Background: “Before medically advised” (BMA) hospital discharge is associated with higher risks for subsequent death and drug overdose. We sought to develop prediction models to estimate the absolute risk of death from any cause and illicit drug overdose after BMA discharge for a given patient.

Methods: We used retrospective population-based administrative health data of a 20% random sample of all British Columbia residents to derive and internally validate regression models to estimate the risks of death (model A) and illicit drug overdose (model B) within 30 days after BMA discharge. Model A included all nonelective, nonobstetrical adult hospitalizations ending in BMA discharge between 2015 and 2019. Model B included only hospitalizations from model A for patients with evidence of prior substance use. We fitted prediction models using logistic regression and validated models using bootstrap-based optimism correction. Candidate predictors included sociodemographic and clinical characteristics available to clinicians during BMA discharge.

Results: Of 6440 hospital admissions included in model A, 102 (1.6%) were associated with the death of a patient within 30 days of BMA discharge. Predictors for death included a Charlson Comorbidity Index of 2 or higher, cancer, and heart disease. Model A exhibited good discrimination (C-statistic = 0.78) and excellent calibration. Of 4466 hospital admissions included in model B, 233 (5.2%) were associated with a patient subsequently overdosing within 30 days of BMA discharge. Predictors for drug overdose included homelessness, receipt of social income assistance, opioid use disorder, non-alcohol substance use disorder, overdose in the past year, and discharge from a surgical service. Model B exhibited good discrimination (C-statistic = 0.79) and excellent calibration.

Interpretation: Risk prediction models may help clinicians and hospitals identify patients who might benefit from intensive support to prevent death or illict drug overdose after BMA discharge.