OBJECTIVES: To determine important risk factors for opioid overdose in a commercially insured population using integrated medical and pharmacy claims data, as well as publically available data published by the CDC and CMS. METHODS: Patients with at least one prescription opioid claim during January 2014 – August 2016 (n = 2,443,769) were selected from an administrative claims database of commercially insured members. Patients were followed on a monthly basis until the conclusion of the study period or until a lapse in eligibility. Patient characteristics including medical diagnosis groups, medical and utilization as well as use of multiple prescribers, and family member opioid use were captured. A binary logistic regression analysis was performed on a stratified random sample to estimate likelihood of overdose in the following month (n=60,867). Recognizing that integrated medical and pharmacy data may not be available, an additional binary logistic regression model was created without medical data. RESULTS: The model using integrated medical and pharmacy data included 16 variables, with prior opioid overdose, substance use and mental health related diagnosis, and utilization of drugs associated with overdose. History of prior opioid overdose (ICD9 965.X and ICD10 T40X) showed the strongest ability to predict overdose, with 25.1% of those overdosing having a prior overdose (OR 16.0 95% CI 8.8 – 28.9). Age was also a strong predictor, with 33.6% of those overdosing in the 15-25 age range (OR 2.5 95% CI 1.9 – 3.3). The model had a c-stat of 0.850 on the training data and 0.881 for the validation data. The model which was restricted to pharmaceutical data had a lower c-stat, with 0.817 for the training data and .839 for the validation data. CONCLUSIONS: Medical and pharmacy utilization patterns can identify members with a history of opioid use at an elevated risk of overdose and thus inform targeted behavioral health interventions.