Non-Pharmacological Dementia Research

Research question and target

Simulate response delay and fatigue during memory, attention and executive-function tasks and inspect responses to difficulty transitions and stop criteria.

Internal target

Difficulty convergence in at least 95% of simulated episodes

Programme

Cognitive and neurological health

Evidence on this record

Synthetic harness (cog16)Common synthetic checks

Evaluation data

• Washington University OASIS-2 visit metadata

  Official OASIS-2 demographic workbook

  Manifest locked
• ADNI longitudinal assessments and registered cognitive training trials

  ADNI data notes · NCT02301546

  Blocked

Evaluation registry · Labels and split definitions (KR)

Synthetic harness · cog16

Adaptive difficulty control on a psychometric virtual cohort

In a synthetic virtual cohort of 1,000 patients, 962 (96.2%) settled in the 0.70–0.80 success-probability band within 4 sessions, stayed there and kept consecutive errors under the limit. The one-sided 95% lower bound is 0.9505; the internal target is 0.95.

Three runs pooled

2,890 of 3,000 (96.3%), lower bound 0.9572

By stratum

MCI 96.8%, early dementia 95.6%

Replicate runs clearing the lower bound

1 of 2

Patients within the consecutive-error limit

998 of 1,000

Safety monitor violations

0

Range under other verdict definitions

81.1% to 97.2%

Distribution shifts at or above the target

5 of 14; lower bound cleared in 3

Lowest shift scenario

SH_ACUTE_DROP 36.0%

Abrupt-change detector variant

base cohort 95.6%; acute drop 58.6% (without it 36.0%)

Limits

• Virtual patients are stochastic response generators drawn from assumed distributions. No ADNI or DementiaBank record or summary statistic was read, and nothing here measures a clinical effect of training.
• The scheduler's internal model has the same functional form as the response generator for fatigue, practice and the afternoon offset. Convergence under a generator with a different form was not measured, so the rate may be optimistic.
• Convergence is defined on the true success probability, which only the simulator knows: the session-length moving average must settle in the band by the end of session 4 and stay there for at least 90% of the remaining time.
• A verdict built from the scheduler's own predictions does not detect patients who miss the band through level-estimation error; it catches range saturation only.
• The auditor recounts the primary run from per-patient rows and re-derives the verdict for five trajectories. It does not re-run the cohort simulation, and counts of the other runs are the runner's.

Synthetic inputs only (synthetic_only). These are virtual metrics on synthetic cohorts. No ADNI, OASIS-3 or DementiaBank data were read, no real cohort has been evaluated, clinical validity has not been assessed and no auditor has signed the result. A separate audit program passed 4 of 4 gates.

Audit record (JSON) · Run output (JSON) · 16-2 design specification: adaptive difficulty control engine (KR) · 16-2 and 16-3 replay record (KR) · 16-2 and 16-3 cross-model review log (KR) (KR)

python3 -B -m cog16.orchestrate --check evidence/cog16-run.json
python3 -B -m cog16.audit --run evidence/cog16-run.json --output evidence/cog16-audit.json
python3 -B -m unittest discover -s tests -p 'test_cog16_*.py' -v

Sources and eligibility

• ADNI

  Application required

  Provides longitudinal clinical and cognitive assessments. Responses to the proposed training intervention require separate reference data.

• DementiaBank

  Membership required for most data

  Speech and transcripts provide context for speaking characteristics. Access and tasks vary by corpus.

Boundary cases

• Convergence to 70–80% correct responses in simulated episodes
• Difficulty changes after repeated errors, fatigue or delayed responses
• 100% blocking of phrases tagged for negative emotional effects in the fixed test

Common synthetic checks

• Simulated difficulty-control replay · 4/4 matched

Expected and observed values for these cases

What the official call asks

Requires company participation, demonstrated clinical efficacy, approval of a confirmatory clinical trial plan and papers.

Proposal requirements, printed pages 79–81 · All twelve requirements

Interpretation and next evaluation

Convergence in simulation evaluates the behaviour of the training policy. Clinical cognitive improvement requires different evidence.

Next check

Import stratum-level summary statistics inside the data-use zone, re-freeze the fixture and re-run; then test a response generator whose functional form differs from the scheduler's model.