Article Text

Download PDFPDF

Association between industry payments for opioid products and physicians’ prescription of opioids: observational study with propensity-score matching
  1. Kosuke Inoue1,
  2. Jose F Figueroa2,3,
  3. E John Orav2,4,
  4. Yusuke Tsugawa5,6,7
  1. 1 Department of Epidemiology, UCLA Fielding School of Public Health, Los Angeles, California, USA
  2. 2 Department of Medicine, Brigham and Women’s Hospital, Boston, Massachusetts, USA
  3. 3 Department of Health Policy and Management, Harvard T. H. Chan School of Public Health, Boston, Massachusetts, USA
  4. 4 Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
  5. 5 Division of General Internal Medicine and Health Services Research, David Geffen School of Medicine at UCLA, Los Angeles, California, USA
  6. 6 Department of Health Policy and Management, UCLA Fielding School of Public Health, Los Angeles, California, USA
  7. 7 UCLA Center for Health Policy Research, Los Angeles, California, USA
  1. Correspondence to Dr Kosuke Inoue, Department of Epidemiology, UCLA Fielding School of Public Health, Los Angeles, CA, USA; koinoue{at}ucla.edu

Abstract

Background Industry marketing to physicians for opioids has received substantial attention as it can potentially influence physicians’ prescription of opioids. However, robust evidence demonstrating a causal link between industry payments for opioids and physicians’ prescription practice for opioids is lacking.

Methods Using the national databases of physicians treating Medicare beneficiaries, we examined the association between physicians’ receipt of opioid-related industry payments in 2016 and (1) the number of opioids prescribed and (2) the annual expenditures for the opioid products by those physicians in 2017, using propensity-score matching in a 1:1 ratio adjusting for the physician characteristics (sex, years in practice, medical school attended, specialty), the number of opioid prescriptions in 2016, and physicians’ financial relationships with industry in 2015. We compared matched pairs of physicians using the estimated effect and paired t-test.

Results Among 43 778 physicians included after propensity-score matching, physicians who received opioid-related industry payments in 2016 prescribed more opioids (153.8 vs 129.7; adjusted difference (95% CI), 24.1 (19.1 to 29.1)) and accounted for more spending due to opioids ($10 476 vs $6983; adjusted difference (95% CI), $3493 (2854 to 4134)) in 2017, compared with physicians who did not receive payments. The association was larger among primary care physicians than surgeons or specialists. The dose–response analysis revealed that even a small amount of industry payments was sufficient to effectively affect physicians’ prescription practice of opioids.

Conclusions Opioid-related industry payments to physicians in the prior year were associated with a higher number of opioid prescriptions and expenditures for opioid products in the subsequent year.

Statistics from Altmetric.com

INTRODUCTION

Opioid overdoses and related deaths have become one of the major public health problems not only in the US,1 but also in many European countries such as the UK (especially Scotland), Estonia, and Sweden.2 3 Although the number of opioid overdose deaths in the US dropped by nearly 3% in 2018 for the first time since 1990,4 there were still 47 590 deaths related to opioids, which suggests that more efforts are needed in addressing this issue. It is estimated that prescribed opioid products are involved in around 30%–40% of all opioid overdoses, and they can potentially serve as a trigger for subsequent transitioning to the use of illicit opioids and heroin.5 Research has shown that physicians’ prescription of opioids can have a long-term effect on individuals’ addiction to opioids.6 Although we observed a wide variation in the number and dose of opioid prescriptions, it remains largely unknown what determines some physicians to prescribe more opioids than others.7 8

The financial relationships between industry and physicians have received attention as they may affect physicians’ prescription practices that could have an impact on the quality of care patients receive. While both industry marketing for opioid products and physicians’ opioid prescription has been declining in recent years,9 10 it is critically important to characterise the causal link between industry payments for opioids and physician prescription of opioids using the latest data given that the industry–physician relationships for opioid products are seen as one of the important upstream determinants of opioid prescription and opioid overdose. Existing studies on this topic were observational studies that adjusted for a limited set of potential confounders (therefore, there are concerns about residual confounding)11 14 or ecological studies using data aggregated at the regional level15 (thus, susceptible to ‘ecological fallacy’16). In addition, while the difference in financial relationships with industry by physician characteristics has been shown in general,17 19 it is still unclear whether some types of physicians are more susceptible to opioid-related industry payments than others. Finally, although it is important to understand how much industry payments are necessary to influence physicians’ prescription practice, the dose–response relationship between industry payments for opioids and physician prescription of opioid has not been well characterised.

