Changes in Opioid Prescribing Following the Implementation of State Policies Limiting Morphine Equivalent Daily Dose in a Commercially Insured Population

Abstract

Background:

Prescription opioid mortality doubled 2002–2016 in the United States. Given the association between high-dose opioid prescribing and opioid mortality, several states have enacted Morphine Equivalent Daily Dose (MEDD) policies to limit high-dose prescribing. The study objective is to evaluate the impact of state-level MEDD policies on opioid prescribing among the privately insured.

Methods:

Claims data, 2010–2015 from nine policy states and two control states and a comparative interrupted time series design were utilized. Primary outcomes were any monthly opioid use and average monthly MEDD. Stratified analyses evaluated theorized weaker policies (guidelines) and theorized stronger policies (passive alert systems, legislative acts, and rules/regulations) separately. Patient groups explicitly excluded from policies (e.g., individuals with cancer diagnoses or receiving hospice care) were also examined separately. Analyses adjusted for covariates, state fixed-effects, and time trends.

Results:

Both guideline and strong policy implementation were associated with 15% lower odds of any opioid use, relative to control states. However, there was no statistically significant change in the use of high dose opioids in policy states relative to control states. There was also no difference in direction and significance of the relationship among targeted patient groups.

Conclusions:

MEDD policies were associated with decreased use of any opioids relative to control states, but no change in high-dose prescribing was observed. While the overall policy environment in treatment states may have discouraged opioid prescribing, there was no evidence of MEDD policy impact, specifically. Further research is needed to understand the mechanisms through which MEDD policies may influence prescribing behavior.

Introduction

Prescription opioids are commonly used for both chronic and acute pain, but have potential for misuse. Previous research has established that high-dose opioid prescribing is a major risk factor for overdose.^1–3 Given this association, a commonly promoted tool is the establishment of Morphine Equivalent Daily Dose (MEDD) threshold policies. MEDD is a measurement that standardizes opioid prescriptions to an equivalent dose in morphine and divides the total prescription by days supply (the number of days the prescription is intended to last).⁴ MEDD allows comparison among different opioid formulations and strengths. MEDD threshold policies set an overall dose over which prescribing is discouraged in some way, though the threshold level and type of policy varies widely by the states and organizations that promote them. These types of policies are typically not implemented in isolation, but rather as part of a comprehensive strategy on opioid prescription reform. From January 1, 2007 to June 1, 2016, 31 MEDD policies have been enacted by 22 states.⁵

Despite the proliferation of MEDD policies, evaluations have been limited. Evaluations of Washington’s MEDD threshold guideline in the Medicaid population found a reduction in opioid use from pre- to post- guideline implementation with the greatest reductions occurring in the proportion of patients receiving over 120 MEDD, (the threshold level set by the guideline).^6,7 The studies did not make use of comparison states, but did note that the reduction was seen at a time in which opioid prescribing was increasing in the United States, overall. A recent study of workers’ compensation MEDD guidelines found that states that implemented MEDD guidelines saw reduced average MEDD as compared to states that did not implement such guidelines over the same timeframe.⁸ To our knowledge, no studies have evaluated MEDD threshold policies in the private insurance population. It is unknown whether policies for the general population will have the same effects as those targeted at Medicaid recipients or Workers’ Compensation claimants, where payers may have more of an influence on prescribing practices. Furthermore, the impact of MEDD policies other than guidelines on prescribed dose have not been studied, though research in other contexts suggests that they may be more impactful than guidelines. For example, studies have generally found that adherence to prescribing guidelines is low, even years after a guideline has been published.^9–12 By contrast, passive alert systems which provide notifications to prescribers through Electronic Health Record (EHR) embedded decision support, e-mail, or physical letters significantly improved provider adherence to prescribing guidelines.^9–11,13,14 Similarly, prescriber mandates may be more effective than guidelines at influencing prescriber behavior. In studies of state prescription drug monitoring programs (PDMPs), mandated use laws reduced certain high-risk practices such as prescribing overlapping opioid prescriptions.^15,16 The goal of the present study is to evaluate the impact of nine states’ MEDD threshold policies on the MEDD of filled prescriptions among individuals with private insurance. These policies include state-level guidelines, rules/regulations, legislative acts, and passive alert systems that are targeted towards the general population (i.e., without any specific type of insurance coverage). Based on prior literature, we hypothesize that all policies will be associated with reductions in prescribed dose and that the reductions associated with rules/regulations, legislative acts, and passive alerts will be greater than the reductions associated with guidelines.

