Across 16,884 adults from three longitudinal cohorts (ELSA, Whitehall II, and Rush MAP), difficulty falling asleep at least once per week in midlife was associated with meaningfully faster 5-year cognitive decline — a 0.12 SD reduction in recall at age 50 — yet this association attenuated and reversed at older ages, with late-life insomnia paradoxically correlating with relatively preserved scores. Short sleep (≤6 hours) showed comparable age-gradient patterns, though these attenuated after adjustment for health and lifestyle confounders. The age-by-sleep interaction was statistically robust (p=0.009 linear, p=0.016 quadratic), and the difficulty-falling-asleep pattern replicated across all three cohorts spanning ages 44 to 102.
The finding reframes a long-contested debate: insomnia and short sleep are not uniformly harmful or benign across the lifespan. The midlife signal supports sleep disturbance as a genuinely modifiable dementia risk factor — consistent with mechanistic evidence linking chronic sleep disruption to amyloid accumulation and neuroinflammation. The late-life reversal is clinically important: it suggests that poor sleep in older adults may reflect prodromal neurodegeneration rather than cause it, meaning treating insomnia at that stage may not yield the same prevention dividend. For adults in their 40s–60s, this argues strongly for addressing difficulty initiating sleep as a priority health behaviour. Limitations include reliance on self-reported sleep, observational design precluding causal inference, and differential attrition in older age bands. As a preprint not yet peer-reviewed, these conclusions remain provisional and warrant replication with objective sleep measurement.