Exercise and Cognition: Part Two: Which Kind, and For Whom

Abstract

Part One established that exercise improves cognitive test scores across a very large trial literature. The obvious next question is which exercise, and the answer turns out to depend on who is being asked.

A network meta-analysis published in the Journal of Sport and Health Science ranked exercise types against each other in people with mild cognitive impairment or dementia — 71 trials and 5,606 participants.

Resistance training came first overall. It also produced an effect size large enough to warrant scepticism, and the ranking reversed in the subgroup that most readers would care about.

1.What a Network Meta-Analysis Does

Ordinary meta-analysis pools trials comparing one thing against a control. It cannot tell you whether aerobic exercise beats resistance training unless trials have compared them directly, and few have.

A network meta-analysis reconstructs those comparisons indirectly. If aerobic exercise beat control by a certain margin in one set of trials, and resistance training beat control by a different margin in another, the two can be ranked against each other through their shared comparator.

The method is powerful and it inherits every weakness of the underlying trials, plus an assumption that the trial populations were similar enough for indirect comparison to be valid. Rankings from network meta-analyses are routinely reported with more confidence than they deserve.

Xin Huang and colleagues searched six databases through September 2019 and included randomised controlled trials in patients with mild cognitive impairment or dementia, with global cognition, executive cognition and memory as primary outcomes.

2.The Ranking

All types of exercise were effective in increasing or maintaining global cognition. That is the first and most robust finding: no modality failed.

Resistance exercise had the highest probability of being the most effective intervention — for slowing decline in global cognition, with a standardised mean difference of 1.05 and a confidence interval from 0.56 to 1.54; for executive function at 0.85, interval 0.21 to 1.49; and for memory at 0.32, interval 0.01 to 0.63.

The memory interval runs from 0.01 to 0.63. Its lower bound is one hundredth of a standard deviation — statistically significant and, at that end of the range, indistinguishable from nothing.

The global cognition figure deserves the opposite scrutiny. A standardised mean difference of 1.05 means the average treated participant scored better than about 85 per cent of controls. For a behavioural intervention in people with dementia, sustained over months, that is larger than the effect of any licensed drug in the condition.

Effect sizes of that magnitude in small trials are usually a signal about the trials rather than the treatment. Blinding is essentially impossible in exercise research — participants know whether they are lifting weights — and outcome assessment in this literature is frequently unblinded too.

3.The Subgroup That Reverses It

Subgroup analysis for patients with mild cognitive impairment — the earlier, milder condition — produced a different answer. There, multicomponent exercise was most likely to be the optimal therapy.

Multicomponent means a programme combining several modalities: aerobic work, resistance, balance, sometimes coordination or dual-task elements.

That two adjacent populations produce different winners is worth more attention than either ranking. It could mean the optimal intervention genuinely differs by disease stage, which would be clinically important. It could also mean the rankings are unstable — which is what one would expect when a network is reconstructed from indirect comparisons across trials of differing quality.

The review does not adjudicate, and neither should a reader.

4.Reading This Against Part One

The two syntheses disagree in an instructive way.

The umbrella review in Part One, across all populations, found larger effects for low- and moderate-intensity interventions. This network meta-analysis, in cognitively impaired patients, ranks resistance training first — which is not a low-intensity modality.

Both can be true if the optimal intervention depends on the population, which is the charitable reading and probably part of the answer. Both can also reflect the instability of subgroup findings in literatures assembled from many small unblinded trials.

The practical consequence is that anyone claiming to know which exercise is best for cognition is over-reading the evidence. What both syntheses agree on is that the modalities all beat doing nothing.

5.A Note on the Population

Everything in this part concerns people with mild cognitive impairment or dementia. These are clinical diagnoses, and the trials measured slowing of decline rather than improvement in healthy function.

Applying a ranking derived from that population to a healthy adult choosing between the gym and a run is exactly the extrapolation this library exists to flag. It may hold. It has not been shown.

Editorial Comment

MindHeaven® makes no claim relating to exercise, and has no product that substitutes for it.

We include this part because the same critical habits apply here as anywhere. A standardised mean difference above 1.0 from unblinded trials in a small literature should raise an eyebrow whether the intervention is a dumbbell or a capsule — and we would say so loudly if a supplement produced that number.

Cognitive impairment and dementia are medical diagnoses, and exercise programmes for people who have them belong with a clinician who knows their cardiovascular and musculoskeletal status. Nothing here is a training prescription.

Part Three asks the question that matters most and has the least encouraging answer: whether being physically active across decades actually changes how cognition ages.