Opioid use in older adults and Medicare Part D

Abstract

Objective

To determine whether the introduction of prescription drug coverage under Medicare Part D increased opioid prescriptions, patient care‐seeking for pain, and pain diagnoses among elderly Medicare‐eligible adults.

Study Setting

Office visits by adults aged 18 years or older from the 2000‐2016 National Ambulatory Medical Care Survey (12 375 207 253 office visits), and respondents from the 2000‐2017 Medical Expenditure Panel Survey (4 023 418 681 individuals).

Study Design

We compared care‐seeking for pain, provider‐assigned pain diagnoses, and opioid prescriptions before and after the Medicare eligibility age of 65, and before and after Part D's implementation using a regression discontinuity, difference‐in‐differences design. Analyses were adjusted for age, sex, race, and year.

Principal Findings

Patient care‐seeking for pain increased by 11.4 office visits per 100 people (95% confidence interval 2.0‐20.8), or 29%, in response to the implementation of Part D. Opioid prescriptions and diagnoses of pain‐related conditions did not change significantly, but the financing of opioid prescriptions shifted from private to public payers at age 65.

Conclusions

The introduction of Medicare Part D was not associated with increased opioid use among older adults. Rather, opioid use among the elderly has been driven by high levels of opioid use among commercially insured adults who subsequently age into Medicare. Our findings raise the question of whether more judicious prescribing to younger adults coupled with concerted efforts to deprescribe opioids when appropriate may prevent problematic opioid use among the elderly.

What is already known on this topic

• One‐third of Medicare beneficiaries received an opioid prescription in 2017, raising questions about what has driven high levels of opioid use and, in particular, the role of Medicare's Part D prescription drug benefit.

• Literature to date has focused on supply‐side factors influencing opioid prescribing—for example, the role of physician preferences, “pill mills,” or promotional activities by pharmaceutical companies.

• Demand‐side factors such as changes in insurance benefit design or pain prevalence have received less attention, but could influence patients' preferences for pain treatment and opioids.

What this study adds

• Patient care‐seeking for pain increased significantly in response to the implementation of Medicare Part D in 2006.

• Diagnoses of pain and opioid prescriptions were not significantly affected by Medicare Part D's implementation.

• The high prevalence of opioid use among Medicare beneficiaries is explained by recent increases in opioid use among commercially insured adults who subsequently age into Medicare.

1. INTRODUCTION

In 2017, one in three Medicare beneficiaries received an opioid prescription. ¹ Given mounting research indicating that long‐term opioid use does not effectively treat chronic pain and abundant evidence of the harms of opioid use—particularly among older adults—both physicians and Medicare have a vital interest in curbing such use where appropriate. ² , ³ Indeed, the Centers for Medicare and Medicaid Services have continued to develop policies targeting the high prevalence of opioid use among Medicare beneficiaries through prevention, treatment, and monitoring. ⁴

A major obstacle to the success of these efforts, however, is that the reasons behind such high levels of opioid use among elderly Medicare enrollees are not well understood. To date, much of the literature examining contributors to the growth in medical opioid use have focused on supply‐side factors influencing opioid prescribing—for example, the role of physician preferences, “pill mills” or promotional activities by pharmaceutical companies directed at changing prescribers' behavior. ⁵ , ⁶ Demand‐side factors that might influence patients' preferences for pain treatment and opioids have received less attention, yet are also potentially an important target of public health initiatives to limit inappropriate opioid use. In the case of older adults, demand‐side factors affecting opioid use may include changes in pain prevalence or, importantly, changes in insurance benefit design with respect to opioids and other treatments for pain. ⁷

Prior research has shown that expanding prescription drug coverage through public insurance programs such as Medicare Part D and Medicaid decreased out‐of‐pocket drug costs and increased prescription drug use among beneficiaries. ⁸ , ⁹ Because Part D plans cover commonly prescribed opioid analgesics, it is therefore possible that the introduction of Part D might have led some older adults to begin or increase their use of opioids. Evidence of the impact of public insurance expansions on opioid analgesic use specifically, however, is limited and mixed. On the one hand, recent Medicaid expansions do not appear to have increased prescription opioid use. ⁸ , ¹⁰ On the other hand, other research has found that opioid abuse grew faster in states where are a larger share of the population was eligible for expanded prescription drug coverage under Part D. ¹¹ This raises the question of whether Part D might have contributed to high levels of prescription opioid use.

