This study found extensive variation in general internal medicine physician prices and that high-priced physicians provided fewer low-value services but had higher spending on these services.
Objectives: To assess the cross-sectional relationship between prices paid to physicians by commercial insurers and the provision of low-value services.
Study Design: Observational study design using Health Care Cost Institute claims representing 3 large national commercial insurers.
Methods: The main outcome was count of 19 potential low-value services in 2014. The secondary outcome was total spending on the low-value services. Independent variables of interest were price quintiles based on each physician’s mean geographically adjusted price of a mid-level office visit, the most commonly billed service by general internal medicine (GIM) physicians. We estimated the association between physician price quintile and provision of low-value services via negative binomial or generalized linear models with adjustments for measure, region, and patient and physician characteristics.
Results: This study included 750,452 commercially insured patients attributed to 28,951 GIM physicians. In 2014, the mean geographically adjusted price for physicians in the highest price quintile was $122.6 vs $54.7 for physicians in the lowest quintile ($67.9 difference; 95% CI, $67.5-$68.3). Relative to patients attributed to the lowest-priced physicians, those attributed to the highest-priced physicians received 3.6, or 22.9%, fewer low-value services per 100 patients (95% CI, 2.7-4.7 services per 100 patients). Spending on low-value services attributed to the highest-priced physicians was 10.9% higher ($520 difference per 100 patients; 95% CI, $167-$872).
Conclusions: Commercially insured patients of high-priced physicians received fewer low-value services, although spending on low-value services was higher. More research is needed to understand why high-priced providers deliver fewer low-value services and whether physician prices are correlated with other measures of quality.
Am J Manag Care. 2022;28(5):e178-e184. https://doi.org/10.37765/ajmc.2022.89149
This study found extensive variation in general internal medicine physician price, and that high-priced physicians provided fewer low-value services but had higher spending on these services.
Despite large variation across physicians in the prices paid by commercial insurers, it is unclear whether higher prices are associated with higher quality of care.1-4 Policy makers have become concerned that prices may be primarily associated with a provider’s negotiating leverage rather than underlying care quality.5,6 As researchers increasingly gain access to commercial claims data sets that report negotiated rates between payers and providers, a limited literature has emerged documenting whether physician quality is associated with price and generally finds a weak relationship.7,8 These studies have focused on a number of well-established quality measures, such as ambulatory care–sensitive hospital admissions and patient Consumer Assessment of Healthcare Providers and Systems survey items. However, quality is multidimensional, necessitating the use of a range of measures. Further, many quality measures require adjustments for patient risk, and this is often imperfect in cross-sectional claims analysis. This study adds to the limited literature that investigates the relationship between physician price and quality of care using another dimension of quality that is less likely to be influenced by patient risk: the provision of low-value services.
Low-value services are services that are thought to provide minimal clinical benefit and, importantly, should rarely be provided to certain types of patients regardless of their clinical risk status.9 Based on the Choosing Wisely campaign and other national guidelines, prior studies have constructed measures of low-value services.10,11 Previous work has identified physician and organizational characteristics associated with low-value service provision. One study found physician characteristics such as age, sex, academic degree, and patient panel size to be weakly associated with low-value services.12 Other studies have identified organizational characteristics associated with more low-value care, including hospital-based rather than office-based care.13,14 To our knowledge, no prior study has examined the relationship between physician prices and the provision of low-value services. We explore this question for general internal medicine (GIM) physicians, who account for the most common primary care specialty.
Data and Sample
We used the 2011-2014 Health Care Cost Institute (HCCI) database, which contains the universe of claims for 3 of the largest national commercial health insurers covering more than 50 million lives across all states. Critically, HCCI data include the allowed amount paid to providers for each service. Physician and practice characteristics were identified by linking HCCI claims to the Medicare Data on Provider Practice and Specialty file and the IQVIA Healthcare Organization Services database using National Provider Identifiers (NPIs). The American Community Survey was merged with claims using patient zip codes to identify the characteristics of zip codes in which patients resided.
