Study: in cognitively healthy adults over 80, computerized brain training was no better than simple games for overall cognition

Adults aged 85 and older are among the fastest-growing segments of the U.S. population, yet most randomized trials of cognitive training had been run in the younger elderly; the few examining the old-old (75+) and oldest-old (85+) had yielded ambiguous results. This 2016 doctoral study tested, in cognitively normal adults aged 80 and older, whether an adaptive computerized cognitive training (CCT) program produced larger cognitive gains than an active control of simpler computer games. After 24 sessions, the two programs were statistically indistinguishable on overall and specific cognition, and the factor associated with improvement turned out to be participant education, not the program.

What the study tested

The trial was a double-blind, randomized, active-controlled clinical trial conducted through the City University of New York with the Alzheimer's Disease Research Center at Mount Sinai. 69 individuals enrolled and completed baseline assessment, randomized by computer to the CCT program (n = 39) or an active games control (n = 30), both produced by CogniFit; participants and examiners were blinded to assignment.^[1]

The sample was old and high-functioning: mean age 85.81 years, 65.2% women, and 71% had more than 15 years of education. Eligibility required a Mini-Mental State Examination (MMSE) score at or above the 25th percentile for age; the enrolled sample had a mean MMSE of 29.06 (range 25–30 out of 30), placing them between the 50th and 75th percentiles for their age on standard tests.^[1]

The intervention: each group trained in 20-minute sessions, every other day, for 24 sessions (roughly 7–8 weeks). The CCT program was adaptive, using a pre-training Neuropsychological Examination to weight each person's training toward their weakest functions (three tasks per session); the active control presented a fixed set of four simple games. The primary outcome was change in a global cognitive composite, with memory, attention/executive, and language composites secondary, all Z-score averages from a CERAD/Uniform Data Set neuropsychological battery administered independently of the training tasks. Analyses used intent-to-treat linear mixed models adjusting for age, sex, education, and sessions completed.

What it found

The headline result was null. No significant interaction of program with time was found on the global cognitive composite (F(1, 55.99) = 0.198, p = .658): the groups did not differ in how their overall cognition changed. The same held for every specific domain, attention/executive (F(1, 53.74) = 0.730, p = .397), language (F(1, 52.86) = 0.251, p = .618), and memory (F(1, 55.85) = 0.092, p = .763), and restricting analysis to those who completed at least 75% of training (31 CCT, 20 games) did not change it (global composite F(1, 48.96) = 0.057, p = .813).^[1]

There was also no significant main effect of time on the global composite (F(1, 53.23) = 2.177, p = .146); memory change reached only a trend level (F(1, 53.37) = 3.761, p = .058). The group as a whole neither improved nor declined over the training period.^[1]

These within-group gains are descriptive: the CCT group significantly improved on three of the 17 individual tests and the games group on one, with no significant declines. But with no program-by-time interaction in the mixed models, these changes "did not differentiate the two groups from each other." Adherence was comparable, 19.20 sessions for CCT vs 16.88 for games, not significant (t(67) = 1.005, p = .319), so training dose does not explain the null result.^[1]

The education effect

The one significant moderator was education, not program. Splitting the sample by college degree, the dissertation found a significant education-by-time interaction on the global cognitive composite (F(1, 55.08) = 4.755, p = .034) and on the language domain (F(1, 52.69) = 5.298, p = .025), with none for the other domains.^[1]

Within-group tests sharpened it: participants without a college degree improved significantly on overall cognition (t(13) = 2.59, p = .023), with trend-level gains for language (p = .066) and memory (p = .069). The college-educated group did not significantly improve on any composite. This benefit appeared regardless of program, the opposite of the original hypothesis, which had expected the training program specifically to help less-educated participants most.^[1]

Limitations, what this does NOT show

Small sample, sufficient only for modest effects. The author wrote that the sample "may have been too limited to identify smaller effect sizes (common in cognitive training studies)," and attrition further reduced statistical power.

A ceiling-prone, unrepresentative sample. Participants were cognitively healthy, computer-literate, well-educated, and "the vast majority were Caucasian." These successful cognitive agers scored highly at baseline, a ceiling effect was noted on Word List Memory recognition, leaving little room to improve. The dissertation states plainly that "results cannot be generalized to all old-old and oldest-old individuals."

Active control and home delivery may mask any difference. Both arms were a novel computer task that made people aware of their cognition, so the control may itself have been beneficial. The author also flags that purely home-based, unsupervised training may underperform supervised, group-based programs, that far transfer to the independent battery is generally weak in the oldest-old, and that no long-term follow-up was conducted.

Where this fits in the broader evidence

This null result sits within a mixed literature. Reviews and meta-analyses such as Papp et al. (2009) and Lampit et al. (2014) reported that effect sizes for computerized cognitive training in healthy older adults are generally small, and that home-based programs tend to underperform supervised, group-based formats, consistent with this trial's failure to separate an adaptive program from a simpler active control in high-functioning adults over 80. By contrast, studies in clinical and impaired populations, where there is more room to improve, have more often reported gains. For the full catalogue of research on CogniFit's technology, see the research index; for population-specific evidence see the study pages on long COVID brain fog and adults over 60.