To investigate whether opioid use among older adults increased in response to the 2006 implementation of Medicare Part D, we used nationally representative survey data to analyze care‐seeking by patients with pain, diagnoses of pain‐related conditions, and opioid prescriptions between 2000 and 2017, a period including Part D's introduction. We improve on prior work by using a regression discontinuity design, which compares pain treatment in adults just above the age 65 threshold for Medicare eligibility, to those just below the age cutoff after adjusting for age, sex, race, and time trends. This approach allows us to more accurately capture changes in pain treatment when Part D was implemented by comparing individuals who have similar ages and characteristics, other than their eligibility for Part D. This combined approach, a regression discontinuity‐difference‐in‐differences design, also limits age‐related confounding that occurs when comparing groups further away from the Medicare eligibility age, a common challenge in standard differences‐in‐differences study designs. ¹² , ¹³

2. METHODS

2.1. Data sources and study measures

We used the most recently available data from two sources: the restricted version of the National Ambulatory Medical Care Survey (NAMCS), and the Medical Expenditure Panel Survey (MEPS).

National Ambulatory Medical Care Survey is a nationally representative, cross‐sectional survey of ambulatory care visits in the United States. It surveys nonfederal, office‐based physicians regarding approximately 30 randomly selected patient visits during a random 1‐week period annually. The sample frame included all visits by patients aged 18‐84 that occurred between January 1, 2000 and December 31, 2016, yielding a sample of 444 433 visits. This corresponded to 12 375 207 253 visits when weighted to be nationally representative using NAMCS survey weights (See Table S1 for observations by year).

To examine both care‐seeking for pain by patients, as well as opioid prescribing by physicians, our primary outcomes were the number of office visits with (a) patient‐reported pain complaints; (b) provider‐assigned diagnoses of pain; or (c) a newly prescribed or continued opioid prescription. Visits for patient‐reported pain complaints were identified by the patient's stated reason for the visit—complaints that explicitly mentioned “pain” (eg, “chest pain”) or painful conditions (eg, “kidney infections”) were included (more details are provided in Appendix S1). We identified visits with provider‐assigned diagnoses of pain based on International Classification of Diseases, Ninth and Tenth Revision diagnosis codes corresponding to any of over 200 conditions that commonly cause pain severe enough to require prescription‐strength analgesics, following recent work (Appendix S1). ¹⁴ Opioid prescriptions were identified by their generic components and included all oral, sublingual, or transdermal formulations (Appendix S1). Covariates included patients' year of visit and self‐reported gender and race (“black,” “white,” or “other”).

To limit selection bias, we focused on the impact of Medicare eligibility on the outcomes of interest, rather than actual Medicare enrollment. Individuals who are still working and have employer‐sponsored insurance might be more likely to delay enrolling in Part B, C, or D compared to individuals who retire earlier as a result of chronic medical conditions—including conditions associated with pain. ¹⁵ Thus, by using Medicare eligibility as the exposure of interest, we can be more confident that any observed changes in pain treatment are not being driven by selection of patients into Parts B, C, or D.

We determined Medicare eligibility on the day of the visit by using the patient's exact birth date and visit date to calculate patient age in quarters. Individuals were determined to be eligible for Medicare on the first day of the month in which they turned 65 (or the prior month if born on the first of the month). For each outcome, we then computed the weighted number of visits by patient age (in quarters relative to Medicare eligibility) and calendar year. To account for differences in the size of age cohorts over time, we normalized the number of visits per quarter by the number of individuals of that age in that year using United States Census population estimates (Appendix S2). ¹⁶ , ¹⁷ These estimates were only available for individuals under age 85, which bounded the maximum age of our sample.

Our second data source was the MEPS, which quantified the number of opioid prescriptions filled in a given year by the same patient. ¹⁸ Since NAMCS surveys visits and does not follow patients longitudinally, the use of the MEPS was valuable for counting how many opioids were filled per person over time. We included respondents aged 18‐84 surveyed between January 1, 2000 and December 31, 2017, yielding a sample of 233 438 individuals. This corresponded to 4 023 418 681 individuals when weighted to be nationally representative using MEPS survey weights (See Table S1 for observations by year).