Methods

Policies

MEDD policies and characteristics were obtained from a prior systematic policy mapping study.⁵ Policy types include guidelines, rules/regulations, legislative acts, and passive alert systems, with guidelines theorized to be less impactful than other types of policies. The effective dates, threshold levels, policy types, sponsoring state organizations, and populations specifically excluded from the policy are detailed in Table 1. Control states were selected on the basis of not implementing Prescription Drug Monitoring Programs (PDMPs),¹⁹ pain clinic laws,²⁰ state-level Medicaid or workers’ compensation MEDD policies,⁵ or passing any major prescription opioid legislation (as determined through an author search of LexisNexis and Westlaw Next) during the study period and parallel trends in average monthly MEDD prior to implementation of the first policy (January 2012).

Table 1.

Effective dates, thresholds, policy types, and excluded patient groups in treatment and control states

State	Effective Month/Year	Threshold	Policy Type	Sponsoring State Organization(s)	Excluded patient groups
AZ	03/2014	120	Legislative act	State Legislature	Inpatients
IN	11/2014	60	Rule/Regulation	Medical Board	Acute pain, terminal/hospice/palliative, long-term care facility, short courses
RI	03/2015	120	Rule/Regulation	Health Department	Acute pain
TN	09/2014	120	Passive alert system	Prescription Drug Monitoring Program	Acute pain, terminal/hospice/palliative, inpatient, emergency room
WA	01/2012	120	Legislative act	State Legislature	Acute pain
AZ	11/2014	100	Guideline	Health Department	Acute pain and terminal/hospice/palliative patients, short courses
CA	11/2014	80	Guideline	Medical Board	Acute pain, terminal/hospice/palliative
CO	07/2014	120	Guideline	Medical Board	Cancer, terminal/hospice/palliative
OH	05/2013	80	Guideline	Medical Board	Acute pain, terminal/hospice/palliative, short courses
SC	11/2014	80	Guideline	Medical, Dentistry, and Nursing Boards	None
WA	04/2007	120	Guideline	Washington Agency Medical Directors Group (collaboration of multiple state agencies)	Acute pain, cancer
VA	N/A	N/A	None (Control)	N/A	N/A
WI	N/A	N/A	None (Control)	N/A	N/A

Policy variables were defined in two ways: First, as a simple pre- and post- indicator for whether or not the policy was in effect at the time. Second, a months since policy implementation variable was developed to allow for gradual policy dissemination over time. Policy variables were also defined as dichotomous variables (any policy and no policy) and as categorical variables based on theorized impact (guideline, strong policy, and no policy). Strong policies included rules/regulations, legislative acts, and passive alert systems. Except for two states, states only implemented one type of policy. Arizona implemented a strong policy prior to implementing a guideline. In analyses, the strong policy takes precedence over the guideline, so individuals in Arizona were considered to be in a “strong policy” state at every time point after the strong policy was implemented. Washington implemented a guideline prior to the study period (2007) and implemented a strong policy during the study period (2012). Individuals in Washington were considered to be living in a “guideline” state until 2012, after which they were considered to be living in a “strong policy” state.

Data

Truven Health’s MarketScan database (hereafter MarketScan) was used to evaluate state-level policies implemented between January 1, 2010 and December 31, 2015. This dataset consists of commercial claims from 350 private payers and include individuals in the United States with employer-sponsored health insurance.¹⁷ The data contain inpatient and outpatient International Classification of Disease, Version 9 Clinical modification (ICD-9-CM) codes, discharge codes, outpatient revenue codes, facility codes, National Drug Codes (NDC), quantity of drug, days supply of drug, age, sex, and state of enrollee residence. MarketScan commercial claims data have been previously used to conduct research on opioid utilization.¹⁸

Population

To determine whether passage of policies was associated with change in odds of receiving any opioids, the main analyses were conducted on a random sample of one million enrollees age <65 from treatment and control states (Table 1). Individuals ≥65 were excluded, as they may also have unobserved Medicare claims for opioid prescriptions.

Additional analyses were conducted on included enrollees with at least one valid, active opioid prescription between January 1, 2010 and December 31, 2015.

Populations for Stratified Analyses

Several indicators were created to identify patient populations or prescriptions that were explicitly excluded from some policies and these indicators were used in subsequent stratified analyses. These previously identified⁵ exclusion groups were: patients with cancer, patients with acute pain, terminal/hospice/palliative patients, inpatients, and short courses of opioids (Table 1). Patients were defined as acute pain or cancer if they had at least one relevant inpatient or outpatient ICD-9-CM code during the study period as defined by Mack et al.²¹ Patients with any inpatient claims at any point during the study period were defined as inpatients, patients with any MarketScan hospice facility codes, hospice outpatient revenue codes, or inpatient discharge status to hospice during the study period were defined as terminal/hospice/palliative care. Patients with any MarketScan emergency room or long-term care outpatient revenue or facility codes during the study period were defined as emergency room and long-term care facility patients, respectively. Any months in which there were no opioid prescriptions in the prior two months were considered short-course opioids, consistent with the policy definitions of short courses as less than three months.⁵

Among patients with any opioid use prior to January 2012 (the first policy implementation date), in sensitivity analyses the population was also stratified by those with and without high baseline opioid use—defined as four indicator variables corresponding to the four MEDD policy thresholds: 60, 80, 100, and 120 MEDD—in at least one month prior to January 2012 (the first policy implementation date). This was to determine whether there were differential policy impacts among patients with and without a history of high dose opioid use.

Outcomes

All outcomes were related to receipt or MEDD of valid, filled opioid prescriptions. Both receipt of any opioids in a given month of enrollment and average monthly MEDD among those who received opioids were tested as outcomes to distinguish between potential effects of the MEDD threshold levels, specifically, and other co-occurring policy features.

Valid opioid prescriptions were defined as having non-missing quantities and days supply. Quantities with values of 0 or >1000 and days supply 0 or >180 were considered missing, consistent with prior studies.²¹ Due to reporting discrepancies, some quantities were misreported and a data cleaning protocol developed with input from the data vendor was applied (Supplementary File 1). Duplicate values on NDC, units, and fill date were deleted. Opioid prescriptions (excluding buprenorphine) and morphine equivalent conversion factors were identified using a Centers for Disease Control crosswalk file.²² Among the sample of one million enrollees, three outcomes were tested: Indicators for any opioid use, opioid use >60 MEDD, and opioid use >120 MEDD during a month in which the individual was enrolled. MEDD was calculated at the person-month level by multiplying quantity, dose, and conversion factor and dividing by days supply, accounting for multiple and overlapping prescriptions (Supplementary File 2).

Among enrollees with any opioid use, the primary outcome of interest was monthly MEDD. As the distribution of MEDD is highly skewed, a log transformed MEDD outcome was also tested. MEDD dichotomized at the 60, 80, 100, and 120 MEDD threshold levels were tested, corresponding to policy thresholds (Table 1).

Analyses

For all analyses, comparative interrupted time series—a special case of difference in differences—were used to evaluate the relationship between policy implementation and opioid prescribing. The unit of analysis was person-month and controls for age, sex, and time since the first opioid prescription filled during the study period (to account for within-subject MEDD changes), state fixed effects, a linear time trend, and clustering at the individual and state level, (to account for correlated outcomes within states and individuals), were included.

First, generalized linear mixed models with a logit link were used with the dichotomous outcomes of any opioid use, MEDD>60 and MEDD>120 in the random sample of one million enrollees using dichotomous guideline and strong policy indicators interacted with time.

Among enrollees with any opioid use, generalized linear mixed models were used with MEDD as the primary outcome and person-month as the unit of analysis. Models included an additional control for months since first opioid prescription and linear time trends. Policy variables tested included both dichotomous and months in effect variables as described in the previous section. All time variables, including months since first opioid prescription, months since policy implementation, and the monthly linear time trends were tested for multicollinearity, defined as variance inflation factors (VIF) >10. All analyses were conducted with SAS 9.4 statistical software (Cary, NC).

Sensitivity Analyses

Great care was taken to examine the opioid policy environment in the states and years selected for inclusion into the analysis. The greatest historical threat to validity identified was in Washington, which implemented a PDMP within a month of the passage of its MEDD threshold legislative act. Therefore, analyses were conducted with and without Washington. All other PDMPs in control and treatment states became active outside of the study period. Among opioid users, the MEDD outcome was tested as continuous, log transformed, and dichotomous (using 60, 80, 100, 120 MEDD cutoffs) and policy variables were also tested as dichotomous and months in effect to allow for the possibility of an implementation lag. Analyses were also stratified by the previously mentioned exclusions groups (e.g., patients with and without malignant pain were examined separately) with the hypothesis that larger decreases in MEDD would be observed in the groups not excluded from the policies while excluded groups would see little to no change, relative to the control group. This was done both in analyses of all opioid users, and state-specific analyses with just the relevant groups for each state excluded. Models were also stratified by high baseline dose, defined as at least one month with a MEDD above each of the four designated thresholds prior to 2012. It was hypothesized that larger decreases in MEDD would be seen among individuals with high baseline use as compared to those without high baseline use.

Results

Baseline—defined as months prior to January 2012—demographic, enrollment, and opioid use characteristics of a random sample of enrollees by policy state (control, guideline, or strong policy) are presented in Table 2. Of the one million individuals in the sample, 719,568 individuals were enrolled prior to 2012 and are included in the baseline table. Demographic and enrollment characteristics were similar among all three groups with females comprising slightly more than half of enrollees and mean age ranging from 30.67 (strong policy states) to 32.62 (control states). However, individuals in the strong policy states had higher opioid use at baseline with 19.29% of enrollees filling at least one opioid prescription prior to 2012 in strong policy states as compared to 15.89% in control states and 15.43% in guideline states. Similarly, enrollees in strong policy states were more likely to have filled at least one prescription >60 MEDD or >120 MEDD than enrollees in control or guideline states.

Table 2.

Baseline characteristics of all users by policy type (Random sample of enrollees, months prior to January 2012, N=719,568 enrollees)

	Control Statesa (N=193,416)	Guideline Statesb (N=363,177)	Strong Policy Statesc (N=162,975)
Male, N(%)	96,058 (49.66)	178,359 (49.11)	80,056 (49.12)
Age, Mean (SD)	32.62 (18.27)	31.36 (18.30)	30.67 (18.46)
Any opioid use, N(%)	30,740 (15.89)	56,044 (15.43)	31,432 (19.29)
≥1 month >60 MEDD	7,698 (3.98)	14,324 (3.94)	9,058 (5.56)
≥1 month >120 MEDD	1,978 (1.02)	3,703 (1.02)	2,328 (1.43)
Months enrolled, Mean (SD)	19.88 (12.07)	20.93 (12.39)	19.80 (12.18)

^aVirginia and Wisconsin

^bCalifornia, Colorado, Ohio, and South Carolina

^cArizona, Indiana, Rhode Island, Tennessee, Washington. Strong policies include legislative acts, rules/regulations, and passive alert systems. Arizona and Washington implemented both guidelines and strong policies, but are classified as strong policy states only for descriptive purposes.

