Answers are grounded in this paper and cite their sources.
Background Postoperative delirium is a serious complication after surgery. This study evaluates prediction from preoperative hospital records and a separate exploratory ICU classification task. Objective To evaluate a shared evolutionary plus surrogate-gradient spiking neural network (E-SNN) for predicting physician-diagnosed delirium within the first postoperative week after orthopedic surgery, with MIMIC-III retained as a distinct exploratory ICU analysis. Methods The hospital cohort included 577 orthopedic surgery patients, including 84 with physician-documented delirium, evaluated across three stratified splits. MIMIC-III was analyzed separately as an exploratory ICU classification dataset. Both datasets used training-only preprocessing, LASSO feature selection, minority oversampling, and the same E-SNN and baseline pipeline. Results Across three hospital splits, the E-SNN achieved mean AUROC 0.629 (SD 0.075) and average precision 0.278 (SD 0.140); mean positive-class F1 at the common threshold of 0.50 was 0.279 (SD 0.167). Histogram gradient boosting performed better on AUROC and average precision. MIMIC-III findings were exploratory and do not constitute external validation for postoperative delirium prediction. Conclusion The E-SNN showed modest, dataset-specific performance and did not outperform histogram gradient boosting on the main discrimination measures. The MIMIC-III analysis should be interpreted as exploratory ICU classification, not external validation of postoperative delirium prediction.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: