Stimulant Overdose Prediction Model for Medicaid-Insured Persons.

Importance: Overdoses involving methamphetamines and cocaine have increased in recent years. Identification of individuals at highest risk could facilitate the implementation of evidence-based interventions to reduce overdose risk. Objective: To develop and internally validate a model that predicts hospitalization or emergency department (ED) treatment for stimulant-involved overdose among the Medicaid-insured population. Design, Setting, and Participants: This was a retrospective case-cohort study using Medicaid claims data from 2016 to 2019 (development) and 2020 (validation) for all Medicaid enrollees age 15 years or older with a cocaine- or other stimulant-involved overdose. A subcohort was created using a simple random sample of the full cohort of all cases. Within the full cohort, cases were identified as those having any inpatient or ED encounter for stimulant-involved overdose during the following year. A case-cohort sample was obtained for each calendar year from 2016 to 2020, each with a subcohort size of 100 000. Each individual contributed only 1 case event (for an individual with multiple overdoses, only the first eligible was selected). For each of the 4 overdose outcomes, a predictive weighted Cox model was first developed among enrollees of sampling years 2016 to 2019 (development set), and its performance was evaluated in our test set of 2020. The prediction models were first developed in November 2023, and the model fairness assessment was performed in April to May 2025. Interventions or Exposures: Individual-level candidate predictors were demographic characteristics, enrollment, health care utilization, and other clinical variables. Area-level variables included social, economic, housing, and demographic characteristics data from the American Community Survey, rural-urban classification, Social Deprivation Index, retail opioid dispensing rates, and health resources. Main Outcomes and Measures: Four types of stimulant-involved overdose associated with hospitalization or ED treatment: cocaine-involved overdose, (1) involving an opioid or (2) not involving an opioid; or methamphetamine-, ecstasy-, or other psychostimulant-involved overdose (hereafter, other stimulant), (3) involving an opioid or (4) not involving an opioid. Results: The analysis included 78 795 enrollees with cocaine- and other stimulant-involved overdose (mean [SD] age, 42.2 [13.7] years; 33 304 [42%] female and 45 491 [58%] male individuals). Weighted Cox regression prediction models showed good calibration and high discriminatory performance (Harrell C statistic): cocaine-involved overdose, with (0.923) or without (0.902) an opioid; other stimulant-involved overdose, with (0.909) or without (0.868) an opioid. For cocaine-involved overdose with opioids, previous individual opioid use disorder diagnosis or cocaine use disorder diagnosis played the largest role in overdose risk prediction. For cocaine-involved overdose without opioids, previous cocaine use disorder diagnosis and area-level income inequality and housing variables contributed most to prediction. For other stimulant-involved overdose with opioids, previous opioid use disorder diagnosis and area-level percentage of those living with a disability contributed most to prediction. For other stimulant-involved overdoses without opioids, previous stimulant-related disorder and area-level proportion of individuals receiving Supplemental Nutrition Assistance Program contributed most to prediction. Conclusions and Relevance: This case-cohort study found that readily available data can be used to identify those at high risk of hospitalization or ED visit for cocaine- or stimulant-involved overdose. These individuals would likely benefit most from evidence-based interventions and awareness of risk factors for overdose.