Opioid prescriptions were identified by Multum prescription codes of “narcotic analgesics” or “narcotic analgesic combinations.” Tramadol, which was not added to the Multum classification until 2011, was classified as an opioid in all years (Appendix S1). ¹⁹ We used respondents' age in years to determine if they were age 65 or older and, thus, Medicare‐eligible. Covariates included survey year, sex, and race (“black,” “white,” or “other”).

All analyses used Stata 16. The ‐svy‐ package was used to calculate summary statistics and estimate the MEPS regressions. The rdrobust package of commands was used for NAMCS regressions. ²⁰

2.2. Statistical analysis

Our main analyses use a regression discontinuity‐difference‐in‐differences (RD‐DD) design. The regression discontinuity (RD) component compares changes in the outcomes of interest for individuals arbitrarily close to the age‐eligibility threshold using individuals' age in quarters. ²¹ A key advantage of this approach is that by comparing individuals just on either side of the Medicare age‐eligibility threshold (ie, age 65), we limit bias from age‐related differences in pain and opioid use. The RD model was estimated separately for the pre‐Part D (2000‐2005) and the post‐Part D periods (2006‐2015). The difference‐in‐differences (DD) component compares the pre‐ and post‐Part D RD estimates, which isolates changes in the outcomes of interest attributable to Part D, as opposed to Medicare generally. The pre‐part D RD estimate measures the impact of becoming Medicare‐eligible at age 65 on the outcomes of interest and the post‐Part D RD estimate measures the impact of becoming Medicare‐eligible at age 65 and also gaining prescription drug coverage through Part D. RD‐DD models were estimated at the level of age‐in‐quarters and year using restricted NAMCS data for 2000‐2015, and including controls for age, year, race, and sex. The estimating equation is shown in Appendix S2. The RD‐DD design assumes that changes that occur at the age of Medicare eligibility would have been the same in the pre‐ and post‐period had Part D not been enacted. ²² To evaluate this assumption visually, outcomes above and below age 65 before 2006 were graphed.

Medical Expenditure Panel Survey analyses used a standard DD design to estimate the difference in the annual number of opioid prescriptions filled per person between ages 60‐64 and 66‐70, before and after 2006. Individuals aged 65 were excluded because, in contrast to NAMCS, respondents' exact age at the time of the prescription is not released by MEPS. The estimating equation, which included controls for age, year, sex, and race, is shown in Appendix S3. This design assumed that the difference in outcomes between ages 66‐70 and ages 60‐64 would have been the same after 2006 as before 2006, had there been no Part D. This assumption was assessed graphically. This was a stronger assumption than that needed for the NAMCS analyses because the threat of confounding due to age‐related differences in pain and opioid use grows as the size of the age group increases.

2.3. Additional analyses

Because the NAMCS only samples office‐based physicians, it does not capture prescribing that occurs in other ambulatory care settings. Analyses were therefore re‐estimated in the National Hospital Ambulatory Medical Care Survey (NHAMCS), which includes hospital outpatient departments. Also, substance abuse clinics are included in the NAMCS sampling frame but not explicitly identified. We therefore also used NHAMCS—which identifies substance abuse clinics—to determine if our results were sensitive to their exclusion. We also did several other robustness checks that varied the time frame, bandwidth around the Medicare eligibility age of 65, and the functional form assumptions. Finally, our preferred MEPS specification used respondents within 5 years of Medicare eligibility (ages 60‐70) to limit confounding from age‐related trends in pain and opioid use. We test the sensitivity of this restriction using samples of adults ages 55‐75 and 50‐80.

3. RESULTS

3.1. Care‐seeking, pain diagnoses, and opioid prescription rates in NAMCS

We first show the age profile of patient care‐seeking for pain, provider‐assigned diagnoses of pain, and opioid prescription rates, for the periods before and after the implementation of Part D. Figure 1 shows that, as individuals age, they are more likely to visit the doctor with pain complaints, to be diagnosed with a pain‐related condition, and to be prescribed an opioid. At every age, the rate of visits with pain complaints or provider‐assigned pain diagnoses was similar before and after 2006. The average number of visits with an opioid prescription, however, was nearly twice as high after 2006 compared to before 2006 across all ages. These trends highlight the importance of secular changes in opioid prescribing patterns—which have affected patients of all ages—in contributing to higher levels of opioid use.

FIGURE 1.