We constructed low-value services for 2014 because the publicly available definitions for identifying low-value services (described below) used International Classification of Diseases, Ninth Revision codes and definitions, as International Classification of Diseases, Tenth Revision updates had not been published. The sample was limited to patients aged 18 to 64 years who were continuously enrolled in a preferred provider organization or point-of-service plan in 2013-2014. Patients were attributed to physicians, regardless of specialty, using the logic developed by CMS for the Value-Based Modifier program (eAppendix Methods 1 [eAppendix available at ajmc.com]).15 To increase the accuracy of patient attribution, we limited the sample to patients with at least 3 claims for evaluation and management (E&M) visits rather than 1. Finally, the sample was limited to patients attributed to a GIM physician (excluding hospitalists).
The primary outcome was the count of low-value services that a patient received, where a patient could receive multiple types of low-value services and multiple low-value services of the same type. Medicare claims-based definitions that identified 26 low-value services were adapted for use with HCCI claims.10 The services were identified as having little clinical benefit by the American Board of Internal Medicine Foundation’s Choosing Wisely initiative, the US Preventive Services Task Force D-graded recommendations, the National Institute for Health and Care Excellence “do not do” recommendations, the Canadian Agency for Drugs and Technologies in Health technology assessments, or peer-reviewed medical literature. Of the original 26 services, we identified 19 that GIM physicians would likely provide directly or refer to another clinician to provide. The final set of 19 measures was similar to that used in a recent study that identified 17 low-value services for Medicare beneficiaries.12
A secondary outcome was total spending on all low-value care services per patient. The low-value care spending outcome included geographically adjusted spending for all 19 low-value services for a patient in 2014 and was constructed by updating a previously developed algorithm (eAppendix Methods 2).10
We also constructed count and spending outcomes for each category of low-value service, which included imaging, preoperative services, and diagnostic testing. For robustness, we included a narrower set of 7 low-value services measures likely to be directly ordered by a GIM physician, as well as a limited set of high-value services, which were 3 diabetic high-value services defined by the National Committee for Quality Assurance’s Healthcare Effectiveness Data and Information Set program. (Measurement specifications for low-value and high-value services are detailed in eAppendix Tables 1 and 2.)
The main variable of interest was a geographically adjusted measure of physician prices.8 The measure was based on prices paid (the allowed amounts) for a mid-level office visit, Current Procedural Terminology (CPT) code 99213. The allowed amount included the amount paid by the insurer and any coinsurance paid by the patient. This is the most commonly billed service by GIM physicians in HCCI data and its price is highly correlated with other E&M service prices (eAppendix Methods 3). Each physician’s mean CPT 99213 price—including applicable facility fees for 0.4% of claims—was geographically adjusted at the county level and winsorized at the top and bottom 5% of the distribution to mitigate the influence of outliers (eAppendix Methods 3). We then divided physicians into quintiles, based on the adjusted price for each physician. Indicators for price quintiles were included in all analyses.
The primary regression model did not include any patient and physician covariates because low-value services should not be provided irrespective of patient risk. In sensitivity analyses, regression models included patient and physician characteristics. We also included a practice characteristic, physician group size, in descriptive statistics. However, we did not include group size in the main regression analyses. An important determinant of price is a practice’s negotiating leverage, which is determined in part by group size. Our primary research question was whether a physician’s price, irrespective of why a physician receives this price, is associated with the provision of low-value services.
Patient characteristics included risk scores, age groups (18-24, 25-34, 35-44, 45-54, and 55-64 years), gender, and zip code–level median household income by quartile. Risk scores were based on the HHS Hierarchical Condition Category risk adjustment model, which is designed specifically for use with commercial claims from the concurrent year.16 Physician characteristics included age and gender and indicators for foreign medical training and attending a top-25 medical school in the United States. The single practice characteristic was a measure of physician group size (number of unique NPIs billing a physician’s primary tax identification number).
All regressions were at the patient level. Negative binomial regression models were used to estimate the relationship between physician price quintiles and the number of low-value services received because the count outcome was overdispersed (likelihood ratio test for α = 0 was P < .001 for all models). A general linear model (GLM) with a log link and gamma distribution was used to estimate the relationship between physician price quintiles and a patient’s total spending on low-value services. All models included indicators for hospital referral region (HRR) and measure eligibility. HRR indicators were included to diminish between-HRR effects. Indicators for an individual’s measure eligibility were included to control for differences in patient panel eligibility across physicians. Further, these indicators help account for some patients being eligible for many measures relative to another patient. In sensitivity analyses, the patient and physician characteristics described in the previous section were included. All models included robust standard errors and were adjusted for clustering at the physician level.