Supplemental material

To address these knowledge gaps, we sought to answer three key questions using the most updated nationally representative database of industry payments to physicians, linked with a comprehensive database of physician characteristics and a national database of physicians’ drug prescriptions. First, is physicians’ receipt of industry payments for opioids associated with their subsequent prescription of opioid products? Second, if there is an association, which types of physicians are more susceptible to industry payments than others? Finally, is there a dose–response relationship between the amount of opioid-related payments from industry and the number of prescription opioids?

METHODS

Data

We created a longitudinal dataset by linking four databases: (1) the Centers for Medicare and Medicaid Services (CMS) Open Payments data in 2015–2016, (2) the CMS National Plan & Provider Enumeration System (NPPES) Database, (3) the CMS Physician Compare Database and (4) the CMS Medicare Provider Utilization and Payment Database in 2016–2017 (figure 1). First, we extracted the data on general payments that listed opioid products approved by the U.S. Food and Drug Administration (FDA) as the primary product from the Open Payments data.20 Second, we linked the Open Payments data with the NPPES Database using physicians’ full name and the zip code for the primary practice location, an approach used in previous studies.17 19 21 Then, we linked the merged database of the Open Payments and NPPES Databases with the Physician Compare database22 and the Medicare Provider Utilization and Payment Database23 using physicians’ National Provider Identifier (NPI).

Figure 1

Database generating process for study population. NPI, National Provider Identifier; NPPES, National Plan & Provider Enumeration System.

Payment data

All non-research general payments per year (eg, food and beverage, travel and lodging, speaker compensation, honoraria, consulting fees, gifts and education materials) were identified with opioid products listed as the primary product, excluding buprenorphine hydrochloride marketed for addiction treatment in 2015 and 2016.11 We did not include research payments because this type of payment could be provided to the institute, but not an individual physician. We also did not include ownership interests because this type of payment was generally not specific to product but pharmaceutical company.

Drug prescription data

Our primary outcome was the total 30-day standardised number of opioid prescriptions in 2017. Our secondary outcome was the total expenditures for the opioid products in 2017. Total claim amount was based on the number of Medicare Part D claims, including original prescriptions and refills. The number was standardised using the number of days supplied on each claim divided by 30 days. The total expenditures were based on the amount paid by the Medicare Part D plan, the Medicare enrollee and other third-party payers or government subsidies, the ingredient cost of the medication, dispensing fees, sales tax and any applicable administration fees. The drug utilisation data in 2016 were also extracted as covariates.

Physician characteristics

We examined the following four characteristics of physicians: sex, years in practice, the ranking of the medical school attended and specialty. Years in practice was defined as years since the graduation of medical school. Based on the 2017 U.S. News & World Report, we categorised medical schools into three groups: those ranked 1 to 20, those ranked 21 to 50 and others (including all unranked and foreign medical schools).24 Physician specialty was classified into 17 categories, as shown in table 1.

Table 1

Physician characteristics according to whether they received opioid-related industry payments in 2016

Statistical analysis

First, we matched each physician who received opioid-related payments in 2016 with a single physician who did not receive the payments using one-to-one propensity-score matching. We used a logistic regression model to derive the propensity scores for the receipt of opioid-related payments in 2016, which included physician characteristics (sex, years in practice, medical school attended and specialty) and relationships with industry (the total amount of industry payments for any drugs received in 2016; the number of opioid prescriptions in 2016 and the receipt of opioid-related industry payments in 2015). For our primary and secondary outcomes, we compared matched pairs of physicians with regard to the number of opioid prescriptions and opioid costs in 2017 using the estimated effect and paired t-test.

