A transient decrease in night-to-night variability of REM sleep before myocardial infarction

Background. Myocardial infarction (MI) outcomes depend on time to treatment, and fatigue and sleep disturbance are often reported in the weeks before the event. Whether wearable-measured sleep changes within individuals before MI, and whether such change is specific to MI, is unknown. Methods. In the All of Us Research Program, adults with a first MI code and at least 180 valid Fitbit sleep nights in the preceding 360 days were matched to up to 10 controls on sex, comorbidities, sleep apnea, calendar time and a propensity score. The full matched cohort of all cases was the primary cohort, and cases with diabetes and/or hypertension formed a subgroup. The main endpoints were the within-person change in night-to-night variability (SD, RMSSD, CV) of the REM-sleep fraction at days 30–16 relative to each person’s baseline (days 360–121) and from days 30–16 to days 15–1, tested against 10,000 randomly placed index dates with allowance for the choice of window (max-T). The same pipeline was applied to cases without recorded symptoms or treatment, to an index date at the first angina code, to a cohort including heart failure and to ischaemic stroke and admissions for heart failure, pneumonia, fracture or appendicitis and any cause, and the analysis was adjusted for nightly heart rate, wear and device. Results. The full matched cohort comprised 45 cases and 450 controls. REM-fraction variability changed transiently before MI, decreasing at days 30–16 and returning towards baseline over days 15–1; allowing for the choice of window, this decrease-then-return pattern was rarely matched at 10,000 randomly placed index dates (max-T p<0.001). Its components were the decrease in SD at days 30–16 (Cliff’s δ −0.30, p=0.004; OR per SD 0.53, 95% CI 0.35–0.80; max-T p=0.0034) and the return in RMSSD over days 15–1 (δ +0.36, p<0.001; max-T p=0.0035); day by day, the decrease was not detected as a continuous change (cluster permutation test p=0.58). The decrease in SD was not seen at the same time before admissions for pneumonia (69 cases), fracture or appendicitis (130) or any cause (859) and was larger before MI than before each of them (difference in δ −0.25 to −0.34), whereas the rise over days 15–1 was shared with pneumonia; it was of similar size in the subgroup with diabetes and/or hypertension, in cases without recorded symptoms or treatment, with the index date at the first angina code, when people with heart failure were included and after adjustment for nightly heart rate, wear and device. Scoring each person against their own baseline discriminated cases from controls with AUCs of up to 0.66, probably an optimistic estimate because the decrease in RMSSD was attenuated in held-out halves of the matched sets. Conclusions. In adults who use consumer wearables, a first MI was preceded by a transient change in night-to-night REM-sleep variability, a decrease at days 30–16 and a return over the following two weeks. The decrease was not seen at the same time before other acute admissions, did not depend on recorded symptoms or treatment and was detectable against each person’s own baseline with simple measures. The timing is an exploratory finding, to be confirmed in new cases.