Estimates reflect differences in the outcome between the lowest-price quintile and other quintiles and were scaled to represent the difference per 100 patients. Differences are the average marginal effects.17
We also ran the main and secondary specifications using outcomes for each category of low-value service. Robustness analyses included using a quasi-Poisson and ordinary least squares (OLS) model instead of negative binomial models, as well as repeating the main negative binomial count and GLM spending specifications using a narrower set of 7 low-value services measures likely to be directly ordered by a GIM physician. Secondary analyses included a narrow set of 3 diabetic high-value services, as well as the inclusion of group size as a covariate.
Analyses were conducted using SAS version 9.4 (SAS Institute) and Stata/MP version 15.1 (StataCorp).
The final sample included 750,452 patients attributed to 28,951 GIM physicians in 2014 (eAppendix Figure 1). The mean adjusted price paid to physicians in the highest quintile for a mid-level office visit was more than double that for physicians in the lowest price quintile ($122.6 vs $54.7; $67.9 difference; 95% CI, $67.5-$68.3) (eAppendix Table 3).
The characteristics of patients attributed to the highest-priced and lowest-priced physicians were similar. For example, the difference in risk scores was less than 2% (highest vs lowest price quintile: 0.99 vs 0.98; P = .007) (Table 1). Physician and practice characteristics varied substantially across physician price quintiles (Table 2). Relative to the lowest-priced physicians, those receiving the highest prices were younger, more likely to be female, and more likely to have graduated from a top-25 medical school. Additionally, the highest-priced physicians were much more likely to practice in larger medical groups (P < .001, χ2 test).
Low-Value Service Provision
In unadjusted comparisons, patients attributed to the highest-priced physicians relative to those attributed to the lowest-priced physicians received 10.0 fewer (41.5%) low-value services per 100 patients (24.1 vs 14.1 services per 100 patients; P < .001) (Table 1).
Differences attenuated in the primary regression-adjusted results using HRR and measure indicators (Figure 1). Patients attributed to the highest-priced physicians received 3.6 fewer (22.9%) low-value services per 100 patients (P < .001) (Figure 1, panel A). The difference in the number of low-value services received between the lowest price quintile and all other quintiles was statistically significant (P < .001 for all). The magnitude of these differences increased with each higher price quintile, suggesting a dose effect. Estimates were similar after adjusting for both patient and physician characteristics (Figure 1, panels B and C), suggesting that the provision of low-value services for patients eligible for low-value services is not strongly influenced by patient risk.
The divergence in the adjusted and unadjusted results demonstrates the likely importance of the composition of a physician’s panel. For example, some physicians or regions may have a higher portion of patients with hypothyroidism or plantar fasciitis, and these patients are eligible for low-value service measures that have rates 2 to 4 times higher than the average low-value service rate (eAppendix Table 4). The regression analyses controlled for potential compositional differences across measures and regions.
Finally, results were similar when using the count of low-value services by service category (eAppendix Table 6).
Low-Value Service Spending
Unadjusted spending on low-value services was $830 (15%) higher per 100 patients attributed to the highest-priced physicians relative to patients attributed to the lowest-priced physicians ($6211 vs $5381; P < .001) (Table 1). In spending estimates adjusted with HRR and measure indicators, 100 patients attributed to the highest-priced physicians spent on average $520 more (10.9%) on low-value services than 100 patients attributed to the lowest-priced physicians (P = .004) (Figure 2, panel A). The differences for all other quintiles were not significant. Results were similar when adjusting for patient and physician characteristics (Figure 2, panels B and C).
Results were similar when using the spending of low-value services for only imaging services. Results were not significant at the 5% level for preoperative services and in the opposite direction for diagnostic services (eAppendix Table 7).
