Linked databases could help identify those at risk for opioid overdoses

Linked administrative databases accurately predicted fatal and nonfatal opioid overdose, according to study results published in JAMA Psychiatry.

“Many people who overdose from opioids have risk factors that can be used to accurately identify who they are and to get them better access to services, such as treatment,” Brendan Saloner, PhD, associate professor in the department of health policy and management at Johns Hopkins Bloomberg School of Public Health told Healio Psychiatry. “We hoped to show that it is possible to develop a model that can be used to find people who have these risk factors. There is a real opportunity to improve public health interventions by merging together multiple data sources that are already being collected through hospital records, criminal justice, behavioral health treatment and controlled substance prescriptions.”

Current predictive risk models often exclude individuals without insurance, individuals with criminal justice involvement and patients who use specialty behavioral health programs derived from health plans. Because overdose risk has largely shifted from those using prescription opioids to those using illicit opioids, prediction risk models may need to incorporate data on indicators linked to illicit substance use to effectively identify populations at high risk, Saloner and colleagues wrote.

Brendan Saloner

The investigators sought to create a predictive risk model to identify opioid overdose using linked criminal justice and clinical data. They used 2015 data from four Maryland databases — criminal justice records for drug- or property-related offenses, the prescription drug monitoring program, public-sector specialty behavioral treatment and all-payer hospital discharges — to create a cross-sectional sample. The researchers included adults living in Maryland who were aged 18 to 80 years with records in any of the four databases and excluded those who had a non-Maryland ZIP code or died in 2015. They separately estimated logistic regression models for risk for fatal and nonfatal opioid overdose in 2016 and assessed model performance using bootstrapping.

Criminal justice, specialty behavioral health and hospital encounters, as well as controlled substance prescription fills, served as exposures. Main outcomes and measures included fatal opioid overdoses defined by the state medical examiner and one or more nonfatal overdoses treated in Maryland hospital during 2016.

Among 2,294,707 individuals included in the sample, 42.3% were men and 53% were younger than 50 years. For 2016, results showed fatal opioid overdoses among 1,204 individuals (0.05%), as well as nonfatal overdoses among 8,430 individuals (0.37%). Adjusted analysis revealed the following factors as most strongly associated with fatal overdose:

• male sex (OR = 2.4; 95% CI, 2.08-2.76);

• diagnosis of opioid use disorder in a hospital (OR = 2.93; 95% CI, 2.17-3.8);

• release from prison in 2015 (OR = 4.23; 95% CI, 2.1-7.11); and

• receiving opioid addiction treatment with medication (OR = 2.81; 95% CI, 2.2-3.86).

Saloner and colleagues noted similar associations for nonfatal overdose. They reported the area under the curve for fatal overdose was 0.82 for a model with hospital variables, 0.86 for a model with both hospital and prescription drug monitoring program variables and 0.89 for a model that also added criminal justice and behavioral health variables. The AUC using all variables was 0.85 for nonfatal overdose.

“Although we do not necessarily think our model can be implemented for clinical interventions in a health system, we do think that it can be used by public health agencies to prioritize resources, such as street-based outreach, increased investments in treatment and distribution of naloxone,” Saloner told Healio Psychiatry.