Visits by cohort in National Ambulatory Medical Care Survey (NAMCS). Using the 2000‐2016 NAMCS, the number of visits with a pain‐related reason for visit, where pain was diagnosed, and with opioid prescriptions are shown. The x‐axis measures patient age at the time of visit. The y‐axis measures the number of visits per 100 people by age. Visits were weighted using NAMCS sample weights and normalized to be per 100 people in each age‐year bin to account for changes in cohort sizes over time [Color figure can be viewed at wileyonlinelibrary.com]

3.2. Role of Medicare Part D

For the NAMCS analyses, Table 1 shows there were no clinically or statistically significant differences in age, sex, and race the quarter before Medicare age‐eligibility compared to the quarter after, consistent with the assumption of the RD estimator (the first difference of the DD estimator). Further, reaching the age of Medicare eligibility significantly increased the number of visits with Medicare as a primary source of payment, a necessary pre‐condition for analyses (21.5%‐53.4%; P < .01). Table 1 also shows that, when comparing individuals in the quarters before and after Medicare eligibility, there was no longer an age‐related trend in the outcome variables of interest, as in Figure 1. In support of the validity of the DD design, Figure S1 shows that pre‐2006 outcome trends were parallel for patients under and over the age of Medicare eligibility.

TABLE 1.

Characteristics of the NAMCS and MEPS study population

Characteristics	NAMCS data a			MEPS data b
	Quarter before medicare eligibility age	Quarter after medicare eligibility age	P value c	Patients aged 60‐64	Patients aged 66‐70	P value d
Number of NAMCS visits, MEPS individuals in millions	46.9	54.0		285.5	213.0
Patient covariates
Mean age, y	64.8	65.1		61.9	67.9
Female, %	56.1	59.0	.24	52.4	52.9	.46
White, %	86.7	87.3	.70	83.3	84.1	.11
Black, %	8.9	9.4	.68	10.5	9.9	.13
Other, %	4.4	3.3	.20	6.3	6.0	.48
Provider experience with patient
Established patient, %	88.7	87.9	.57	‐	‐	‐
New patient/Blank, %	11.3	12.1	.57	‐	‐	‐
Insurance e
Medicare, %	21.5	53.4	<.01	9.7	98.4	<.00
Private, %	63.3	36.1	<.01	71.7	1.1	<.00
Medicaid, %	4.1	2.3	.04	5.1	0.0	<.00
Other, %	11.0	8.2	.03	13.5	0.5	<.00
Outcomes
Visits for pain, %	31.2	32.2	.64	‐	‐	‐
Visits where pain‐related conditions were diagnosed, %	43.0	47.3	.08	‐	‐	‐
Prescribed opioid f , %	9.0	8.2	.57	17.2	18.7	.01
Prescribed opioid g , prescriptions per person, mean	‐	‐		4.5	4.3	.24

Sample characteristics use NAMCS data from 2000‐2015 and MEPS data from 2000‐2017. Estimates used survey‐provided sampling weights to account for complex sampling. Cells are left blank if no relevant data are available.

Abbreviations: MEPS, Medical Expenditure Panel Survey; NAMCS, National Ambulatory Medical Care Survey.

^aThe unit of observation in the NAMCS was at the visit level.

^bThe unit of observation in the MEPS was at the individual‐year level, where age was measured at the end of the survey year (in years). Respondents who were age 65 in a given survey year were excluded because it was not possible to identify whether they received the opioid prescription before or after their 65th birthday.

^c P‐values based on t tests comparing the mean the last quarter before to the mean the first quarter after Medicare eligibility.

^d t Tests determined if means between visits in NAMCS and individuals in MEPS aged 60‐64 and 66‐70 were significantly different.

^e“Insurance” refers to a patient's primary source of payment in the NAMCS and self‐reported insurance status in the MEPS. In both the NAMCS and the MEPS, primary source of payment was coded in mutually exclusive categories of Medicare, private, Medicaid, and other.

^f“Prescribed Opioid” includes visits where at least one opioid was prescribed in the NAMCS or individuals who reported at least one opioid prescription in the MEPS.

^g“Prescriptions per person” is the number of prescriptions, conditional on individuals filling at least one opioid prescription in that survey year in the MEPS.

Table 2 shows our main estimates which differenced the pre‐ and post‐Part D estimates to isolate changes attributable to Medicare Part D alone. We found that the number of visits for pain complaints increased by 11.4 visits (CI, 2.0 to 20.8) per 100 people per quarter of age—a statistically significant increase. Although care‐seeking for pain increased after Part D, there was no significant increase in the number of visits per 100 people per quarter of age that resulted in either a provider‐assigned pain diagnosis or an opioid prescription.