The results of sensitivity analyses that used OLS and quasi-Poisson models and used a narrower set of low-value services measures yielded results consistent with our primary analyses (eAppendix Tables 11 through 14). In our secondary analysis examining a limited number of diabetic high-value services, patients attributed to the highest-priced physicians received 12.7 additional (5.8%) high-value services per 100 patients (P < .001) (eAppendix Table 15). In our secondary analysis adjusting for group size, spending results were similar (eAppendix Table 16). Results for the count of low-value services were smaller in magnitude and not statistically significant; coefficients associated with group size were highly significant (eAppendix Table 17).
In this study of the relationship between the prices paid to GIM physicians and their provision of low-value services, insurers paid physicians in the highest price quintile prices that were more than double those paid to physicians in the lowest price quintile ($122.6 vs $54.7). In adjusted analyses, patients attributed to the highest-priced physicians received nearly one-fourth fewer low-value services compared with patients attributed to the lowest-priced physicians (12.1 vs 15.7 low-value services per 100 patients); this difference was driven by reductions in all categories of low-value services.
Although patients attributed to the highest-priced physicians were less likely to receive a low-value service, total spending per 100 patients on these services was more than 10% higher ($5266 vs $4746). However, only spending on imaging services drove the increase in low-value spending.
We found that higher-priced physicians were much more likely to be part of larger physician groups and, in our secondary analysis adjusting for group size, group size rather than a physician’s price quintile was significantly associated with the provision of low-value services. This suggests that larger groups, which likely have more leverage to negotiate higher prices, deliver fewer low-value services. Large groups may have more resources to invest in information technology infrastructure and quality improvement activities that affect the provision of low-value services. Alternatively, physicians who provide fewer low-value services could be more likely to join larger groups. Future work should explore why high-priced providers deliver fewer low-value services. Identifying the underlying mechanisms may aid organizational and policy maker efforts to promote the reduction of low-value services.
One explanation for the finding that higher prices are associated with fewer low-value services could be the so-called target income hypothesis (ie, low-priced physicians provide more services overall to make up for the fact that they receive a lower price per service).18 This hypothesis suggests that the low-priced physicians would provide more of all services irrespective of whether the services are of high, intermediary, or low value. However, our analysis of a limited number of diabetic high-value services does not support this explanation. We find the opposite relationship in the high-value setting—high prices were associated with slightly more diabetic high-value services—suggesting that high-priced physicians are better able to distinguish between high-value and low-value services than low-priced physicians and provide more of one and fewer of the other.
Total spending on low-value services, driven by imaging services, was higher for patients attributed to high-priced physicians, even though they provided a lower volume of low-value services. As an example, for physicians with a panel of 100 patients, a high-priced physician provided 3.8 fewer services relative to a low-priced physician, but those fewer services cost $520 or more than 10% more. The higher cost could be due to the higher prices that primary care physicians receive or due to higher-priced primary care physicians referring to higher-priced providers, inside or outside their own practice.19 Higher spending only for imaging services suggests that high-priced physicians are referring their patients to higher-priced imaging facilities, as documented in other studies.20 As provider groups enter new contracts that incentivize higher-quality care at lower costs, practices may find that their higher prices and the higher prices of their referral network prevent them from reducing overall spending even when providing services with more value.
These results add to the mixed findings of the only 2 previous studies on the relationship between prices and quality of health care. One study found that higher practice prices were associated with better quality for a few process-of-care measures, including higher vaccination rates, but were not associated with patient ratings of care and ratings of their physicians.7 The second study found that higher physician prices were associated with higher costs but not with higher quality as measured by 30-day readmissions and potentially avoidable hospitalizations.8
This work also adds to previous limited literature on the provision of low-value services that has identified a limited number of physician and practice characteristics associated with low-value care. One study found that physician characteristics explained little of the variation in the provision of low-value services, but it did not examine physician prices.12 The same researchers found modest consistency within a provider organization in the provision of low-value care across specific services and over time.14 A final study found that practice location (community based vs hospital based), but not practice ownership, was associated with use of low-value care.13 Our study findings suggest that the price a physician receives, which is typically accepted or negotiated with insurers at a practice or provider organization level, is correlated with the provision of low-value services.
