Development and internal validation of an early pre-sleep prediction model for first-night postoperative sleep disturbance: a prospective cohort study

Abstract

Abstract Background Postoperative sleep disturbance (PSD) is common and may impair recovery. Sleep quality during the first postoperative night is important in early recovery and offers an immediate opportunity for symptom-directed care. Several PSD prediction models have been reported, but most predict sleep disturbance at 1 week or within 7 days after surgery. We developed and internally validated a pre-sleep model for first-night PSD using variables available before the first postoperative night. Methods This single-center prospective cohort study included 205 adults undergoing elective surgery. First-night PSD was assessed on the morning of postoperative day 1 and defined as an Athens Insomnia Scale (AIS) score ≥ 6. After exclusion of patients with missing primary outcome data, patients were randomly divided in a 7:3 ratio into a training cohort ( n = 143) and an internal testing cohort ( n = 62). Candidate predictors available before the first postoperative night were selected using LASSO regression and entered into multivariable logistic regression to construct the prediction model and nomogram. Model performance was assessed using receiver operating characteristic (ROC) curves, calibration analysis, decision curve analysis (DCA), and bootstrap internal validation. Results First-night PSD occurred in 135 patients (65.9%). Multivariable analysis showed that a higher pre-sleep VAS score was independently associated with first-night PSD (OR 1.23, 95% CI 1.02–1.51, P = 0.039). Non-minimally invasive surgery (OR 2.11, 95% CI 0.99–4.49, P = 0.052) and higher preoperative AIS score (OR 1.13, 95% CI 1.00–1.31, P = 0.077) showed positive but statistically non-significant associations with first-night PSD. The final LASSO-selected model incorporated sex, surgical approach, preoperative AIS score, and pre-sleep VAS score; however, only pre-sleep VAS score reached conventional statistical significance in the multivariable model. The model demonstrated modest-to-moderate discriminative performance, with an apparent AUC of 0.703 in the training cohort, an optimism-corrected AUC of 0.674 after bootstrap validation, and an AUC of 0.699 in the internal testing cohort. Calibration analysis suggested acceptable agreement between predicted and observed risks, and DCA indicated potential clinical utility. Conclusions First-night PSD was common among adult surgical patients. Pre-sleep pain intensity was a dynamic, clinically modifiable factor before nocturnal sleep. The proposed pre-sleep prediction model showed modest-to-moderate discriminative performance after internal validation and may assist early risk stratification before the first postoperative night. Further multicenter studies incorporating objective sleep monitoring, longer follow-up, and external validation are needed.