Does APOE4 modify how lifestyle interventions affect cognition?

Three trials pooled into one interaction test

The paper asks a specific question: does carrying the apolipoprotein E ε4 allele (APOE4), the main genetic risk factor for dementia incidence and earlier onset, change how much a multidomain lifestyle intervention improves cognition? To answer it, the authors combined harmonised estimates from three completed randomised trials that tested multidomain lifestyle interventions against control: FINGER in Finland, MAPT in France and J-MINT in Japan.

The trials enrolled different risk profiles. FINGER selected older adults at risk of dementia using the CAIDE dementia risk score plus average or below-average cognitive performance; MAPT enrolled community-dwelling older adults with subjective cognitive complaints or signs of frailty; J-MINT enrolled people with mild cognitive impairment (MCI). Participants were aged 60 to 85. APOE genotype was dichotomised, so carriers of one or two APOE4 alleles were compared with non-carriers.

The design matters for reading the result. This is not a new trial but a meta-analysis of existing trial data, using a random-effects model with maximum likelihood estimation. The main analysis used 24-month data to harmonise trials whose intervention durations ran from 18 months in J-MINT to 3 years in MAPT; a secondary analysis used 12 months because that was the last time point observed in all three trials.

The outcome is a difference in differences

The key statistic is not simply how much carriers improved. For each trial the authors computed an effect size comparing the intervention-minus-control change in APOE4 carriers with the intervention-minus-control change in non-carriers. A positive value therefore means carriers benefited more from the intervention than non-carriers; it does not mean carriers improved in absolute terms more than everyone else.

This distinction is easy to lose. The reported interaction p-values test whether that carrier-versus-non-carrier difference in intervention effect is statistically distinguishable from zero, using a two-sided z-test without adjustment for multiple comparisons. Only when an interaction signal appeared did the authors run stratified analyses of the intervention effect separately within carriers and within non-carriers.

The cognitive outcomes also differ in coverage. Global cognition was each trial's own primary composite, but the tests making up that composite were not the same across studies. Executive function was a prespecified secondary outcome in FINGER and J-MINT and was calculated for this meta-analysis in MAPT. Memory and processing speed were available only from FINGER and J-MINT, so those domains rest on two trials rather than three.

Global cognition favours carriers at 12 months

For global cognition, the pooled difference in effect size between carriers and non-carriers was 0.188 (95% CI 0.009 to 0.437) at 12 months, with an interaction p-value of 0.039. At 24 months the pooled difference was 0.173 (95% CI −0.006 to 0.352) and the interaction p-value was 0.058, described by the authors as borderline.

The direction was consistent across the three trials even though no individual trial's confidence interval excluded zero: FINGER 0.151, J-MINT 0.362 and MAPT 0.147 at 12 months. The pooled estimate is what crossed the significance threshold, which is why the paper frames the finding as meta-analytic rather than as a result established in any single trial.

The stratified analyses add a second layer. Among APOE4 carriers, the intervention effect on global cognition was significant at both 12 and 24 months; among non-carriers it was not. That pattern is the basis for the authors' conclusion that carriers are a key target group for lifestyle-based prevention.

Executive function shows the clearest domain signal

When the authors split cognition into domains, the APOE4 effect modification looked more related to executive function than to memory or processing speed. For executive function the pooled carrier-versus-non-carrier difference was 0.179 (95% CI −0.001 to 0.358) at 12 months with p = 0.050, and 0.168 (95% CI −0.012 to 0.347) at 24 months with p = 0.067.

Those p-values sit right at or just above the conventional threshold, so the honest reading is a consistent direction rather than a firmly established domain-specific effect. The contrast with the other domains is starker. Memory at 24 months gave a pooled difference of 0.150 (95% CI −0.066 to 0.366, p = 0.174), and processing speed gave 0.006 (95% CI −0.211 to 0.222, p = 0.959), essentially no signal.

Because memory and processing speed came only from FINGER and J-MINT, the domain comparison is also uneven in evidence base. The paper's claim is therefore best stated as the APOE4 effect appearing more related to executive function than to memory or processing speed, not as proof that other domains are unaffected.

The MCI trial shows the widest carrier gap

The authors report that the APOE4 effect on intervention benefit was greatest in J-MINT, the trial targeting MCI, where effect sizes among carriers were roughly twofold those in FINGER or MAPT and benefit was less evident among non-carriers. In the table, J-MINT's global cognition difference is 0.362 at both 12 and 24 months, against 0.151 in FINGER and 0.147 or 0.100 in MAPT.

The authors interpret this pattern as suggesting that the greater benefit for carriers may be more evident in early cognitive risk stages such as MCI than in broader at-risk groups identified by other risk factors or dementia risk scores. That is the authors' explanation of a between-trial contrast, not a tested head-to-head comparison of risk stages, and the trial-level confidence intervals overlap.

What the meta-analysis does support is breadth: the modification appeared across genetically and culturally different populations from Asia and Europe and across target groups ranging from at-risk older adults without substantial impairment to people with MCI. The paper describes this as the first meta-analytic assessment of the question.

What the pooled estimate does not settle

Several features limit how far the result travels. The interaction tests were not adjusted for multiple comparisons, and the 24-month global cognition result is borderline rather than significant. The executive function result at 12 months sits exactly at p = 0.050. The authors themselves frame the domain finding as the effect appearing more related to executive function, language that stops short of a confirmed domain-specific mechanism.

The harmonisation choices also shape the numbers. MAPT combined its two lifestyle arms and excluded the supplement-only control arm, with a sensitivity analysis treating the arms separately; the third year of MAPT was excluded; and MAPT's non-linear change was modelled with separate year 1 and year 2 estimates while FINGER and J-MINT change was treated as linear. Global cognition composites were not identical across trials.

Finally, the outcome is cognitive change on trial composites, not incident dementia. The paper's conclusion concerns APOE4 carriers as a target group for lifestyle-based prevention among people with MCI or earlier at-risk stages, and that is the scope the evidence covers.

Carriers with MCI are the clearest priority

The practical weight of this paper falls on a comparison rather than a general recommendation. Within the same three-trial evidence base, the intervention benefit on global cognition and executive function was significant among APOE4 carriers at both 12 and 24 months and not among non-carriers, and the carrier advantage was largest in the MCI trial.

So the priority group the data point to is APOE4 carriers already showing early cognitive risk, particularly MCI, rather than all older adults at risk. The 12-month global cognition interaction (p = 0.039) is the firmest single number; the 24-month result (p = 0.058) and the executive function results (p = 0.050 and 0.067) should be read as consistent but not decisive.

For anyone weighing whether APOE4 genotyping changes how a multidomain lifestyle programme should be targeted, this paper's contribution is that the modifier question has now been tested meta-analytically across FINGER, MAPT and J-MINT, with the carrier advantage concentrated in the earliest risk stage and in executive function rather than memory or processing speed.

Sources used in this answer