Among individuals with at least one opioid prescription during the study period, the final population for analysis was 27,391,637 person-months representing 7,030,785 individuals. Of these individuals, 4,961,599 had at least one opioid prescription filled prior to January 2012. Baseline characteristics of these users by policy group are presented in Table 3. In general, opioid users, regardless of policy group, were older and more likely to be female than the overall population of enrollees. As with the overall population of enrollees, baseline demographic characteristics of opioid users did not differ greatly by policy state. Opioid users in strong policy states were more likely than opioid users in control or guideline states to have high dose use defined as >60 MEDD (28.86% in strong policy states vs. 25.46% in control states and 25.28% in guideline states) and >120 MEDD (7.48% in strong policy states vs. 6.44% in guideline states and 6.45% in control states). In all policy states, most opioid users had at least one exclusion—defined as an acute pain or cancer diagnosis, or hospice, inpatient, long-term care, or emergency department use—during the study period, but opioid users in strong policy states were less likely to have one of these exclusions (56.96%) as compared to opioid users in guideline (58.62%) or control states (62.07%).

Table 3.

Baseline characteristics of opioid users by policy type (Individuals using opioids prior to January 2012, N=4,961,599 enrollees)

	Control Statesa (N=1,287,321)	Guideline Statesb (N=2,356,122)	Strong Policy Statesc (N=1,318,303)
Male, N(%)	575,690 (44.72)	1,034,102 (43.89)	575,835 (43.68)
Age, Mean (SD)	40.30 (15.52)	40.22 (15.44)	39.94 (15.80)
≥1 month >60 MEDD, N(%)	327,737 (25.46)	595,474 (25.28)	380,447 (28.86)
≥1 month >120 MEDD, N(%)	82,971 (6.45)	151,763 (6.44)	98,549 (7.48)
Months with ≥1 opioid, Mean (SD)	3.29 (5.48)	3.55 (5.95)	3.66 (5.96)
Acute pain diagnosis, N(%)	448,458 (34.84)	749,220 (31.80)	394,538 (29.93)
Cancer diagnosis, N(%)	206,204 (16.02)	355,930 (15.11)	187,757 (14.24)
Hospice, N(%)	4,058 (0.32)	5,712 (0.24)	3,883 (0.29)
Inpatient visit, N(%)	197,693 (15.36)	354,727 (15.06)	191,035 (14.49)
Long-term care, N(%)	5,852 (0.45)	8,255 (0.35)	3,777 (0.29)
ED visit, N(%)	388,058 (30.14)	616,931 (26.19)	350,476 (26.59)
Any person-level exclusion, N(%)	799,066 (62.07)	1,381,147 (58.62)	750,952 (56.96)

^aVirginia and Wisconsin

^bCalifornia, Colorado, Ohio, and South Carolina

^cArizona, Indiana, Rhode Island, Tennessee, Washington. Strong policies include legislative acts, rules/regulations, and passive alert systems. Arizona and Washington implemented both guidelines and strong policies, but are classified as strong policy states only for descriptive purposes.

Abbreviations: ED, Emergency Department; MEDD, Morphine Equivalent Daily Dose

The unadjusted proportion of enrollees receiving any opioid, opioids >60 MEDD, and opioids >120 MEDD decreased across all three policy groups over time (Figure 1). At baseline, a higher proportion of enrollees in strong policy states received any opioids and high dose opioids than did enrollees in guideline and control states, and the proportion of enrollees with any opioid use appears to have decreased more rapidly in the strong policy states than in the guideline or control states.

Figure 1.

Percent of random sample of enrollees with any opioid use, Morphine Equivalent Daily Dose (MEDD) >120, and MEDD>60 in MEDD guideline states (Ohio (OH), South Carolina (SC), California (CA)) and strong policy (rules/regulations, legislative acts, and passive alert systems) states (Washington (WA), Tennessee (TN), Indiana (IN), Arizona (AZ), and Rhode Island (RI)), vertical lines indicate policy enactment dates (N=1,000,000 people; 25,628,772 person-months)

Regression results using the dichotomous outcomes of any opioid use, MEDD>60, and MEDD>120 on a random sample of one million enrollees are presented in Table 4. Passage of both guidelines and strong policies were associated with lower odds of any opioid use relative to control states. Guideline passage was associated with 15% lower odds of any opioid use (95% CI: 18% lower to 13% lower, p<0.001) and strong policy passage was associated with 15% lower odds of any opioid use (18% lower to 12% lower, p<0.001) relative to control states. However, neither guideline passage nor strong policy passage was associated with a statistically significant change in odds of high dose use relative to control states.

