Early prediction of post-stroke depression using polysomnography parameters: a prospective cohort study

Background Post-stroke depression (PSD) affects approximately one-third of stroke survivors and significantly impacts functional recovery. This study aimed to develop and validate a prediction model for PSD using polysomnography (PSG) parameters combined with clinical characteristics. Methods We conducted a prospective cohort study enrolling 437 acute ischemic stroke patients who underwent PSG assessment within 7 days of stroke onset between January 2022 and December 2024. The primary outcome was PSD at 3-month follow-up, defined as Patient Health Questionnaire-9 (PHQ-9) score ≥10. Multivariate logistic regression identified independent predictors, and machine learning algorithms were compared for model performance. Results Among 390 patients completing follow-up, 139 (35.6%) developed PSD. Independent predictors included sleep latency (OR = 2.14, 95% CI: 1.68–2.73), arousal index (OR = 2.13, 95% CI: 1.65–2.94), sleep efficiency (OR = 0.74, 95% CI: 0.55–0.96), heart rate variability RMSSD (OR = 0.77, 95% CI: 0.61–0.96), NIHSS score (OR = 1.35, 95% CI: 1.04–1.76), and prior stroke history (OR = 1.35, 95% CI: 1.09–1.71). The gradient boosting model achieved the highest discriminative performance (AUC = 0.763; bootstrap-validated AUC = 0.738, 95% CI: 0.682–0.794). Risk stratification demonstrated a five-fold gradient in PSD rates across probability categories (14.3 to 70.3%). Conclusion PSG parameters, particularly sleep efficiency, sleep latency, and arousal index, are significant independent predictors of PSD. Integrating objective sleep assessment into early stroke management may facilitate identification of high-risk patients for targeted preventive interventions.