A gradient-boosted trees algorithm using preoperative, surgical, and postoperative laboratory data predicted postoperative delirium in older patients with an AUC of 0.83.
Does a gradient-boosted trees algorithm accurately predict postoperative delirium in older surgical patients?
A machine learning algorithm combining preoperative factors, surgical characteristics, and postoperative lab changes can accurately predict postoperative delirium in older patients.
AbstractBackground Postoperative delirium (POD) affects ∼20% of older surgical patients. It is associated with poor clinical outcome and increased mortality. We aimed to identify the major POD risk factors and to develop and validate a multivariate algorithm for individual POD risk prediction and risk evaluation in the very early postoperative period. Methods BioCog is a prospective cohort study conducted in the anaesthesiology departments of two tertiary care centres in Germany and The Netherlands. Patients aged ≥65 yr with no preoperative dementia (Mini-Mental Status Examination ≥24) undergoing surgery with an expected duration of at least 60 min were enrolled and screened for POD according to DSM 5 until the seventh postoperative day. Clinical, neuropsychological, neuroimaging data, and blood were measured before and after surgery. We evaluated several models by sequentially adding blocks of variables. Gradient-boosted trees (GBT) with nested cross-validation were used for POD prediction. Model accuracy (area under the receiver-operating curve, AUC) and calibration were assessed (Brier score). Results Out of 929 patients, 184 (20%) experienced POD. A GBT algorithm using both preoperative data, characteristics of the intervention, and postoperative changes in laboratory parameters achieved the highest AUC (0.83, 0.79–0.86) with a Brier score of 0.12 (0.12–0.13). Conclusions Models combining preoperative with precipitating factors during surgery predict POD with high accuracy. This suggests that the resulting algorithms eventually may become useful to support clinical decision-making. Clinical trial registration NCT02265263.
Lammers-Lietz et al. (2026) studied this question. A gradient-boosted trees algorithm using preoperative, surgical, and postoperative laboratory data predicted postoperative delirium in older patients with an AUC of 0.83.