Predicting psychiatric drug initiation after craniectomy: a retrospective cohort study

Psychiatric morbidity following craniectomy remains underexplored, with physical disability outcome measures forming the mainstay of follow-up tests used during rehabilitation. There are no predictive models to identify patients at a higher risk of new psychiatric drug initiation. We investigated new psychiatric drug initiation in a mixed-indication cohort following craniectomy and developed a risk stratification model. Our retrospective single-centre study identified 145 patients who underwent a supratentorial craniectomy between 2020 and 2025. Predictor variables were selected based on clinical plausibility, including age, Glasgow Coma Scale (GCS), pupillary response, American Society of Anaesthesiologists (ASA) grade, Index of Multiple Deprivation (IMD) rank, indication for surgery, laterality, and comorbidities. The primary outcome was new psychiatric medication initiation in patients without a pre-existing psychiatric diagnosis. Logistic regression and LASSO penalised modelling were performed. The model was internally validated using bootstrap optimism correction and regression coefficient shrinkage. Across the whole cohort, 33.8% of patients were prescribed psychiatric medication. Of those without a pre-existing psychiatric diagnosis, 18.6% of patients were commenced on new psychiatric medication. Whilst no independent predictors were significant in multivariate analyses, LASSO regression retained higher GCS (coefficient + 0.0585), fewer comorbidities (coefficient - 0.1749), increasing age (coefficient + 0.0109), and a lower ASA (coefficient - 0.0838) as non-zero coefficients under penalisation. Our final model offered limited discrimination (apparent AUC 0.645; Brier score 0.144) with bootstrap internal validation revealing a mean optimism of 0.065 and an optimism-corrected AUC of 0.580 (optimism-corrected Brier score 0.156). We present the first internally validated risk prediction model for new psychiatric drug initiation in a mixed indication craniectomy cohort. Model discrimination was limited, indicating that baseline clinical variables are insufficient to identify patients at increased risk. Prospective studies incorporating gold-standard neuropsychiatric testing are needed before risk stratification could facilitate earlier neuropsychiatric referrals and support more holistic rehabilitation care.