Abstract
BACKGROUND: People with opioid use disorder (OUD) have substantially higher standardised mortality rates compared with the general population. However, lack of individualised prognostic information presents challenges in personalisation of addiction treatment delivery. AIMS: To develop and validate the first prognostic models to estimate 6-month all-cause and drug-related mortality risk for people diagnosed with OUD using indicators recorded at baseline assessment in addiction services in England. METHOD: Thirteen candidate prognostic variables, including sociodemographic, injecting status and health and mental health factors, were identified from nationally linked addiction treatment, hospital admission and death records from 1 April 2013 to 1 April 2022. Multivariable Cox regression models were developed with a fractional polynomial approach for continuous variables, and missing data were addressed using multiple imputation by chained equations. Validation was undertaken using bootstrapping methods. Discrimination was assessed using Harrel's C and D statistics alongside examination of observed-to-predicted event rates and calibration curve slopes. RESULTS: Data were available for 236 064 people with OUD, with 2427 deaths due to any cause, including 1289 due to drug-related causes. Both final models demonstrated good optimism-adjusted discrimination and calibration, with all-cause and drug-related models, respectively, demonstrating Harrell's C statistics of 0.73 (95% CI 0.71-0.75) and 0.74 (95% CI 0.72-0.76), D-statistics of 1.01 (95% CI 0.95-1.08) and 1.07 (95% CI 0.98-1.16) and calibration slopes of 1.01 (95% CI 0.95-1.08) and 1.01 (95% CI 0.94-1.10). CONCLUSIONS: We developed and internally validated Roberts' OUD mortality risk, with the first models to accurately quantify individualised absolute 6-month mortality risks in people with OUD presenting to addiction services. Independent validation is warranted to ensure these models have the optimal utility to assist wider future policy, commissioning and clinical decision-making.