Holding other factors constant, the authors find that older providers prescribe significantly more opioids, with the gap between older and younger providers increasing from 2010 to 2015.
Objectives: The relationship between provider age and quality of care is theoretically indeterminate. Older providers are more experienced, which could lead to a positive relationship between age and quality, but providers’ practice patterns could become outdated as technology and scientific knowledge change, which could lead to a negative relationship between age and quality. However, little work has investigated the provider age/quality relationship, and no work has investigated the relationship between provider age and opioid prescribing behavior.
Study Design: We analyze Medicare Part D data to investigate how opioid prescribing differs by provider age.
Methods: We use regression analysis to estimate the effect of provider age, holding other factors constant.
Results: We find that older providers prescribe significantly more opioids, with the gap between older and younger providers increasing from 2010 to 2015.
Conclusions: Assuming that older physicians follow patterns of previous generations, anticipated retirement of older providers and entry by younger providers will tend to reduce opioid volumes, undoing at least in part the rapid increase since 2000.
Am J Manag Care. 2022;28(5):223-228. https://doi.org/10.37765/ajmc.2022.89143
The increase in the volume of opioids prescribed and the harms from opioid use over the past 2 decades are well known.1,2 Evidence indicates that provider behavior has played a key role in expanding opioid use (we refer to prescribers of opioids as providers, which includes physicians, nurse practitioners, and physicians’ assistants) and that at least some of this expansion is medically inappropriate.3
Although the volume of opioids prescribed peaked in 2011 and then decreased each year through 2016,4 opioid use remains far above its 2000 level. Understanding the role of provider characteristics in opioid prescribing and the potential role of changing provider characteristics on prescribing trends is essential for developing an appropriate policy response.
Available evidence shows that there is variation across providers in opioid prescribing. There are differences in opioid prescribing among specialties,5 among physicians in the same specialty, and even among physicians in the same emergency department, with consequent increases in long-term opioid use among patients treated by physicians who prescribe large amounts of opioids.6 Other work shows that physicians trained at less selective medical schools write significantly more opioid prescriptions than those trained at more selective schools.7
Although this work provides important insights, no work has investigated whether the decline in opioid volume since 2011 has been due to changes in provider characteristics—or in spite of them. In particular, no work has investigated the association between opioid prescribing and one of the most basic provider characteristics: age.
Provider age might be an important determinant of opioid prescribing behavior for several reasons. On one hand, older providers’ greater experience might lead them to write fewer inappropriate opioid prescriptions.8 On the other hand, evidence suggests that older physicians can be less likely to adhere to and be aware of current evidence-based standards of care, such as the recommendation to use episiotomy only in limited circumstances.9 (Episiotomy is a surgical procedure to enlarge the vaginal opening during childbirth. Episiotomy was used routinely into the 1990s, but has since been shown to be clinically beneficial only in limited circumstances.) Other work has found that older primary care physicians have higher rates of use of low-value services10 and higher mortality of patients treated in the hospital.11 To the extent that at least part of the expansion of opioid prescribing was not medically appropriate, this work would predict a positive association between provider age and opioid prescription volume.
To investigate whether opioid prescribing behavior differs by provider age, we analyze prescription drug claims from a 20% random sample of Medicare Part D beneficiaries in 2010 and 2015. Using the methods described below, we calculate for each provider the number of days supplied of opioids they prescribed per attributed patient, along with several other measures of the intensity and appropriateness of opioid prescribing. We examine whether these measures vary across providers of different age cohorts, adjusted for the characteristics of the providers and their patients. Then, we create illustrative forecasts of opioid prescription volume for each age cohort of providers in 2020 and 2025 under different assumptions about cohorts’ future prescribing behavior, and aggregate to the population of providers, in order to provide insights into the effects on the overall future volume of prescribed opioids that could be possible based on our estimates. Our forecasts are not intended to represent what will actually occur in the future. Instead, they are intended to represent what would occur under simple assumptions about trends in provider and prescribing behavior, in order to illustrate how these factors might contribute to future opioid volume.
Differences in opioid prescribing behavior by provider age have several implications for policy. Differences by provider age can suggest targeted approaches to improving prescribing practice and may indicate how future opioid prescribing behavior could be affected by the age distribution of the physician workforce as older doctors retire over time and younger doctors enter practice.
