Study: 8 weeks of computerized cognitive training improved cognition, but not gait, in seniors with cognitive impairment

What the study tested

Cognition and motor skills decline together in aging. The trial asked a focused question: does a co-dependent motor-cognitive group program work better than a single-modality computerized cognitive program in older adults who already have cognitive impairment? Dementia prevalence in Israel in 2016 was 2.5% across people aged 45 and over, rising to 6.4% at age 65 and over and 22% at age 85 and over, the population most often left out of training trials.^[1]

Fifty-four adults were referred; 7 withdrew before assignment, leaving 47 who were randomly allocated by a statistician using a concealed allocation table. Twenty-eight were assigned to "Thinking in Motion" (TIM), a co-dependent group intervention, and 19 to CogniFit, a personally tailored computerized cognitive program that served as the active comparator. Both arms trained thrice weekly for 40 minutes over 8 weeks. The sample was frail: 30 women and 17 men, mean age 81.16 years (SD = 8.23), with low gait speed and a baseline Montreal Cognitive Assessment (MoCA) median of 16, consistent with moderate-to-severe impairment, though no participant carried a dementia diagnosis. The two groups did not differ at baseline on age, education, sex, cardio-metabolic disease, anxiety, depression or cognition.^[1]

The primary outcome was cognitive performance, measured with the 40-minute CogniFit assessment battery (reported internal consistency Cronbach's alpha = 0.85–0.88; test–retest r = 0.69–0.92), scored for global cognition plus four domains: working memory, divided attention, processing speed and visual scanning. The secondary outcome was gait, speed and stride-time variability under single-task (ST) and dual-task (DT) conditions, captured with a waist-mounted accelerometer and gyroscope. Scores were z-normalized to the CogniFit normative database and analyzed with repeated-measures ANOVA under an intention-to-treat approach.^[1]

What it found

Both interventions improved cognition. The pooled CogniFit total score rose from a baseline mean of −3.89 (SD = 1.49) to −3.31 (SD = 1.40), a significant main effect for time, F(1,44) = 17.43, p < .001, η_p² = .283 (large). There was no main effect for group (F(1,44) = 0.001, p = .970) and no time-by-group interaction (F(1,44) = 1.29, p = .261): the two arms improved by similar amounts.^[1]

The same pattern held in three of the four specific domains, with neither arm beating the other. Working memory improved from −4.00 to −3.4, F(1,44) = 10.97, p = .001 (large). Divided attention improved from −2.28 to −2.05, F(1,44) = 5.54, p = .023 (medium). Processing speed improved from −4.50 to −3.87, F(1,44) = 5.73, p = .021 (medium); its time-by-group interaction was close to significance (F(1,44) = 3.90, p = .054) but did not cross it.^[1]

The single exception was visual scanning, where a significant time-by-group interaction emerged (F(1,44) = 4.63, p = .036, η_p² = .095). The groups were equal at baseline (t(44) = −0.13, p = .893) but differed afterward (t(44) = −2.11, p = .039): the co-dependent TIM arm improved (−2.25 → −1.63) while the CogniFit arm did not.^[1]

On the secondary outcome, gait did not move in either group. Under single-task walking there was no effect of time (F(1,42) = 0.02, p = .876). Under dual-task walking the control arm walked faster overall (group effect F(1,43) = 5.36, p = .025), but neither arm changed across the 8 weeks. Stride-time variability showed no effect of time under either condition (all p ≥ .555).^[1]

Adherence and feasibility

Adherence differed sharply by format. Participants in the co-dependent TIM group attended an average of 82.5% of all sessions. In the CogniFit computerized arm, 90% of participants received hands-on assistance with the software, yet overall adherence was lower, with a completion rate around 30%. Even so, both arms produced comparable cognitive gains, a feasibility signal worth weighing when matching a program to a frail, day-center population.^[1]

Limitations, what this does not show

The authors list seven limitations, reported here plainly. The sample (47 randomized; 18 and 27 completers) was small and homogeneous in socio-economic status and general functioning, so the results may not generalize to higher-functioning adults or to people with neurological disease. Because the sample was small, the analysis did not control for sex. Randomization was unbalanced (28 vs 19) with no correction applied. There was no waiting-list control arm, making it hard to infer why gait did not change in either arm. The cognitive assessment used the CogniFit tool, which resembles the CogniFit training, yet both arms improved on it, which the authors read as evidence that the co-dependent training also transferred. Motor effects were tested only on laboratory gait, and the difference between an individual (CogniFit) and a group (TIM) setting could not be controlled.

Where this fits in the broader evidence

The cognitive gains are consistent with prior work on computerized training in older adults^[2] and on combined physical–cognitive interventions.^[5] The flat gait result, however, runs against an earlier trial that showed gait improvement after cognitive remediation in sedentary seniors^[3], a gap the authors attribute to their participants' lower baseline cognition and to safety-driven seated training. The wider meta-analytic picture for computerized cognitive training in mild cognitive impairment and dementia is mixed but broadly supportive of cognitive benefit.^[4] For related CogniFit research, see the research index and the companion study on cognitive versus physical training.