Table 4.

Regression results, random sample of 1,000,000 enrollees (N=25,628,772 person-months)^a

	Any opioid use			MEDD>60			MEDD>120
	OR	95% CI	p-value	OR	95% CI	p-value	OR	95% CI	p-value
Guidelineb	0.845	(0.822, 0.868)	<0.001	0.946	(0.894, 1.006)	0.078	0.984	(0.893, 1.084)	0.741
Strong Policyc	0.847	(0.818, 0.877)	<0.001	1.034	(0.962, 1.112)	0.367	1.033	(0.909, 1.174)	0.623
Age	1.040	(1.039, 1.040)	<0.001	1.042	(1.040, 1.044)	<0.001	1.043	(1.040, 1.046)	<0.001
Male	0.848	(0.827, 0.870)	<0.001	0.916	(0.870, 0.963)	<0.001	0.944	(0.864, 1.032)	0.208
Monthsd	0.994	(0.994, 0.994)	<0.001	0.991	(0.990, 0.992)	<0.001	0.991	(0.989, 0.992)	<0.001

^aIncludes state fixed effects and clustering at the individual and state level, unit of analysis is person-month

^bCalifornia, Colorado, Ohio, South Carolina, and Washington (pre-2012)

^cArizona, Indiana, Rhode Island, Tennessee, Washington (2012–2015). Strong policies include legislative acts, rules/regulations, and passive alert systems. Arizona implemented a guideline following a strong policy, but is considered a strong policy state at every time point after the strong policy was implemented. Washington implemented a guideline prior to the study period (2007) and implemented a strong policy during the study period (2012). Washington was considered a guideline state until 2012, after which they were considered a strong policy state.

^dCentered at January, 2012

Several additional regression analyses were conducted among opioid users in treatment and control states. These included analyses using continuous, log transformed, and dichotomous MEDD outcomes and policy variables that allowed for lagged dissemination, and excluded Washington State. Results of these regressions are presented in Tables 1–8 of Supplementary File 3. Among opioid users, policy passage was associated with a small but statistically significant increase in MEDD (3.72 mg MEDD, 95% CI: 3.51, 3.92; p<0.001) (Supplementary File 3, Table 1). The direction and significance of this relationship did not change after excluding Washington State or when using different outcome and policy definitions. Each month the policy was in effect was associated with a 0.08 mg MEDD increase relative to control states, (95% CI: 0.07, 0.09; p<0.001) (Supplementary File 3, Table 4).

Several stratified regression analyses were also conducted including separate analyses by policy type (guideline or strong policy), baseline opioid use, and policy exclusion groups. The results of these analyses are summarized in Supplementary File 3, Figure 1. In each of the policy type and policy exclusion stratified analyses, the direction and significance of the relationship between policy and MEDD was consistent, although the magnitude of the relationship was larger in some excluded groups (hospice, cancer diagnosis, inpatient, long-term care) and smaller in other excluded groups (short courses of opioids and patients with acute pain diagnoses). Among patients with any exclusion, the policy was associated with a 3.10 mg MEDD increase relative to control states (95% CI: 2.90, 3.31; p<0.001) and among patients with no exclusions, the policy was associated with a 5.44 mg MEDD increase (95% CI: 4.92, 5.95; p<0.001). There was a small decrease in MEDD associated with the policy among those with baseline use >60 MEDD (−1.26 mg MEDD; 95% CI: −2.28, −0.24; p=0.016) and >80 MEDD (−1.18 mg MEDD; 95% CI: −1.95, −0.42; p=0.003), but no significant change in MEDD associated with the policy among those with baseline use >100 or >120.

Individual regressions where each state was individually compared to control states were also performed for opioid users. Due to data use agreement reporting restrictions, the results from individual states may not be presented. Therefore, a summary of state-level results are presented in Supplementary File 3, Table 9. In guideline states, guideline passage was associated with increased odds of any opioid use in one state and decreased odds of any opioid use in three states. Similarly, in strong policy states, policy passage was associated with increased odds of any opioid use in one state and decreased odds of any opioid use in four states. However, when looking specifically at change in dose among opioid users, guideline passage was associated with increased MEDD in all states, regardless of outcome or whether individuals fell into groups specifically excluded from the policies. In strong policy states, the associations between policy passage and MEDD among opioid users were mixed. Policy passage was associated with increased MEDD in one state, decreased MEDD in another state, and no significant relationship with MEDD in the remaining three states. Results were also mixed when examining dichotomous outcomes and stratification by state-specific exclusion groups.