TABLE 2.

NAMCS estimates of the changes in opioid‐related outcomes at age 65, before and after 2006

Quarterly visits per 100	Regression discontinuity estimate at age 65 2000‐2005			Regression discontinuity estimate at age 65 2006‐2015			Difference‐in‐differences estimate a
	Unadjusted rate, quarter before medicare eligibility, mean	Absolute risk difference b (95% CI)		Unadjusted rate, quarter before medicare eligibility, mean	Absolute risk difference b (95% CI)		Absolute risk difference (95% CI)
		Unadjusted	Adjusted c		Unadjusted	Adjusted c	Unadjusted	Adjusted c
Visits for pain	39.8	−3.2 (−10.8 to 4.3)	−3.5 (−11.6 to 4.6)	31.0	8.2 (3.4‐13.1)	7.9 (3.2‐12.7)	11.5 (2.5‐20.4)	11.4 (2.0‐20.8)
Visits where pain‐related conditions were diagnosed	52.7	2.0 (−7.1 to 11.1)	2.0 (−7.6 to 11.6)	43.6	9.9 (2.5‐17.3)	9.9 (3.1‐16.6)	7.9 (−3.8 to 19.7)	7.9 (−3.9 to 19.6)
Visits with an opioid prescription	6.3	0.1 (−3.5 to 3.7)	−0.1 (−3.6 to 3.5)	11.1	2.2 (−1.6 to 6.0)	2.2 (−1.6 to 5.9)	2.1 (−3.2 to 7.3)	2.2 (−3.0 to 7.4)

Analyses of 433 630 visits (11 697 090 694 weighted visits) surveyed from 2000 to 2015 in the NAMCS.

Abbreviations: CI, confidence interval; NAMCS, National Ambulatory Medical Care Survey.

^aDifference‐in‐differences estimate compares the pre‐Part D and post‐Part D regression discontinuity (RD) estimates. Appendix S2 describes the calculation of standard errors.

^bTo determine the number of age‐year bins included in each RD model, we used optimal bandwidth selection (Stata command “rdrobust”). This procedure selected an age bandwidth of 7.63 years on either side of age 65 pre‐part D and 7.70 years on either side of age 65 post‐part D for “Visits for Pain,” 8.53 and 7.46 for “Visits where Pain‐Related Conditions were Diagnosed,” and 8.35 and 8.58 for “Visits with an Opioid Prescription.” These bandwidths yielded the following numbers of observations for each model in the pre‐Part D and post‐Part D periods, respectively: 366 and 610 age‐year bins for “Visits for Pain,” 414 and 590 age‐year bins for “Visits where Pain‐Related Conditions were Diagnosed,” and 402 and 690 age‐year bins for “Visits with an Opioid Prescription.” Appendix S2 describes additional specification details.

^cAdjusted estimates control for survey year as well as patient race and sex.

For the MEPS analyses, Table 1 shows that 9.7% of 60‐ to 64‐year‐olds had Medicare coverage, compared to 98.4% of 66‐ to 70‐year‐olds (P < .01). The percent of individuals prescribed an opioid was also 1.3 percentage points higher among 66‐ to 70‐year‐olds per year, compared to 60‐ to 64‐year‐olds (18.7% vs 17.2%; P = .01). Other respondent characteristics were similar across age groups. Figure S2 shows that the yearly difference in the number of opioid prescriptions per person and the share filling any opioid prescription trended similarly from 2000‐2005 for those aged 60‐64 and 66‐70, providing support for our parallel trends assumption.

Table 3 illustrates that the introduction of Part D did not significantly affect the number of opioids prescribed per person or the likelihood that an individual filled at least one opioid prescription in a given year. This suggests that NAMCS visit‐level and MEPS person‐level estimates yield similar results.

TABLE 3.