Next, to evaluate whether the relationship between the receipt of opioid-related industry payments in 2016 and the number of opioid prescriptions in 2017 varies by the types of physicians, we conducted stratified analyses by each physician characteristic (ie, sex, years in practice, medical school attended and specialty), and also by whether they received opioid-related industry payments in 2015. We formally tested whether the interaction is statistically significant using a Wald test.

Finally, we examined the dose–response relationship categorising the physicians into five groups based on the amount of opioid-related industry payments received in 2016: $0, $1-$14 (the first quartile<Q1/> for physicians who received payments), $15-$26 (the second quartile <Q2/>), $27-$66 (the third quartile <Q3/>) and over $66 (the fourth quartile <Q4/>). Then, we employed multivariable negative binomial regression models (to account for the right-skewed distribution of the prescription data), adjusting for the same covariates used in the propensity-score matching. Given the possible bidirectional relationship between physicians’ prescription of opioids and the industry payments in the same year, opioid prescriptions in 2016 could be seen as a mediator, rather than a confounder, in the relationship between industry payments in 2016 and prescriptions in 2017. To address this problem, we also examined the dose–response relationship without adjusting for the number of opioid prescriptions in 2016.

Sensitivity analyses

Given the potential bias due to residual confounding, we re-analysed the propensity score-matched samples using: (i) negative binomial regression models to account for skewed outcome distributions and (ii) ordinary least squares (OLS) regression models with Huber-White robust SEs to allow for simple interpretations, adjusting for the covariates used to estimate the propensity scores (ie, physician characteristics and relationships with industry). Given the potential influence of prescriptions in the prior year on industry payments, we also analysed the data restricting physicians to those who did not prescribe opioids in 2015. Finally, to assess the robustness of our findings without dual-eligible beneficiaries aged <65 years old, we also analysed the data restricting to Medicare beneficiaries aged 65 years and older.

All analyses were performed using Stata, version 15 (Stata-Corp). The study was exempted by the UCLA Office of the Human Research Protection Program, Institutional Review Board.

RESULTS

Physician characteristics

In our entire cohort (without propensity-score matching), opioid-related general payments worth $7.1 million (mean, $274; min, $3; max, $210 508) were made to 25 923 physicians caring for Medicare beneficiaries in 2016. Characteristics of physicians are shown in table 1, and the propensity-score matching improved the covariate balance considerably.

Industry payments and prescription of opioid products

In the final dataset before propensity-score matching, 21 million claims were filed for opioid products by 157 873 physicians caring for Medicare beneficiaries, with a total opioid-related expenditure of $870 million. After the propensity-score matching, we found that physicians who received industry payments for opioid products in 2016 prescribed 24.1 (95% CI 19.1 to 29.1) more opioids in 2017 on average, compared with physicians who did not receive payments in 2016, after adjusting for potential confounders (table 2). We also found higher mean annual expenditures for opioids for physicians who received opioid-related industry payments in 2016 (adjusted difference, $3493; 95% CI $2854 to $4134) compared with those who did not receive any payments in 2016.

Table 2

Comparison of mean number of opioid prescriptions and expenditure due to opioid prescriptions in 2017, physicians who received opioid-related industry payments vs matched physicians who did not receive payments

Stratified analysis by physician characteristics

In the stratified analyses, the association between opioid-related industry payments and opioid prescriptions did not differ by physician’s sex, years in practice and medical school attended (figure 2, Supplementary Table A). As for specialty, we found smaller associations among surgeons and specialists compared with primary care physicians (surgery vs primary care, p value for interaction =0.001; specialists vs primary care, p value for interaction <0.001). Among specialists receiving the highest amount of opioid-related industry payments, we found no evidence that industry payments were associated with opioid prescriptions among pain medicine physicians (adjusted difference, −13.5; 95% CI −136.1 to 109.1) and rehabilitation physicians (adjusted difference, 44.5; 95% CI −12.3 to 101.3), and even an inverse relationship between industry payments and opioid prescriptions among anesthesiologists (adjusted difference, −36.2; 95% CI −50.1 to −22.4). The difference in the mean number of opioid prescriptions was observed in physicians regardless of receiving opioid-related industry payments in 2015.