Discussion

Overall, both guidelines and strong policies were associated with lower odds of using opioids relative to control states. However, there was no statistically significant relationship between either type of policy and high dose opioid use. Among opioid users, there was, contrary to our hypotheses, a small, but statistically significant increase in dose, and excluding certain individuals for whom the policy was not specifically targeted did not change the magnitude or direction of the relationship. The small increase in dose may have been due to lower proportions of individuals receiving any opioids leaving only patients with higher pain management needs still receiving opioids. It is possible that MEDD thresholds as part of a larger set of opioid policies and accompanied by an increased awareness of the risks of opioid prescribing may have led to a decrease in the proportion of individuals prescribed any opioids. However, there is no evidence of decreased prescribing above the threshold levels set by the policies. In subpopulation analyses, we also did not see decreases in dose among patients with high dose baseline use or chronic, non-cancer pain. There was also no evidence that policies with a higher theorized impact—rules/regulations, legislative acts, and passive alert systems—had a greater impact on prescribed MEDD than did guidelines. Therefore, it is unlikely that any observed changes in overall opioid prescribing were due specifically to setting MEDD thresholds. Our findings of a small increase in opioid dose associated with MEDD policy implementation among opioid users in a privately insured population contrast with previous research on MEDD workers compensation.⁸ Among injured workers receiving workers’ compensation insurance, state MEDD threshold guidelines were associated with a decrease in prescribed MEDD.

This study had a number of important strengths, including a large, multi-state study population. This allowed for adequate sample size, even in the smallest sample subgroups, for example, individuals receiving hospice care. The study also makes use of longitudinal data over a period of six years and includes multiple pre- and post- observations for multiple policies of interest. The analysis improves upon prior evaluations of MEDD policies by making use of comparison states that did not implement policies. This study also builds upon previous work which systematically examined policy structure, threshold level, and excluded groups and used this information to evaluate several different aspects of the policies.

This study must also be evaluated in light of its limitations. From 2010 to 2016, the United States has experienced a complex and dynamic policy environment surrounding opioids. While care was taken to select control states that did not pass major opioid legislation, MEDD policies, or implement PDMPs, policies at the local level were not systematically captured and may have had an influence in certain states. Other types of state policies may have also influenced results. For example, while Tennessee did not implement a PDMP during the study time period, they did implement PDMP must-access mandate which may have influenced prescribing practices.¹⁶ Individual insurers are also increasingly instituting a variety of coverage and utilization management policies to reduce high risk opioid use and some of these practices may include use of MEDD thresholds, which may bias the results. It is also important to note that only opioids from commercial claims covered by MarketScan were observed. It was not possible to observe if individuals obtained opioids from other payers or sources. Furthermore, while MarketScan includes commercially insured individuals from all states, the percent of the population they coverage varies by state and may not be representative of the commercially insured population in that state as a whole. While we do not expect the composition of our population to vary substantially within a state from year to year, this may limit the generalizability of our findings.

Future work should evaluate the impact of MEDD policies in other contexts and populations. Given previous research showing decreases in high dose prescribing in workers’ compensation and Medicaid populations,^6–8 it is possible that guidelines targeted towards a specific population may be more effective than those aimed at the general population. Furthermore, other types of policies not included in this study, such as prior authorization requirements for high dose opioid, may be more effective at changing prescribing behavior.

In addition to evaluating the impact of MEDD policies on prescribing behavior, further research should evaluate their impact on patient outcomes. Surveys or qualitative research may illustrate differences in dissemination efforts and policy awareness and help understand why certain policies are more influential. This work, combined with the present study will elucidate how state-level MEDD policies may influence opioid prescribing and outcomes.