Point-of-care electroencephalography for prediction of postoperative delirium in older adults undergoing elective surgery: protocol for a prospective cohort study.

Postoperative delirium (POD) is a complication of surgery in older adults associated with adverse outcomes. Current screening methods demonstrate poor interrater reliability, and conventional electroencephalography (EEG)-based screening requires intensive setup. Point-of-care (POC) EEG technology offers a rapid and objective alternative that may capture neurophysiological signatures of delirium risk. When combined with baseline and perioperative variables, POC EEG may enable the prediction of POD before clinical manifestation. In this study, we aim to develop a POD prediction model using POC EEG as well as explore secondary outcomes such as longer-term cognitive impairment and postoperative pain. This is a prospective cohort study enrolling older adults (≥60 years) undergoing elective non-cranial inpatient surgery at two academic hospitals. The target cohort size is 150 participants, determined by an events-per-parameter approach. All participants undergo baseline cognitive testing and pain assessment using the Montreal Cognitive Assessment (MoCA) and Numeric Rating Scale. The primary outcome is POD, while secondary outcomes include follow-up MoCA scores and postoperative pain scores. POD is assessed immediately after surgery and every 12 h during the admission with the 4AT tool. Perioperative EEG is acquired using the Ceribell EEG system (Ceribell, Inc.) across standardized preoperative, intraoperative, and postoperative phases. EEG features such as spectral power, alpha/delta ratio, and burst suppression ratio are analyzed in relation to outcomes. Predictive models will be developed using regularized logistic regression with nested feature sets, and model performance will be evaluated. This study evaluates whether POC EEG can accurately predict POD in older adults undergoing elective surgery, as well as longer-term cognitive impairment and postoperative pain. This approach could enable early identification of high-risk patients and facilitate targeted preventive strategies. By generating a validated risk model, multimodal exploratory analyses, and openly available datasets, this work aims to advance the practical management of perioperative outcomes.