MEPS estimates of the change in opioid‐related outcomes before and after 2006

	Difference between ages 60‐64 and 66‐70 2000‐2005			Difference between ages 60‐64 and 66‐70 2006‐2017			Difference‐in‐differences estimate
	Unadjusted mean, before medicare eligibility, 2000‐2005	Absolute risk difference a (95% CI)		Unadjusted mean, before medicare eligibility, 2006‐2017	Absolute risk difference a (95% CI)		Absolute risk difference a (95% CI)
		Unadjusted	Adjusted b		Unadjusted	Adjusted b	Unadjusted	Adjusted b
Number of opioid prescriptions per person c	4.03	−0.20 (−0.84 to 0.43)	−0.46 (−1.49 to 0.56)	4.62	−0.20 (−0.57 to 0.17)	0.04 (−0.76 to 0.83)	0.01 (−0.71 to 0.72)	−0.01 (−0.73 to 0.71)
Percent of people who filled opioid prescriptions d	13.97	0.67 (−0.78 to 2.11)	0.48 (−2.47 to 3.43)	18.36	1.76 (0.47‐3.04)	2.47 (0.41‐4.53)	1.09 (−0.84 to 3.02)	0.98 (−0.93 to 2.90)

Analyses of 31 053 individuals for a total of 47 562 individual‐year observations (498 527 298 weighted individual‐year observations) from 2000 to 2017 in the MEPS. Respondents age 65 were excluded from analyses because it was not possible to identify whether opioid prescriptions were filled before or after individuals' 65th birthday. Age in years was measured at the end of the survey year. Estimates are weighted to account for complex sampling design using MEPS sampling weights. See Appendix S3.

Abbreviations: CI, confidence interval; MEPS, Medical Expenditure Panel Survey.

^aFor the pre‐ and post‐Part D columns, the absolute risk difference compares 60‐ to 64‐year‐olds to 66‐ to 70‐year‐olds. For the difference column, the absolute risk difference compares the difference between 60‐ and 64‐year‐olds and 66‐ and 70‐year‐olds before and after Part D. Further details are provided in Appendix S3.

^bAdjusted estimates control for survey year as well as respondent age, sex, and race.

^c“Number of Opioid Prescriptions per Person” is the number of opioid prescriptions received per individual conditional on filling at least one opioid prescription. Estimates for this outcome were based on 6730 individuals who filled at least one opioid (88 929 222 weighted individual‐year observations).

^d“Percent of People Who Filled Opioid Prescriptions” is an indicator of whether an individual received at least one opioid prescription in a given year.

The results of all sensitivity analyses and falsification tests supported the conclusions of the main analyses. Using the NHAMCS, we found no significant change in the number of hospital outpatient department visits for pain complaints, or the number of visits with pain diagnoses or opioid prescriptions at age 65 in response to Part D's implementation (Table S2). Excluding the 0.7% of visits with opioid prescriptions originating in substance abuse clinic visits did not significantly change our results (Table S3). Other variations in the time frame and the bandwidth around Medicare eligibility as well as modifying functional form assumptions supported the main results (Tables S4‐S7). In the MEPS, widening the age band did not change the main estimates (Table S8). This suggests age and year trend controls reduced confounding from secular and age‐related increases in opioid use, respectively. We also show DD estimates for each age cohort in Figure S3 (“event study estimates”), which support the findings in Table 3.

3.3. Explaining high rates of opioid use in medicare

Figure 2 plots the age profile of visits where opioids were prescribed by payer. Although the total number of opioids trends smoothly through age 65, there is an abrupt decrease in visits covered by private insurance, and a corresponding increase in those covered by Medicare. That is, while the amount of opioid use does not change at Medicare eligibility, its financing shifts from private to public payers. Figure 2 also shows that opioid use among the commercially insured is already high before Medicare eligibility. Further supporting this finding, Figure S4 shows that the number of individuals who filled an opioid prescription in year t who also filled an opioid prescription in year t + 1 is smooth through the age 65 threshold, ranging from 50% at age 60 to 62% at age 70.

FIGURE 2.

Visits with an opioid prescription by age in National Ambulatory Medical Care Survey (NAMCS). Using the 2000‐2016 NAMCS, the number of visits with a recorded new or continued opioid prescription is shown. The x‐axis measures patient age at the time of visit. The y‐axis measures the number of visits per 100 people by age. Visits were weighted using NAMCS sample weights and normalized to be per 100 people in each age‐year bin to account for changes in cohort sizes over time. Payment categories were mutually exclusive and the underlying expected primary source of payment was prioritized in the following order (from highest to lowest): Medicare, private, Medicaid, and other [Color figure can be viewed at wileyonlinelibrary.com]