Figure 2

Mean number of opioid prescriptions in 2017 according to whether the physician received opioid-related industry payments in 2016 stratified by physician characteristics. *P values for difference are described. Further details, including p values for interaction between subgroups, are shown in Supplementary Table A.

Dose–response relationship between industry payments and opioid prescription

We found a dose–response relationship between the amount of opioid-related industry payments in 2016 and the number of opioid prescriptions in 2017, after adjusting for physician characteristics and relationships with industry (figure 3); a strong positive relationship was observed below industry payments of the second quartile ($15 to $26). We also found the dose–response relationship in the model without adjusting for the number of opioid prescriptions in 2016 (Supplementary Fig. A).

Figure 3

Dose–response relationship between the total amount of opioid-related industry payments received in 2016 and the number of opioid prescriptions in 2017.

Sensitivity analyses

Our main findings were qualitatively unaffected by the use of the negative binomial regression model or the OLS regression model with additional adjustments for physician characteristics and relationships with industry (Supplementary Table B, C). We found consistent results with smaller point estimates even when restricting physicians to those who did not prescribe opioids in 2015, the year prior to the receipt of industry payments (Supplementary Table D) and when using the outcome only among beneficiaries aged 65 years and older (Supplementary Table E).

DISCUSSION

Using the national datasets on industry payments to physicians, we found that opioid-related industry payments to physicians in 2016 were associated with a higher number of opioid prescriptions, as well as expenditures for opioid products in 2017. The association was consistent across subgroups of the physicians being prominent among primary care physicians and physicians who did not receive the payments in 2015. We also found the dose–response relationship between industry payments and opioid prescriptions. Taken together, these findings provide stronger evidence of the relationship between the receipt of industry payments for opioid products and physicians’ prescription of opioids.

Our findings suggest that industry payments to physicians for opioids may be one of the key upstream determinants of the current opioid overdose crisis. Previous studies reported that opioid dependence is commonly triggered by initial exposure to opioid prescriptions by physicians.25 27 Moreover, the financial relationships between physicians and industry have been shown to affect their prescribing practices even through meals.28 29 This is also the case for opioids, and it is possible that opioid-related industry marketing may potentially counter national efforts to curb the overprescribing of opioids. To the extent that these relationships might be causal, our work suggests that further interventions to regulate industry–physician relationships for opioids may be an effective strategy to reduce inappropriate opioid prescriptions.

Although we found no evidence that the association of industry payments with opioid prescriptions varied by physicians’ sex, year in practice and medical school graduated, we found a larger difference in opioid prescriptions among primary care providers compared with other specialists. A prior work has found that adherence to clinical guidelines for pain management is suboptimal among general internists,30 and therefore, they might prescribe more opioids from the desire to help patients after all previous interventions fail. Of note, given that primary care physicians account for almost half of total opioid prescriptions in the US.,31 our study suggests that industry marketing targeting clinical practice on such physicians should be immediately addressed. We also observed the larger difference among physicians who did not receive opioid-related industry payments in 2015 compared with physicians who received, indicating that the initial impact of industry marketing could be larger than the secondary impact. Our dose–response findings suggested that even a small amount of industry payments was sufficient to effectively affect physicians’ prescription practice of opioids. Given the shape of the association between payments and prescriptions after adjusting for the number of opioid prescriptions in the prior year, the contact with the industry itself, rather than the actual size of the industry payments, may have a direct impact on physicians’ practice patterns regarding the prescription of opioids in the subsequent year.

