Sleep Medicine

Machine-learning analysis reveals major heterogeneity in CPAP cardiovascular effects in non-sleepy OSA

A causal survival forest was applied post hoc to SAVE trial data from 2,687 non-sleepy OSA patients to estimate individualized CPAP treatment effects. The model showed substantial heterogeneity in cardiovascular response, with strong discrimination between predicted benefit and harm (target operator AUC 2.6; 95% CI 2.03–4.55). Patients in the tertile predicted to benefit from CPAP had roughly 100-fold better event-free survival when randomized to CPAP. Those in the tertile predicted to be harmed had more than 100-fold higher major adverse cardiovascular event rates on CPAP. 1

Inpatient PSG associates with higher 1-year mortality and costs versus early post-discharge PSG

This Ontario population-based cohort compared adults receiving inpatient PSG, delayed PSG within one month of discharge, or no PSG after hospitalization. In overlap-weighted analyses, inpatient PSG versus delayed PSG was associated with higher 1-year mortality (HR 2.18; 95% CI 1.68–2.82) and higher costs (RoM 1.82; 95% CI 1.64–2.03). Compared with matched non-PSG controls, inpatient PSG showed lower mortality (HR 0.72; 95% CI 0.55–0.94) but higher costs (RoM 1.73; 95% CI 1.50–1.99). Delayed PSG done within 3 or 6 months after discharge was associated with lower costs than no PSG, although not when limited to one month. 2

Digital CBT-I has a non-linear dose–response with near-maximal benefit at about 250 treatment minutes

A model-based network meta-analysis of 73 RCTs (14,465 participants) quantified the dose–response between digital CBT-I duration and Insomnia Severity Index improvement. Maximum estimated efficacy was a 5.77-point ISI reduction versus control (95% CrI −7.51 to −4.68). The ED50 was 44.55 minutes, with comorbid insomnia showing greater sensitivity than isolated insomnia. Protocols around 250 minutes achieved optimal effectiveness (mean ISI difference −7.95; 95% CrI −11.77 to −4.22), with effects plateauing beyond 300 minutes. 3

Sleep EEG brain age index independently predicts incident dementia across large community cohorts

This individual participant data meta-analysis pooled 7,105 adults from five longitudinal cohorts with home polysomnography and dementia adjudication. A machine-learning brain age index quantified the deviation between EEG-derived brain age and chronological age using central sleep EEG channels. Across cohorts, each 10-year increase in brain age index was associated with a substantially higher risk of incident dementia, accounting for death as a competing risk. Median time to dementia after PSG ranged from roughly 3.6 to 16.9 years across cohorts, indicating long-term prognostic value. 4