DATA AND METHODS
We begin with all the providers serving Medicare beneficiaries with Part D coverage in our 20% sample in 2010 or 2015. We include providers serving both traditional Medicare fee-for-service and Medicare Advantage beneficiaries. We report the fraction of patients enrolled in Medicare Advantage by provider age group in the eAppendix (available at ajmc.com).
We match each provider in the sample by National Provider Identifier (NPI) to records in the Medicare Provider Practice and Specialty file (MD-PPAS) to obtain the provider’s age, gender, and 62-category specialty. We match each provider by NPI to records in the National Plan and Provider Enumeration System (NPPES) to obtain county of practice. We limit our analysis sample to those providers with valid age, gender, and location information in MD-PPAS and NPPES who were aged at least 30 years in 2010 or 2015. We assign providers to nine 5-year age cohorts: 30 to 34, 35 to 39, 40 to 44, 45 to 49, 50 to 54, 55 to 59, 60 to 64, 65 to 69, and 70 years and older. We also present the full distribution of provider ages in the eAppendix.
In 2010 and again in 2015, we attribute to each provider all the unique beneficiaries for whom the provider wrote any prescriptions in that year. Based on this, we calculate the age cohort, gender, race (Black vs non-Black), and original Medicare eligibility status (disabled vs aged) mix of the provider’s patients in the year. We group patients into different age cohorts (0-39, 40-54, 55-64, 65-69, 70-74, 75-79, 80-84, 85-98, ≥ 99 years) than providers because the distribution of patients is highly skewed to older ages.
We also calculate days supplied of opioids prescribed by the provider to their attributed patients in that year. We determine which drug claims are for opioids based on the National Drug Code on the claim and the CDC’s drug classification system,12 and we sum days supplied of each opioid prescription across claims from each provider’s attributed patients.
To determine the relationship between provider age group and opioid prescription patterns, we estimate regressions across providers, by year, of days supplied of opioids per attributed patient on a set of indicator variables for the characteristics of each provider and their patients. Provider characteristics include age cohort; gender; specialty; county of practice; and patients’ age cohort, gender, race, and eligibility status mix.
We provide the full set of regression results, and the mean and SD of each variable, in the eAppendix. We calculate the adjusted days supplied per patient for each provider age cohort in a given year as the predicted value of days supplied per patient at the sample average of all the indicator variables in the regression (except those for provider age cohort).
To obtain insight into how changes over time in the age distribution of physicians in practice could affect opioid prescribing, we use our regression results to construct illustrative forecasts of days supplied per patient for 2020 and 2025 using 2 methods. The first method assumes that number of days supplied by a provider age cohort changes from 2015 to 2020 and from 2020 to 2025 at the same rate that it changed between 2010 and 2015. So, for example, the forecast days supplied by providers aged 35 to 39 years in 2020 would be equal to days supplied by providers aged 30 to 34 years in 2015, multiplied by the ratio of days supplied by providers aged 35 to 39 years in 2015 to days supplied by providers aged 30 to 34 years in 2010. Days supplied by providers aged 30 to 34 years in 2020 is not determined by this model, so we assume it is equal to days supplied by providers aged 30 to 34 years in 2015. We describe this as the “cohort-specific growth” model.
The second method assumes that the numbers of days supplied by a provider age cohort in 2020 and 2025 are exactly the same as they were for that cohort in 2015. So, for example, the forecast days supplied by providers aged 35 to 39 years in 2020 would be equal to days supplied by providers aged 30 to 34 years in 2015. Numbers of days supplied by providers aged 30 to 34 years in 2020 and 2025 are also not determined by this model, so we assume they are equal to days supplied by providers aged 30 to 34 years in 2015. We describe this as the “cohort-stable growth” model.
We aggregate forecast days supplied by provider age to the population of providers based on the 2015 age distribution of providers. In doing so, our forecasts implicitly assume that the age distribution of providers will remain stable into the future. This abstracts away from the 2 opposing forces that are likely to drive the future provider age distribution: the aging of the physician population and the increase in the share of providers who are (younger) nurse practitioners and physician assistants.
