Health & function · Evidence Explainer

Cognitive training improved practiced skills—not survival or proven dementia prevention

ACTIVE supports protocol-specific gains in trained abilities and some long-term function measures. Twenty-year mortality was null, while a lower claims-diagnosed dementia estimate appeared only in a nonrandomized booster-completer subgroup that methodologists dispute.

Reviewed under standing publication authorization

Discovery, counterevidence, claim mapping, identifiers, funding, correction state and prose were reviewed before release.

Evidence box

Three evidence lanes—not one brain-training verdict

Study type
Randomized cognitive-training trial with ten- and twenty-year linked follow-ups
Studied in
Independent US adults age 65–94 assigned to memory, reasoning, speed-of-processing training or no-contact control
Participants / sample
2,832 randomized; 2,802 in mortality analysis; 2,021 in the claims-linked dementia analysis
Publication status
Peer-reviewed primary reports and current registry records; correction status checked August 2, 2026
Outcome type
Trained abilities, daily function, all-cause mortality and claims-diagnosed ADRD
Development stage
Protocol-specific research interventions; generic games and commercial descendants are not interchangeable
Conflicts / funding
NIH support; mortality report declared no conflicts; commercial descriptions are not efficacy evidence
Our assessment
Trained-ability benefits and some function signals; no mortality benefit and no reliable randomized dementia-prevention conclusion
Reporting confidence
High for trained abilities and null mortality; low for the disputed booster subgroup

Start here

Key takeaways

  • Training improved the abilities participants practiced; ten-year persistence was detectable for reasoning and speed, not memory.
  • The 20-year analysis found no significant all-cause mortality effect for any training arm.
  • The reported HR 0.75 for claims-diagnosed ADRD appeared only among booster-completing speed trainees; randomized arms were null, non-boosters had HR 1.01, and the subgroup analysis is contested.

Keep the outcomes separate

What we know / what we do not know

What we know

  • The trial provides protocol-specific randomized evidence for trained abilities and some long-term daily-function measures.
  • The long-term mortality analysis directly found no significant survival effect from memory, reasoning or speed training.

What we do not know

  • Whether booster completion caused the claims-based dementia difference, because booster completion was not randomized.
  • Whether generic brain games or commercial implementations reproduce the studied protocols or prevent dementia.

Evidence at a glance

The answer changes with the outcome.

Rows preserve outcome boundaries; they are not a score, ranking, or recommendation.

Evidence by outcome
OutcomeResultBoundary
Trained ability and functionMemory, reasoning and speed improved initially; reasoning and speed persisted at ten years, with reported function signals.Practiced-domain performance is not generalized rejuvenation.
All-cause mortalityNo significant effect for memory, reasoning or speed training.The trial does not support a lifespan claim.
Claims-diagnosed dementiaHR 0.75 only in booster completers; randomized arms null and non-boosters HR 1.01.Post-randomization, claims-based and methodologically contested.
01

The bottom line

ACTIVE has three evidence lanes that must remain separate. Training improved specifically practiced abilities and some long-term function measures. The 20-year mortality analysis found no significant effect on all-cause mortality. A lower claims-diagnosed dementia estimate appeared only in speed trainees who completed boosters—a post-randomization subgroup whose interpretation is contested.

02

Why this matters

“Brain training works” can refer to doing better on practiced tasks, preserving daily function, preventing dementia, or living longer. Those are different claims. ACTIVE is valuable because its long follow-up shows both where evidence persists and where the stronger story fails.

03

What researchers did

ACTIVE randomized 2,832 independent adults age 65–94 to memory, reasoning, speed-of-processing training or no-contact control. Some eligible trainees were later offered boosters. Booster completion was not a fresh randomized assignment, so completers and non-completers cannot be assumed comparable.

04

What they found

At ten years, reasoning and speed—but not memory—retained detectable trained-ability effects. Intervention groups reported less decline in instrumental daily activities, while self-reported and performance-based function remained distinct. In the mortality analysis, 2,021 of 2,802 participants died through 2019; no training arm significantly changed all-cause mortality.

05

How strong is the evidence?

Randomization supports protocol-specific trained-ability conclusions. Long follow-up strengthens the mortality result. The claims-based dementia analysis is weaker: randomized arm comparisons were null, while HR 0.75 (95% CI 0.59–0.95) appeared only among booster-completing speed trainees; non-booster speed trainees had HR 1.01 (95% CI 0.81–1.27).

06

What this evidence does not show

ACTIVE does not establish that generic brain games prevent dementia, extend life, or rejuvenate the brain. It does not show that the booster subgroup difference was caused by boosters rather than post-randomization selection or other differences.

07

Safety, interpretation, and funding

The claims-based dementia paper, published methodological critique and authors’ response belong together. Critics raised comparability, attrition and selection, claims diagnosis, lack of prespecification and multiple unadjusted subgroup comparisons. NIH supported ACTIVE and the long-term work; commercial product descriptions are not independent evidence.

08

What happens next

A stronger dementia test would preserve randomization, prespecify the clinical endpoint and booster comparison, retain participants, verify diagnoses beyond claims, adjust planned multiplicity and replicate the protocol independently.

Primary sources

Sources, roles and limits

  1. Ball et al., original ACTIVE randomized report
    Identifier
    PMID:12425704 · DOI:10.1001/jama.288.18.2271 · NCT00298558
    Role
    Primary randomized trained-ability evidence
    Limitation
    Protocol-specific cognitive outcomes; not dementia or survival.
  2. Rebok et al., ten-year ACTIVE follow-up
    Identifier
    PMID:24417410 · PMCID:PMC4055506 · DOI:10.1111/jgs.12607
    Role
    Long-term ability and function evidence
    Limitation
    Self-reported and performance-based function remain distinct.
  3. Chen et al., twenty-year mortality analysis
    Identifier
    PMID:39358997 · PMCID:PMC12234222 · DOI:10.1093/aje/kwae381
    Role
    Primary long-term null counterevidence
    Limitation
    Linked follow-up; no significant all-cause mortality effect by training arm.
  4. ACTIVE claims-based dementia analysis
    Identifier
    PMID:41669119 · PMCID:PMC12884427 · DOI:10.1002/trc2.70197
    Role
    Secondary claims-linked subgroup evidence
    Limitation
    Randomized arms were null; booster completion was not randomized.
  5. Published methodological critique
    Identifier
    DOI:10.1002/trc2.70280
    Role
    Contradictory methodological assessment
    Limitation
    Raises selection, claims, prespecification and multiplicity concerns.
  6. Authors’ response to critique
    Identifier
    PMID:42454025 · DOI:10.1002/trc2.70290
    Role
    Response and dispute context
    Limitation
    A response does not resolve the need for randomized replication.

Connected topics

Continue through the knowledge system

Intervention categoriesStudy and trial recordsHow evidence differsHealth & function stories

Continue with real coverage

Keep reading

These are learning paths, not rankings, recommendations, or evidence that the stories support one another.

Small cognitive-test differences without a dementia endpoint

Research AnalysisHealth & function

Multivitamins moved some tests and aging clocks. What does that mean?

COSMOS found small cognitive-test differences and modest movement in two DNA-methylation clocks; neither establishes dementia prevention or longer life.

Evidence scopeHuman cognitive tests and biomarkers · correlated randomized substudies

Disclosures and history

  • August 2, 2026 — source discovery, verification and counterevidence completed before prose drafting.
  • August 2, 2026 — atomic claim map, article draft, identifiers, funding, correction state and inference boundaries reviewed.
  • Corrections: none as of publication; ACTIVE’s published methodological dispute is shown in the evidence record.