4. DISCUSSION

Prescription opioid use among the elderly is prevalent and has increased substantially over time. ¹ While there are many possible contributors to this shift, it has been hypothesized that the expansion of prescription drug coverage through Medicare Part D may have increased patients' demand for opioid analgesics. ¹¹ Our study is among the few to assess the impact of insurance coverage expansions, such as Part D, on the use of opioid analgesics. ⁸ , ¹⁰ , ¹¹ , ²³

Our principal finding is that while care‐seeking for pain increased after Part D's implementation, provider‐assigned diagnoses of pain and opioid prescriptions did not correspondingly increase. This finding contrasts with prior work based on fewer years of data, which found that Part D expanded the supply of prescription opioids. ¹⁰ Our analysis does not support the hypothesis that the high prevalence of opioid use observed in the elderly Medicare population is due to Medicare Part D. Instead, we find that from 2000 to 2017, there was a large secular increase in opioid prescribing across all ages, not just among the elderly, which lead to a high prevalence of opioid use among commercially insured adults aged 50‐64. This suggests that prescription opioid use among new Medicare beneficiaries began prior to Medicare enrollment, and that the aging of privately insured opioid users into Medicare has contributed to the increased prevalence of opioid use in Medicare over time. This pattern may not be unique to opioids; since the financing of all prescription drugs shifts from private payers to Medicare Part D at age 65, use of other controversial medications may be similarly persistent. Lastly, the fact that patient care‐seeking for pain increased without a corresponding increase in provider pain diagnoses or prescription opioid use suggests that the additional care‐seeking may have been for mild pain and could have been appropriately treated with non‐opioid and/or non‐pharmacologic treatments. ²⁴

Last, the reasons for an increase in pain‐related visits without a corresponding increase in opioid prescriptions are uncertain, but one possibility is that through the prospect of more affordable prescription medications, Part D might have indirectly incentivized an increase in outpatient visits as a pathway to obtaining prescriptions. If the patients who responded to this incentive through increased care‐seeking disproportionately had milder pain that could have been appropriately treated with non‐opioid and/or non‐pharmacologic treatments, this could explain the discrepancy between the change in pain‐related visits and opioid prescriptions.

This study has several limitations. NAMCS prescription data come from patient medical records and may not reflect actual prescription fills. It also captures prescribing at a single visit and does not follow patients longitudinally. Additionally, the estimated impact of Part D on opioid prescribing is not a precise null result, suggesting that the effect could be as large as an increase of 7.4 opioids or a decrease of 3.0 per 100 people per quarter of age. We address these limitations by replicating our findings in the MEPS, where the number of filled opioid prescriptions per individual can be tracked across multiple visits. A limitation specific to the MEPS is that while self‐reported prescriptions are verified by pharmacies, prescriptions not reported by respondents are missed. To the extent that misreporting is consistent across years and ages, our estimates are likely unaffected. Finally, while we examine the impact of Part D on overall opioid prescribing, we do not identify or assess impacts on problematic opioid use specifically, as this requires more detailed information about the clinical context than is readily ascertainable from the NAMCS and MEPS alone. This is an important area for future research.

While the introduction of Medicare Part D was not associated with increased medical opioid use among older adults, Medicare is inheriting high levels of opioid use from commercial insurers as large numbers of near‐elderly adults on opioid therapy age into Medicare. For providers, there is a clear need to screen for potentially inappropriate opioid use among newly eligible Medicare beneficiaries. The “Welcome to Medicare” preventive visit is one setting in which pain assessment and the evaluation of indications for long‐term opioid therapy could be formally incorporated. For the Medicare program, once potentially problematic opioid use is identified, Medicare can guide providers on how to safely wean patients off opioids as well as connect patients to treatment for opioid dependence and alternative evidence‐based treatments for chronic pain. Further, partnerships between Medicare and commercial insurers to jointly address opioid prescribing may also be needed to achieve safer pain management over the life course.

CONFLICT OF INTEREST

No other disclosures.

Supporting information

Author matrix

Supplementary Material

Sabety AH, Sherry TB, Maestas N. Opioid use in older adults and Medicare Part D. Health Serv Res. 2021;56:289–298. 10.1111/1475-6773.13623

DATA AVAILABILITY STATEMENT

The NAMCS data are publically available, although the restricted version requires authors to apply for restricted access. The MEPS is publically available. The accompanying code for this research can be provided to researchers who contact the corresponding author.