To both validate and extend our main analysis, we calculate for 2010 and 2015 several other measures of the intensity and appropriateness of opioid prescribing in addition to days supplied. Other measures of intensity and appropriateness include the count of opioid prescriptions per patient, the average daily dose in morphine milligram equivalents (MME) per prescribed day, the count of high-dose opioid prescriptions (> 90 MME/day) per patient,13 and the share of patients with an opioid prescription who also had an overlapping prescription for a benzodiazepine.14 (This is a standard measure of inappropriate prescribing because of the danger, especially in older individuals, of concomitant use of opioids and benzodiazepines.)
We also calculate the number of days supplied of 2 nonopioid pain medications, nonsteroidal anti-inflammatory drugs (NSAIDs) and gabapentin/pregabalin, to investigate whether differences in opioid prescribing by age are driven by opioid prescribing specifically or could be related to differences in providers’ willingness to treat pain more broadly.15
Figure 1 presents, by provider age cohort, opioid days supplied per patient for 2010 and 2015, as well as forecast days supplied per patient for 2020 according to the 2 methods described above. All values in Figure 1 are adjusted for differences across age cohorts in the patient and provider characteristics described above.
Figure 1 shows that, on average, days supplied in 2010 and 2015 rise with provider age, steeply as providers age into their 40s and gradually thereafter. For example, in 2010, the average provider aged 30 to 34 years prescribed 6.3 days per patient, rising to approximately 8.0 days for providers aged 35 to 39 years and 9.1 days for providers aged 40 to 44 years. After age 40 to 44 years, the number of days supplied continues to rise but more gradually. In 2010, the average provider aged 45 to 49 years prescribed 9.8 days, increasing to 10.8 days for providers aged 60 to 64 years, 11.0 days for providers aged 65 to 69 years, and 11.5 days for providers 70 years and older. All differences between providers aged 30 to 34 years and older age cohorts are statistically significant at the 1% level.
Figure 1 also shows that the number of days supplied per patient by providers in a given age group in 2015 was lower than the number of days supplied by providers in the same age group in 2010. This difference is apparent for all providers but is larger for younger than for older providers. In particular, for the average provider aged 30 to 34 years, days supplied per patient declined from approximately 6.3 in 2010 to 5.3 in 2015, or 15.9%, but for the average provider aged 60 to 64 years, days supplied per patient declined from approximately 10.8 in 2010 to 10.4 in 2015, or 3.7%. This foreshadows our main finding—that as older providers retire and younger providers enter practice, the replacement of older with younger providers is expected to lead to declines in the volume of opioids prescribed.
Figure 1 also reports our illustrative forecasts of 2020 days supplied per patient based on our 2 methods. The dark green line reports 2020 forecast days supplied, assuming that days supplied by a provider age cohort changes from 2015 to 2020 at the same rate that it changed from 2010 to 2015. Over the middle range of provider ages, days supplied continues to fall from 2015 to 2020; at the top of the provider age distribution, days supplied remains relatively stable. The largest decline in days supplied occurs for providers aged 40 to 44 years, for whom it declined by 1.3 from 2015 to 2020 forecast. The light green line reports our illustrative forecast for 2020 days supplied, assuming that days supplied by a provider age cohort in 2020 is exactly the same as it was in 2015. By construction, the number of days supplied by the average provider aged 35 to 39 years in the 2020 forecast is exactly the same as the number of days supplied by providers aged 30 to 34 years in the 2020 forecast: Both are equal to the number of days supplied by providers aged 30 to 34 years in 2015. With the exception of providers aged 35 to 39 years, 2020 forecast days are remarkably similar across forecast methods. This suggests that the most important determinant of our forecasts is the (common) replacement of higher-prescribing older providers with lower-prescribing younger providers.
The Table aggregates the age-specific results presented in Figure 1. The first row of the Table shows that, for the average provider, actual days supplied per patient decline from 2010 to 2015 by approximately 9%, from 9.5 to 8.6. The second and third rows of the Table show that, at the 2015 age distribution of providers, forecast 2020 days supplied per patient would decline between approximately 8.6% and 9.1%. The 2 methods forecast continuing declines in days supplied through 2025 because of continuing retirements that lead to the effective replacement of older higher-prescribing providers with younger providers.
