Developing and validating a machine-learning algorithm to predict opioid overdose in Medicaid beneficiaries in two US states: a prognostic modelling study

Changes in predicted opioid overdose risk over time in a state Medicaid program: a group‐based trajectory modeling analysis

“…Using 2013-16 Pennsylvania Medicaid claims, our group previously developed and validated a claim-based machine-learning algorithm to predict opioid overdose among Medicaid beneficiaries who filled opioid prescriptions [13]. Our machine-learning algorithm using a gradient-boosting machine achieved a concordance statistic (C-statistic) of more than 0.8 in predicting overdose in the subsequent 3 months after initiating prescription opioids [C-statistic = 0.841, 95% confidence interval (CI) = 0.834-0.848 for 2013-16…”

“…To our knowledge, the present work is the first to study longitudinal changes in opioid overdose risk. In contrast to many published opioid overdose prediction models that predicted one's risk in the next year or even longer periods [1,21], our machine-learning model is unique in that it predicts the overdose risk in the subsequent 3 months [13]. This has allowed us to evaluate the dynamic patterns of overdose risk quarterly.…”

Opportunities for opioid overdose prediction: building a population health approach

“…We commend Lo-Ciganic and colleagues for their model, which holds potential to prevent opioid overdose at the individual level. 3 Ending the opioid overdose epidemic is possible, and the shared toolkits of medicine, epidemiology, and data science have the potential to bend the curve through innovation and interdisciplinary engagement.…”

“…In a prognostic modelling study published in The Lancet Digital Health, Wei-Hsuan Lo-Ciganic and colleagues present the development and validation of a machinelearning algorithm to predict opioid overdose among Medicaid beneficiaries. 3 This study builds on a growing body of work at the intersection of medicine and data science-to which Lo-Ciganic and colleagues are leading contributors [4][5][6] -showing the utility of machine learning to inform clinical care for diverse patients with opioid use disorder or at risk for opioid overdose.…”

“…This portable tool for providers who treat Medicaid populations fills a pressing gap in the literature. 3 The authors notably predict patient opioid overdose risk with high predictive performance over a short-run period. Prediction across an abbreviated, 3-month risk period has the potential for strong clinical effect.…”