National Demographic Trends in Hospitalizations for Heroin- and Opioid-Related Overdoses, 1993-2009

Background: The rapid growth in the abuse of prescription opiates (PO) has dramatically increased PO overdoses (POD).  Currently, POs are the most likely cause of death for 35 -54 year olds.  Public and political awareness of PO misuse has increased law enforcement supply-based interdiction efforts as well as changes to the delivery technologies for drugs such as oxycontin, making them harder to misuse. While these efforts may reduce harm associated with PO misuse, addicted individuals may switch to other opiates such as heroin.  In this analysis, we test the hypothesis that changes in the rates of POD are correlated with changes in the rate of heroin overdoses (HOD).

Methods: Data come from the Nationwide Inpatient Sample (NIS), a nationally representative sample of all-payer hospital data for years 1993 to 2009.  ICD-9 codes for POD and HOD were used to code overdose events.  Census data and NIS data were used to construct rates of POD and HOD overall and by race, gender, and age.  Regression methods were used to test for linear trends in rates within demographic groups over the 17 years of the study period.  Two lagged negative binominal regression models were used to predict changes in a hospital’s count of HOD and POD.  Model 1 used year t’s count of hospital POD to predict year t+1’s count of HOD, controlling for year t’s count of HOD and hospital urbanicity. Model 2 used year t’s count of hospital HOD to predict year t+1’s count of POD, controlling for year t’s count of POD and hospital urbanicity.

Results:  Regression analysis of overdose rates show that whites, women, and middle aged individuals had the highest increase in both POD and HOD over the study period. In lagged model 1, each increase in a hospital’s year t count of POD increases the hospital’s year t+1 rate of HOD by a factor of 1.26 (p<0.001), after adjusting for year t’s HOD and the hospitals urbanicity. A standard deviation increase in year t's PODs, roughly 7 overdoses per hospital, increases the mean expected count of HODs for a hospital by 193%, a nearly three-fold increase in HOD.  In lagged model 2, predicting POD, each increase in a hospital’s year t HOD results in an increase in year t+1’s rate of POD by a factor of 1.57 (p<0.001). A one standard deviation in year t’s HODs, approximately 4 overdoses per hospital, results in an expected mean increase of a hospital’s t+1 rate of POD by 205.6%.

Conclusion: Our hypothesis, that there is fungibility between prescription opiates and heroin, was confirmed by these analyses. Our finding suggests that increases in types of opiate use are strongly associated with each other. These findings suggest that focusing PO abuse prevention efforts on supply-based interventions may simply lead to a shift in use to heroin rather than reduced harm.  The alternative approach of using drug abuse prevention resources on treatment and demand-side reduction is likely to be more productive at reducing opiate abuse related harm.