Working Paper: NBER ID: w24947
Authors: Jill Horwitz; Corey S. Davis; Lynn S. McClelland; Rebecca S. Fordon; Ellen Meara
Abstract: States, which have the primary legal role in regulating the prescribing and dispensing of prescription medications, have created Prescription Drug Monitoring Programs (PDMP) to try to reduce inappropriate prescribing, dispensing, and related harm. Research assessing whether these interventions are effective has produced inconclusive and contradictory results. Here we examine whether different data sources may have contributed to the varying results. Specifically, we: 1) identify the decisions inherent in creating such a dataset; 2) discuss the public data sources used by researchers in previous work; 3) develop and apply a detailed research protocol to create a novel PDMP law dataset; and 4) to illustrate potential consequences of data choice, apply various data sources to analyze the relationship between PDMP laws and prescribing and dispensing of opioids among disabled Medicare beneficiaries. We find that our dates differ from those in existing datasets, sometimes by many years. The regression analyses generated a twofold difference in point estimates, as well as different signed estimates, depending on the data used. We conclude that the lack of transparency about data assembly in existing datasets, differences among dates by source, and the regression results raise concerns for PDMP researchers and policymakers.
Keywords: Prescription Drug Monitoring Programs; Opioid Regulation; Data Quality; Health Policy
JEL Codes: I11; I12; I18; K32; K42
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| PDMP laws (I18) | opioid prescribing patterns (I11) |
| differences in data sources (Y10) | variations in point estimates for opioid prescriptions (C46) |
| lack of transparency about data assembly methods (C80) | concerns for researchers and policymakers (J18) |
| inconsistent or poorly defined data sources (C80) | misrepresentation of effectiveness of PDMPs (C83) |