized signal on a hard clinical outcome 20 years later is unusual, which is part of why it received attention.

It is also worth remembering what a single trial can establish. ACTIVE is a well-designed, federally funded trial, supported by National Institute on Aging grants including R01AG056486 for this analysis. But the dementia outcome here was a secondary analysis using linked administrative data, not a pre-specified primary endpoint measured by clinical examination at the time the trial was designed. Replication, ideally in a trial built around dementia incidence with adjudicated diagnoses, would carry far more weight.

The NIA’s leadership has been enthusiastic. NIA director Richard Hodes said moderate cognitive training could delay the onset of dementia over subsequent years. That is a reasonable reading of the point estimate. It is also the most optimistic reading, and the confidence interval permits smaller effects.

Open questions for researchers

Several questions will determine whether this finding becomes a clinical tool or a footnote:

• Replication. Does speed-of-processing training with boosters reduce clinically adjudicated dementia in a new, adequately powered trial?
• Mechanism. Do imaging or fluid biomarkers show differences in people who trained? Blood-based Alzheimer’s biomarkers, which are increasingly practical to collect, could show whether the training influences underlying pathology or only how it manifests.
• Dose and schedule. How many boosters are needed, how often, and for how long? Would a lifelong, low-intensity schedule work better than a handful of sessions?
• Heterogeneity. Who benefits most? Baseline processing speed, education, vascular risk and genetics such as APOE status may modify the effect.
• Delivery. The original program ran in supervised sessions. Home-based or tablet-based delivery would scale better, but it must be tested, not assumed to be equivalent.

What This Means For You

If you are an older adult, or you are caring for one, the practical takeaways from this trial are modest and useful.

Treat it as a promising add-on, not a primary strategy. The strongest evidence for lowering dementia risk still points to controlling blood pressure, treating hearing loss, staying physically active, avoiding smoking, managing diabetes and keeping social connections. If those are not in order, they come first.

Be skeptical of generic brain-training products. The ACTIVE speed program was adaptive, targeted visual attention and divided attention, and was delivered on a defined schedule with follow-up boosters. A game that merely feels mentally busy is not the same thing. If you are interested in speed-of-processing training, ask whether a program is based on the type used in ACTIVE and whether it has been tested in published trials, and be wary of marketing claims that cite this study without that connection.

Consistency may matter more than intensity. The benefit appeared with boosters, not the initial course alone. If you try a structured program, plan for returning to it periodically rather than treating it as a one-time course.

Do not use training to delay evaluation. If you or a family member notices memory or thinking changes, talk with a clinician. A cognitive exercise program is not a diagnostic test or a treatment for existing disease.

Watch for follow-up research. The question the field needs answered is whether this holds up in a new trial with clinically verified outcomes. Until then, the sensible stance is interest without certainty.

Twenty years is a long time to wait for an answer, and the answer is still partial. What ACTIVE shows is that a specific, inexpensive, low-risk intervention may leave a trace in dementia rates decades later, and that the trace depends on the kind of training and on keeping it going. That is enough to justify more research. It is not yet enough to change medical practice.

This article is for informational purposes and is not medical advice. Discuss cognitive health and any training program with a qualified clinician.

Sources

• Coe NB, Miller KEM, Sun C, et al. Impact of cognitive training on claims-based diagnosed dementia over 20 years: evidence from the ACTIVE study. Alzheimer’s & Dementia: Translational Research & Clinical Interventions, 2026. DOI 10.1002/trc2.70197.
• Johns Hopkins Medicine news release, February 2026: Cognitive speed training linked to lower dementia incidence up to 20 years later.
• News-Medical coverage, February and March 2026, and PsyPost coverage, February 2026, of the ACTIVE 20-year analysis.
• Ball K, et al. Effects of cognitive training interventions with older adults: a randomized controlled trial. JAMA, 2002. DOI 10.1001/jama.288.18.2271, PubMed 12425704.

Free Daily Briefing

The Latest Longevity Science.
Delivered Every Morning.

Join researchers, physicians, and health professionals getting daily breakthroughs in AI-driven medicine, epigenetics, and longevity research.

Support the research that powers this editorial

Subscribe Free

No spam. Unsubscribe anytime. We respect your inbox.

Keep Reading on Healthcare Discovery

• A Nasal Spray That Reverses Brain Aging: The Texas A&M Breakthrough Reshaping Alzheimer's Research
• Nurvv Run Smart Insoles: Pressure-Based Gait Analysis and Injury Prevention
• The Daily Rounds: Longevity & Health Care Brief | June 6, 2026