Why do large CPAP trials show little cognitive benefit while small trials show gains, and could high baseline scores explain the difference?

A flat average can hide opposite responses

A randomized trial reports one number: the average change in the treated group minus the average change in the comparison group. If treatment helps some participants and hurts others by similar amounts, those effects cancel and the average sits near zero even though real, opposing responses exist. That is the pattern reported in a secondary analysis of APPLES, a large sham-controlled CPAP trial. The primary analysis had found no significant difference in a working-memory measure at six months, and the reanalysis confirmed a non-significant overall effect [1].

When the analysts split participants by estimated individual treatment effect, the picture changed. One group of 285 people had an estimated average gain of about 0.68 points on the cognitive index, and another group of 328 had an estimated average loss of about 0.64 points, both statistically distinguishable from zero [1]. The units are index points on a working-memory score, so the sizes are modest — but the direction is what matters: the overall null was an average of gains and losses, not evidence that CPAP does nothing for thinking.

Who looked like a responder, and who did not

The group that appeared to benefit was characterized by more males, higher baseline cognitive scores, a lower EEG-derived brain age index (a measure built from brain-wave patterns that estimates how "old" the brain looks), and slightly more depression. The group that appeared to worsen had the opposite profile [1]. This is the strongest direct support for the idea in the question: baseline cognitive score was one of the features separating likely improvers from likely non-improvers, alongside sex and the brain-age measure.

Two cautions belong right next to that finding. First, the exposure in this reanalysis was not the original random assignment: because people assigned to active CPAP adhered better than those on sham, the analysts compared adherent active-CPAP users against everyone else, a non-randomized contrast, and adjusted for confounders by matching on clustered covariates [1]. The authors themselves frame the conclusion as conditional — "assuming the ITE is unbiased" — so the subgroup split is a hypothesis about who responds, not a proven causal map. Second, the responder profile is a description of a group, not a test you could apply to an individual patient.

Why small studies can still show gains

Small studies tend to enroll patients who are clearly affected and often moderately to severely ill, and they measure change within those patients rather than averaging across a broad, heterogeneous sample. A study of 21 patients with moderate-to-severe OSA found that one month of CPAP adherence was accompanied by changes in regional brain-activity synchrony across several areas, and that the pre-treatment abnormalities in those measures correlated with Montreal Cognitive Assessment scores, daytime sleepiness, and apnea severity [2]. That is a neural correlate of treatment response in a selected, symptomatic group — consistent with the idea that a narrowly chosen sample can show a signal that a broad sample dilutes.

It is worth being precise about what that study shows. It reports associations between brain-activity measures and clinical scores before and after short-term treatment; it does not demonstrate that CPAP caused cognitive improvement, and it did not compare large and small trials [2]. The same restraint applies to work on predicting who improves in daytime function after six months of CPAP, which examined baseline polysomnographic predictors of response and noted that several studies had found no clear cognitive effect [3]. Together these point to baseline patient characteristics — severity, symptoms, and starting cognitive level — as the place to look for the split, rather than to trial size itself.

What to take away if you or someone you know is starting CPAP

The practical reading is that "CPAP doesn't improve thinking" is an average statement about a mixed population, not a verdict for every patient. The evidence is consistent with a subset — plausibly those with relatively intact cognition to begin with and a brain-activity profile closer to healthy — gaining something, while others gain little or even do worse on some measures [1]. That is a possibility supported by subgroup estimates, not an established rule, and no study here shows that starting with a high cognitive score causes a better response.

The honest state of the evidence is that the responder profile is still being defined. The APPLES reanalysis gives a concrete candidate set of markers — sex, baseline cognition, brain age index, mood — but derived them from a non-randomized contrast within one trial [1]. The smaller imaging and predictor studies add plausibility by showing that treatment-related brain and cognitive changes track baseline severity and scores [2][3], yet they were not designed to explain the large-versus-small trial discrepancy. A pilot study comparing driving, eye movements, and cognition before and after CPAP adds another small, within-patient data point but does not resolve the question [4]. The most useful next step for the field would be testing the responder hypothesis prospectively in a trial designed to detect subgroup effects, rather than reading the average as the whole story.

Sources used in this answer

Related research questions