These findings advance our current state of knowledge about the influence of opioid drug manufacturers on physicians’ daily clinical practice. Previous studies have reported that opioid-related industry payments to physicians were associated with subsequently increased prescription of opioids and deaths from opioid overdoses.11 13 15 However, these studies did not employ robust causal design with insufficient adjustment for physician-level confounders. Given the potential association of such physician characteristics with both industry payments and prescription behaviour,17 19 32 their lack of data could potentially lead to biased estimates. For example, a recent study reported the association between lower ranked medical school graduated and higher opioid prescriptions32; lack of this information might induce positive confounding (ie, bias away from the null) if medical school graduated is also associated with the receipt of industry payments in the same direction. Moreover, we adjusted for factors related to the physician–industry financial relationship in the prior year. As these factors are strongly associated with subsequent payments and prescriptions of opioids, failing to adjust for them also induced bias away from the null. While one study used a difference-in-difference (DID) design,12 it is likely that physicians who received industry payments (the treatment group) were different in a meaningful way from physicians who did not receive payments (the control group), which could not be sufficiently accounted for using this study design. In addition, given that this study only included 1–2 years of data for the pre-intervention period, whether the ‘parallel trends assumption’ of the DID hold was unclear (ie, it is possible that they were underpowered to detect meaningful differences in pre-intervention trends between the treatment and control groups).33 Finally, most of the existing studies used Open Payments data published during 2013–2015, and the accuracy of the dataset has improved after 2014 due to several efforts performed by the CMS20; for example, (i) they encouraged all physicians to review and dispute incorrect information before publication, (ii) they undertook audits to improve data integrity and corroborate additional elements of the reported data and (iii) they undertook outreach and education activities to increase awareness of the Open Payments Program. Our findings based on more updated national databases combined with a set of physician characteristics, along with a robust study design to test the causality, provide more robust evidence about the impact of industry marketing on opioid prescriptions.

Limitations of the study

Our study has limitations. First, given the nature of the observational design, the differences in outcomes could be affected by unmeasured confounding (not included in our analyses), even though we attempted to address this by including physician characteristics in our models. Second, the value of payments listed in the Open Payments may not be completely accurate. Although the validity of the Open Payments data has improved over time,20 we cannot deny the possibility that pharmaceutical industries under-report the value of payments to physicians. Third, due to the lack of detailed information, we could not evaluate the impact of industry marketing to physicians on their prescription pattern in the form of research payments or ownership interests. Given that general payments account for the largest share of industry payments to physicians, this limitation is unlikely to change our findings. Last, this study did not include physicians who had no Medicare claims, and therefore, our findings may not be generalizable to physicians not caring for Medicare beneficiaries.

CONCLUSIONS

Using nationally representative databases of industry payments to physicians, physician characteristics and drug utilisation by physicians, we found that pharmaceutical industry marketing of opioid products to physicians was associated with higher utilisation of opioids in clinical practice. These findings support the current efforts to improve transparency and may suggest the potential impact of the opioid-related industry payments on physicians’ prescription of opioids in the US. Future investigations are warranted to understand how industry payments for opioids influence the number of opioid overdoses, and whether the prescriptions of opioids are largely driven by physicians’ encouragement to prescribe opioids or patients’ preference for pain management.

What is already known on this topic

  • Industry marketing to physicians for opioids has received substantial attention as it can potentially influence physicians’ prescription of opioids.

  • Existing studies are simple observational studies or ecological studies; therefore, the causal impact of physicians’ financial relationships with industry on opioid prescription remains unclear.

  • Moreover, little is known about the dose–response relationship of industry payments for opioids.

What this study adds

  • Using propensity- score matching, we found that industry payments to physicians for opioid products were associated with a larger number of opioid prescriptions, and higher expenditures for opioid products, in the subsequent year.

  • The association was larger among primary care physicians than surgeons or specialists.

  • Even a small amount of industry payments for opioids might be sufficient to effectively affect physicians’ prescription practice of opioids.

REFERENCES

Footnotes

  • Contributors All authors had full access to all of the data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis. KI, JF and YT contributed to the design and conduct of the study, data collection and management and analysis and interpretation of the data; and preparation, review or approval of the manuscript. EJO contributed to the analysis and interpretation of the data and preparation, review or approval of the manuscript.

  • Funding KI was supported by the Burroughs Wellcome Fund Interschool Training Program in Chronic Diseases (BWF-CHIP) and Honjo International Scholarship Foundation.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethical approval The study was exempted from human subjects review by the institutional review board at University of California, Los Angeles.

  • Data sharing statement All data are publicly available: CMS.gov (https://www.cms.gov/openpayments), Medicare.gov (https://www.medicare.gov/physiciancompare) and CMS.gov (https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Medicare-Provider-Charge-Data/Part-D-Prescriber.html).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.