To further investigate the importance of entries and retirements to the overall trend in days supplied, we redid the analysis underlying the Table using only those providers who appeared in both the 2010 and 2015 samples. Days supplied from this fixed cohort therefore include the effects of within-cohort trends but exclude the effect of entries and retirements. Among providers in this fixed cohort, days supplied by the average provider actually increased slightly from 9.6 in 2010 to 9.8 in 2015 (not in any exhibit)—suggesting that entries and retirements were largely responsible for the observed overall decline in days supplied. Although the fixed cohort excludes entries and retirements, it also excludes (1) providers who wrote so few prescriptions that they wrote no prescription for a Part D beneficiary in our 20% random sample in 1 but not both years and (2) providers who fail to appear in both years for reasons other than entry or retirement. If these providers are more likely not to prescribe opioids, then the fixed-cohort analysis would overstate the effect of entries and retirements relative to other factors.
We examine 2 other measures of the intensity of opioid prescribing to validate our main result: the regression-adjusted count of opioid prescriptions per patient and mean daily dose in MME per prescribed day. The results of these analyses, presented in the eAppendix, show the same pattern across provider cohorts and over time as do days supplied.
Other analyses of inappropriate prescribing show a significant gradient across provider age cohorts. For example, in 2015, whereas 4.8% of patients with an opioid prescription from the average provider aged 30 to 34 years had an overlapping benzodiazepine prescription from that provider (regression adjusted), 10.2% of patients with an opioid from the average provider 70 years or older had an overlapping benzodiazepine—more than double the level (Figure 2). (Because benzodiazepines were not covered by Part D in 2010, we calculate only the 2015 rate of opioid/benzodiazepine overlap.)
In addition, although the count of high-dose opioid prescriptions per patient declined dramatically from 2010 to 2015 (from 0.093 to 0.057, or by 38.7%), the provider age gradient remained substantial. In 2015, whereas providers aged 30 to 34 years wrote 0.027 high-dose opioid prescriptions per patient on average, providers 70 years and older wrote 0.088—more than triple the level (Figure 3).
The provider-age gradient in opioid prescribing is also present for other pain relievers. For example, in both 2010 and 2015, the average provider aged 30 to 34 years prescribed approximately 3 days of NSAIDs per patient, rising to approximately 5 days for providers aged 50 to 54 years and 6 days for providers 70 years and older (Figure 4). Although the provider-age gradient in gabapentin/pregabalin prescribing is less pronounced, particularly in 2010, older providers prescribe more of these drugs as well (eAppendix).
Medically inappropriate use of opioids in the United States is an important social problem, and provider prescribing behavior plays a key role. However, no work has investigated the association between opioid prescribing and provider age. This gap is surprising, as other work has found less adherence to evidence-based standards of care and lower-quality outcomes on average for older providers compared with their younger counterparts. Extended to the context of opioid prescriptions, such work suggests that a disproportionate share of opioid overuse could be attributable to the prescribing behavior of older providers.
We analyze data from Medicare Part D to investigate whether opioid prescribing behavior differs by provider age. We find that it does. In 2010, adjusted for the characteristics of providers and their patients, the average provider 70 years or older prescribed 11.5 days supplied per patient vs 6.3 days for the average provider aged 30 to 34 years—approximately 83% more. Although days supplied fell in every provider age cohort from 2010 to 2015, consistent with the overall decline in opioid prescribing documented elsewhere, they fell more for younger than for older providers. Indeed, even as days fell overall from 2010 to 2015, the gap in days across provider cohorts widened: from 5.2 days for providers 70 years and older vs 30 to 34 years in 2010 to 6.1 days for providers 70 years and older vs 30 to 34 years in 2015.
On average, older providers not only prescribe a greater volume of opioids in terms of days supplied, but also prescribe opioids less appropriately. In 2015, compared with providers aged 30 to 34 years, the average provider 70 years or older wrote more than twice as many concomitant benzodiazepine prescriptions for patients for whom they had prescribed an opioid. Similarly, in 2015, compared with providers aged 30 to 34 years, the average provider 70 years or older wrote more than 3 times as many high-dose opioid prescriptions per patient. Although comparisons of potentially inappropriate prescribing were most dramatic at the end points of the provider age distribution, there were also very large differences in the distribution’s interior. This appears to be due to a higher level of pain treatment prescribing more broadly, rather than opioid prescribing specifically: Older providers also prescribe more of other, nonopioid pain relievers.
