How increasing medical access to opioids contributes to the opioid epidemic: Evidence from Medicare Part D

Abstract

Drug overdoses involving opioid analgesics have increased dramatically since 1999, representing one of the United States' top public health crises. Opioids have legitimate medical functions, but they are often diverted, suggesting a tradeoff between improving medical access and nonmedical abuse. We provide causal estimates of the relationship between the medical opioid supply and drug overdoses using Medicare Part D as a differential shock to the geographic distribution of opioids. Our estimates imply that a 10% increase in opioid medical supply leads to a 7.1% increase in opioid-related deaths among the Medicare-ineligible population, suggesting substantial diversion from medical markets.

Keywords: Diversion; Opioid crisis; opioid supply.

Figure A.1:

Elderly Share in 2003

Figure A.2:

Relationship between % Elderly in 2003 and % Enrolled in Part D Notes: We regress the percentage of the population enrolled in Part D on the percentage of the 2003 population ages 65+. We perform this cross-sectional regression by year.

Figure A.3:

Pain Reliever Misuse Rate in 2004 by Age Group Source: 2004 National Survey on Drug Use and Health

Figure A.4:

Relationship between % Elderly in 2003 and Mortality Rate by Age Notes: We estimate equation (1) for each age between ages 1 and 85. The models include all covariates, including the policy variables. Confidence intervals are adjusted for within-state clustering.

Figure A.5:

Placebo Event Studies using TEDS Sources: Treatment Episode Data Set (2000–2011) Notes: Outcomes are defined as per 100,000 people. Each estimate refers to the effect of 2003 Elderly Share in that year. All specifications include controls for time and state fixed effects. We also include all controls used in Table 3, Column 4. Regressions are population-weighted. Estimates are normalized to 0 in 2003. 95% confidence intervals adjusted for clustering at the state level.

Figure A.6:

Event Studies for Overdoses not involving Prescription Opioids and Alcohol Poisoning Deaths Sources: National Vital Statistics System Notes: Mortality is defined as per 100,000 people. Each estimate refers to the effect of 2003 Elderly Share in that year. All specifications include controls for time and state fixed effects. We also include all controls used in Table 3, Column 4. Regressions are population-weighted. Estimates are normalized to 0 in 2003. Non-opioid overdoses exclude overdoses involving opioids and unspecified drugs. 95% confidence intervals adjusted for clustering at the state level.

Figure A.7:

Event Studies for Other Deaths of Despair Sources: National Vital Statistics System Notes: Mortality is defined as per 100,000 people. Each estimate refers to the effect of 2003 Elderly Share in that year. All specifications include controls for time and state fixed effects. We also include all controls used in Table 3, Column 4. Regressions are population-weighted. Estimates are normalized to 0 in 2003. 95% confidence intervals adjusted for clustering at the state level.

Figure A.8:

Mortality Estimates When Excluding One State Notes: We replicate our main mortality result while excluding one state at a time. Each estimate above is marked by the state that is excluded. 95% confidence intervals adjusted for clustering at the state level.

Figure A.9:

Event Study Estimates for Buprenorphine Distribution Sources: ARCOS Notes: The outcome is buprenorphine grams per capita. Each estimate refers to the effect of 2003 Elderly Share in that year. All specifications include controls for time and state fixed effects. We also include all controls used in Table 3, Column 4. Regressions are population-weighted. Estimates are normalized to 0 in 2003. 95% confidence intervals adjusted for clustering at the state level.

Figure 1:

Opioid Use and Abuse Notes: We use ARCOS data to generate per capita opioid distribution, NVSS to create per capita opioid-related mortality, and TEDS to calculate per capita substance abuse treatments for opiates. We normalize each time series to 100 in 2000.

Figure 2:

Opioid Distribution: Event Study Notes: We estimate equation (1) but allow the effect of Elderly Share in 2003 to vary by year, normalizing the coefficient for 2003 to zero. The outcome is morphine equivalent doses per capita. State and time fixed effects included. We also include all controls used in Table 2, Column 4. 95% confidence intervals adjusted for clustering at the state level.

Figure 3:

Main Event Study Estimates Sources: National Vital Statistics System (2000–2011) and Treatment Episode Data Set (2000–2011) Notes: Outcomes are specific to opioid-related mortality and opioid-related substance abuse treatments (per 100,000). Each estimate refers to the effect of 2003 Elderly Share in that year. All specifications include controls for time and state fixed effects. We also include all controls used in Table 3, Column 4. Regressions are population-weighted. Estimates are normalized to 0 in 2003. 95% confidence intervals adjusted for clustering at the state level.