Study: Eight weeks of cognitive training improved sleep quality and cognition in older adults with insomnia

What the study tested

Chronic insomnia is one of the most common complaints of later life: the authors cite prevalence estimates of 20% to nearly 50% in older adults. The authors note that the default treatment is medication, yet observe that sleeping pills carry risks of adverse effects and dependence and that their effectiveness "wanes rapidly after 30 days of use" (Haimov & Shatil, 2013). This trial asked whether structured new learning, a non-drug activity, was associated with changes in the participants' actigraphy-measured sleep.^[1]

Haimov and Shatil ran a randomized controlled, eleven-week proof-of-concept clinical trial, registered as NCT00901641. From 144 applicants recruited through local senior centres, 84 met the inclusion criteria: chronic insomnia under American Academy of Sleep Medicine (AASM) criteria, confirmed by a week of baseline wrist actigraphy, with poor sleep at least three nights per week for at least six months. Fifty-one participants aged 65–85 completed the study: 34 in the cognitive training group (mean age 73.2) and 17 in the active control group (mean age 69.9). At baseline, 31% of the training group and 21% of controls were using sleeping pills.^[1]

The intervention was an eight-week, home-based, personalized computerized cognitive training program (the CogniFit cognitive training program): 24 sessions of 20 to 30 minutes, three times a week, drawn from 21 training tasks at three difficulty levels, with difficulty continually adapted to each participant after a baseline evaluation. The active control group completed a time-matched, identically structured program ("Word and Paint"), twelve Microsoft Word and ten Microsoft Paint assignments that do not engage high-level cognitive functioning, adapt, or give feedback.

The primary outcome was sleep quality: five actigraph-derived parameters (sleep onset latency, sleep efficiency, total sleep time, wake after sleep onset, and number of awakenings), recorded over seven consecutive nights before and after the intervention. The secondary outcome was a 17-ability computerized neurocognitive evaluation administered in three 20-minute sessions at each time point. Because the training group was about 3 years older on average (p = 0.02), all mixed-models analyses controlled for age.^[1]

What it found

After eight weeks, the cognitive training group showed significant between-group improvements over the active control on two sleep parameters: sleep onset latency (F = 5.49, Cohen's d = −0.70) and sleep efficiency (F = 6.86, Cohen's d = 0.70), both medium-sized effects on Cohen's benchmarks.^[1]

Within the training group, four of the five sleep parameters improved significantly: sleep onset latency fell by 13.7 minutes (p < 0.001), sleep efficiency rose 6.7 percentage points (p < 0.001), wake after sleep onset dropped from 72.1 to 58.9 minutes (p < 0.05), and nightly awakenings fell from 10.7 to 9.0 (p < 0.05). Total sleep time rose about 14 minutes but did not reach significance. In the active control group, no significant change was observed on any of the five sleep variables.^[1]

Cognition moved in parallel. After controlling for age, the training group improved significantly on five cognitive measures, avoiding distractions, naming, general memory, visual memory and working memory, with Cohen's d values in the medium-to-high range. Of special interest, the control group showed a significant decline in working memory over the same eight weeks, a measure that had improved considerably under cognitive training.

Adherence was nearly identical across arms: 58% of those assigned to cognitive training and 68% of those assigned to the active control finished, for a 61% overall completion rate.^[1]

Did the cognitive gains drive the sleep gains?

The authors correlated each participant's change on the 16 analyzable cognitive abilities with their change on the five sleep parameters, then ran hierarchical linear regressions. In the training group, improved visual scanning predicted earlier sleep onset, improved naming predicted both less wake after sleep onset and fewer awakenings, and improved "avoiding distractions" predicted longer total sleep time (slope = 252.33, F = 5.71, adjusted R² = 0.20). In the control group, the steeper a participant's working-memory decline, the longer they took to fall asleep (adjusted R² = 0.46).^[1]

The authors are careful about mechanism: causal direction "cannot be inferred from correlation analysis," though they argue it is "quite likely" the cognitive improvements drove the sleep improvements, since the control group experienced neither.

Limitations, what this study does not show

The authors name attrition as the main limitation: only 51 of 84 eligible participants (61%) adhered to the training program. About half the non-completers cited technical problems, software crashes and glitches on their home computers, and 39% of the rest left for health reasons. Missing data trimmed the cognitive analyses further, to 45 participants (29 training, 16 control).^[1]

Other constraints: the sample was small, especially the 17-person control arm, and the groups were unequal at baseline on age and several cognitive scores, handled statistically rather than by design. Sleep was measured for one week before and one week after training, so the trial says nothing about whether the gains persist; the authors explicitly call for research into long-term effects and into combining cognitive training with cognitive behavioural therapy for insomnia (CBT-I). The cognition–sleep link is correlational, not proven causal. Finally, the program tested was CogniFit's own and a co-author was a CogniFit employee, an independent replication would carry more weight.

Where this fits in the broader evidence

The paper describes itself as "the first prospective study investigating the relation between learning, here operationalized as personalized cognitive training, and sleep quality in older individuals with insomnia." Its cognitive results replicate earlier trials of the same personalized training in other populations, healthy older adults, multiple sclerosis patients, and people with mood disorders [2, 3, 6], while its sleep results extend a literature showing learning reshapes sleep architecture in people without insomnia [5]. For all peer-reviewed CogniFit studies, see the research index.