Our findings have at least 3 limitations. First, we cannot definitively rule out the possibility that the observed difference in opioid volume by provider age is due to unobserved differences in the health of the 2 groups’ patient populations. To address this concern, we control for the variables described above, but other differences in the health of the groups’ patient populations may remain. However, even if there are differences in the groups’ patient populations, those differences would have to be very large and growing to explain our findings.
We took an additional step to further investigate whether the observed difference in opioid volume by provider age is due to unobserved differences in patients’ health. The Medicare Fee-for-Service (FFS) Provider Utilization and Payment Data Part D Prescriber Public Use File (PUF) reports the mean Hierarchical Condition Category (HCC) risk score for each provider serving Part D beneficiaries. These PUFs exist for 2013 to 2017 only. We reestimated the 2015 regression on which our estimates of adjusted differences in opioid volume by provider age are based including the providers’ FFS HCC score as reported in the PUF. When we did this, adjusted differences in opioid volume between younger and older providers actually increased, although by a relatively small amount (approximately 5%).
Second, our construction of the size of each provider’s patient panel based on Part D prescription drug data rather than other Medicare data could bias our results. If older providers are more likely to have patients with no prescribed medications, then our use of Part D data to construct panel sizes would lead us to understate older providers’ panel sizes and in turn overstate older providers’ days supplied per patient. However, this source of bias is unlikely to be important: Of the 39.5 million beneficiaries enrolled in Part D in 2015, 38.9 million (or 98.5%) had at least 1 outpatient prescription drug claim.16
Third, because our study population is disproportionately older, we cannot identify the extent to which the provider age effect accounts for trends in opioid prescribing in the younger population. Other work shows that the decline in opioid prescriptions in recent years was much greater for younger than for older patients.17 Differences between older and younger patients may be important, particularly in light of work showing a large recent decline in initial opioid prescriptions in the younger population,18 which could translate into even greater declines in overall opioid use in the future, given opioids’ addictive nature. Future work might seek to investigate this hypothesis, along with the mechanisms through which provider age affects prescribing behavior, in order to improve public policy.
Holding other factors constant, we find that older providers prescribe significantly more opioids, with the gap between older and younger providers increasing from 2010 to 2015. We conclude that policies to improve prescribing practice may have a greater impact if they are targeted at older providers. In addition, we conclude that the anticipated retirement of older providers and entry by younger providers will tend to reduce opioid volumes, undoing at least in part the rapid increase since 2000.
The authors would like to thank seminar participants at Stanford University for helpful comments and Takaaki Sagawa for exceptional research support. All authors contributed equally to all aspects of the paper.
Author Affiliations: Stanford University (LCB, DPK), Stanford, CA; National Bureau of Economic Research (LCB, DPK), Cambridge, MA; University of Chicago (GKV), Chicago, IL.
Source of Funding: None.
Author Disclosures: Dr Baker has received consulting fees from insurers, hospitals, integrated delivery systems, and other providers of medical products and services including the distributor defendants in the national prescription opiate litigation; serves as a senior adviser at Cornerstone Research, a provider of litigation consulting services; and has received grant support from the National Institutes of Health, The Agency for Healthcare Research and Quality, the Robert Wood Johnson Foundation, and the National Institute of Health Care Management. Dr Kessler has received speaking and consulting fees from insurers, integrated delivery systems, and other providers of and investors in medical products and services including the distributor defendants in the national prescription opiate litigation; serves as a senior adviser at Cornerstone Research; and has received grant support from the Agency for Healthcare Research and Quality. Mr Vaska reports no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (LCB, DPK, GKV); acquisition of data (LCB, DPK, GKV); analysis and interpretation of data (LCB, DPK, GKV); drafting of the manuscript (DPK, GKV); critical revision of the manuscript for important intellectual content (LCB, DPK, GKV); statistical analysis (LCB, DPK, GKV); provision of patients or study materials (DPK, GKV); administrative, technical, or logistic support (LCB, DPK, GKV).
Address Correspondence to: Daniel P. Kessler, PhD, Stanford University, 434 Galvez Mall, Stanford, CA 94305. Email: email@example.com.
