消化器手術患者における術後せん妄予測モデルの開発と検証

Postoperative delirium (POD) is a critical and prevalent postoperative complication challenging to major surgeries, associated with the increased morbidity and mortality rate, prolonged hospitalization, and higher healthcare expenditure. Despite the availability of POD prediction tools, none of the models deals particularly with the unique preoperative risks of gastrointestinal surgery patients, which is the critical gap in the present clinical practice. Thus, the objectives of the study were the development and validation of an easy-to-use preoperative prediction model for POD specifically for patients undergoing gastrointestinal surgery, to narrow this gap by a specific population tool.

A total of 369 patients undergoing gastrointestinal surgery were analyzed retrospectively. Univariate analysis was used to identify candidate variables, and a multivariate logistic regression model was built using with statistically significant predictors (P<0.05). The area under the receiver operating characteristic curve (AUC) was used as a measure of model performance. Internal validation was done though 20 times 5-fold cross-validation of the models as a measure of robustness.

Univariate analysis identified four statistically significant factors for POD: advanced age (OR = 4.19, 95% CI: 1.68-11.91), lower educational level (OR = 0.25, 95% CI: 0.10-0.60), history of cerebrovascular disease (OR = 6.59, 95% CI: 2.55-16.47), and elevated preoperative serum glucose level (OR = 2.68, 95% CI: 1.03-6.54). All these factors were included in the final POD prediction model, which proved to have strong discriminative power (AUC = 0.810, 95% CI: 0.700-0.921) and good calibration (Hosmer-Lemeshow test, P = 0.436).

This model is the first easy-to-use preoperative prediction model of POD particularly developed for gastrointestinal surgical patients and exhibits high discriminatory accuracy. It enables anesthesiologists to promptly stratify the risk of POD, which will allow them to implement preventive interventions in time.