Predicting the risk for postoperative delirium using routine EEG monitoring data

Postoperative delirium (POD) is a common complication in elderly surgical patients, causing prolonged hospitalization, cognitive decline, and increased institutionalization. Current prediction models often rely on demographic factors that lack sufficient granularity in age-homogeneous populations. We hypothesized that routine intraoperative EEG monitoring could detect underlying brain vulnerability patterns predictive of POD beyond clinical risk factors. We analyzed two cohorts of elderly patients from Charité Berlin: SuDoCo (n=1032, ≥60 years) and ePOD (n=263, ≥70 years), recorded 10 years apart with different anesthetic protocols. We developed a novel two-step burst suppression detection algorithm and extracted three EEG feature sets: burst suppression duration, power spectral densities, and signal covariances. Maintenance anesthetic choice created profound EEG differences that classifiers initially exploited as POD proxies rather than detecting genuine vulnerability. Training medication-specific models eliminated these pharmacological confounds. All three EEG features captured meaningful vulnerability patterns. The meta-classifier integrating all approaches achieved balanced accuracy of 0.691 (AUC: 0.759) on SuDoCo, with robust transfer to the demographically distinct and homogeneous ePOD cohort. Our findings demonstrate that EEG-based vulnerability markers add crucial discriminative power when demographic predictors fail, particularly in homogeneous elderly populations where age and clinical scores lack sufficient resolution.

Author dc:creator

• Röhr, Vera

Advisor dc:contributor.advisor

• Blankertz, Benjamin

Licence dc:rights.uri

• https://creativecommons.org/licenses/by/4.0/

Language dc:language.iso

en

Identifier URI

https://doi.org/10.14279/depositonce-26065

OAI identifier oai:identifier

oai:depositonce.tu-berlin.de:11303/27230