1. Guy GP Jr, Zhang K, Bohm MK, et al. Vital signs: changes in opioid prescribing in the United States, 2006-2015. MMWR Morb Mortal Wkly Rep. 2017;66(26):697-704. doi:10.15585/mmwr.mm6626a4
2. Paulozzi LJ, Jones CM, Mack KA, Rudd RA. Vital signs: overdoses of prescription opioid pain relievers – United States, 1999-2008. MMWR Morb Mortal Wkly Rep. 2011;60(43):1487-1492.
3. Liu Y, Logan JE, Paulozzi LJ, Zhang K, Jones CM. Potential misuse and inappropriate prescription practices involving opioid analgesics. Am J Manag Care. 2013;19(8):648-658.
4. Piper BJ, Shah DT, Simoyan OM, McCall KL, Nichols SD. Trends in medical use of opioids in the U.S., 2006-2016. Am J Prev Med. 2018;54(5):652-660. doi:10.1016/j.amepre.2018.01.034
5. Levy B, Paulozzi L, Mack KA, Jones CM. Trends in opioid analgesic-prescribing rates by specialty, U.S., 2007-2012. Am J Prev Med. 2015;49(3):409-413. doi:10.1016/j.amepre.2015.02.020
6. Barnett ML, Olenski AR, Jena AB. Opioid-prescribing patterns of emergency physicians and risk of long-term use. N Engl J Med. 2017;376(7):663-673. doi:10.1056/NEJMsa1610524
7. Schell M, Currie J. Addressing the opioid epidemic: is there a role for physician education? Am J Health Econ. 2018;4(3):383-410. doi:10.1162/ajhe_a_00113
8. Choudhry NK, Fletcher RH, Soumerai SB. Systematic review: the relationship between clinical experience and quality of health care. Ann Intern Med. 2005;142(4):260-273. doi:10.7326/0003-4819-142-4-200502150-00008
9. Howard DH, Hockenberry J. Physician age and the abandonment of episiotomy. Health Serv Res. 2019;54(3):650-657. doi:10.1111/1475-6773.13132
10. Schwartz AL, Jena AB, Zaslavsky AM, McWilliams JM. Analysis of physician variation in provision of low-value services. JAMA Intern Med. 2019;179(1):16-25. doi:10.1001/jamainternmed.2018.5086
11. Tsugawa Y, Newhouse JP, Zaslavsky AM, Blumenthal DM, Jena AB. Physician age and outcomes in elderly patients in hospital in the US: observational study. BMJ. 2017;357:j1797. doi:10.1136/bmj.j1797.
12. Opioid prescription data resources. CDC. Accessed April 14, 2022. https://www.cdc.gov/opioids/data-resources/index.html
13. Strickler GK, Kreiner PW, Halpin JF, Doyle E, Paulozzi LJ, Opioid prescribing behavior – prescription behavior surveillance system, 11 states, 2010-2016. MMWR Surveill Summ. 2020;69(1):1-14. doi:10.15585/mmwr.ss6901a1
14. Hwang CS, Kang EM, Kornegay CJ, Staffa JA, Jones CM, McAninch JK. Trends in the concomitant prescribing of opioids and benzodiazepines, 2002-2014. Am J Prev Med. 2016;51(2):151-160. doi:10.1016/j.amepre.2016.02.014
15. Table DDE-E. National Committee for Quality Assurance. 2017. Accessed April 14, 2022. https://www.ncqa.org/wp-content/uploads/2018/08/DDE-E_2017-final.xlsx
16. Medicare Part D enrollment and utilization by state, calendar year 2015. CMS. Accessed April 14, 2022. https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Medicare-Provider-Charge-Data/Downloads/Medicare_Part_D_Enrollees_and_Utilizers_by_State_CY2015.zip
17. Schieber LZ, Guy GP Jr, Seth P, Losby JL. Variation in adult outpatient opioid prescription dispensing by age and sex – United States, 2008-2018. MMWR Morb Mortal Wkly Rep. 2020;69(11):298-302. doi:10.15585/mmwr.mm6911a5
18. Zhu W, Chernew ME, Sherry TB, Maestas N. Initial opioid prescriptions among U.S. commercially insured patients, 2012-2017. N Engl J Med. 2019;380(11):1043-1052. doi:10.1056/NEJMsa1807069