This study has 5 main limitations. First, our measures of low-value services do not encompass all aspects of quality. Second, our results are based on 1 year of data in 2014. However, previous research using the same commercial database found similar rates of low-value services over time.11 Third, our results are based on cross-sectional analysis and therefore should not be interpreted as causal. Fourth, it is possible that some of the low-value services provided were provided by specialist physicians rather than the GIM physician. Our sensitivity analysis focused on services likely to be provided only by GIM physicians produced similar results. Finally, our results may not be generalizable to other commercial insurers. Further, although requiring patients to have at least 3 E&M visits adds confidence on the attribution of care to a physician, this may limit generalizability to patients who interact with the health system frequently.
We found extensive variation in prices paid to GIM physicians and that high-priced physicians provided fewer low-value services but had higher spending on low-value services. More research is needed to understand why high-priced providers deliver fewer low-value services and whether physician prices are correlated with other measures of quality.
Preliminary results from this study were presented at the International Health Economics Conference in Basel, Switzerland, in July 2019. The authors would like to thank Dr Aaron L. Schwartz for sharing his thoughts and prior work on measuring low-value service spending. The authors acknowledge the assistance of the Health Care Cost Institute and its data contributors, Aetna, Humana, and UnitedHealthcare, in providing the claims data analyzed in this study.
Author Affiliations: Division of Health Policy and Economics, Department of Population Health Sciences, Weill Cornell Medical College (AMB, YZ, MZ, PJ, MAU, LPC), New York, NY; Department of Internal Medicine, Montefiore Medical Center (FT), New York, NY; Health Policy and Management Department, Yale University (YQ), New Haven, CT.
Source of Funding: This research was supported by the Physicians Foundation Center for the Study of Physician Practice and Leadership at Weill Cornell Medical College. The Foundation had no role in the selection of the research topic, the design of the project, analysis of the data, or preparation of the manuscript.
Author Disclosures: The authors report 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 (AMB, YZ, FT, MZ, MAU, LPC); acquisition of data (AMB, FT, MZ, YQ, LPC); analysis and interpretation of data (AMB, YZ, MZ, PJ, MAU, LPC); drafting of the manuscript (AMB, MAU); critical revision of the manuscript for important intellectual content (YZ, PJ, MAU, LPC); statistical analysis (AMB, YZ, PJ); provision of patients or study materials (FT, YQ); obtaining funding (LPC); and administrative, technical, or logistic support (YQ).
Address Correspondence to: Amelia M. Bond, PhD, Division of Health Policy and Economics, Department of Population Health Sciences, Weill Cornell Medical College, 402 E 67th St, New York, NY 10065. Email: email@example.com.
1. Baker LC, Bundorf MK, Royalty AB, Levin Z. Physician practice competition and prices paid by private insurers for office visits. JAMA. 2014;312(16):1653-1662. doi:10.1001/jama.2014.10921
2. Capps C, Dranove D, Ody C. Physician practice consolidation driven by small acquisitions, so antitrust agencies have few tools to intervene. Health Aff (Millwood). 2017;36(9):1556-1563. doi:10.1377/hlthaff.2017.0054
3. Chernew ME, Hicks AL, Shah SA. Wide state-level variation in commercial health care prices suggests uneven impact of price regulation. Health Aff (Millwood). 2020;39(5):791-799. doi:10.1377/hlthaff.2019.01377
4. Pelech D. An analysis of private-sector prices for physicians’ services. Congressional Budget Office. January 12, 2018. Accessed March 4, 2021. https://www.cbo.gov/publication/53441
5. FTC to study the impact of physician group and healthcare facility mergers. News release; Federal Trade Commission. January 14, 2021. Accessed July 9, 2021. https://www.ftc.gov/news-events/press-releases/2021/01/ftc-study-impact-physician-group-healthcare-facility-mergers
6. Gudiksen KL, Montague AD, King JS. Who can rein in health care prices? state and federal efforts to address health care provider consolidation. Milbank Memorial Fund. June 24, 2021. Accessed July 26, 2021. https://www.milbank.org/publications/who-can-rein-in-health-care-prices-state-and-federal-efforts-to-address-health-care-provider-consolidation/
7. Roberts ET, Mehrotra A, McWilliams JM. High-price and low-price physician practices do not differ significantly on care quality or efficiency. Health Aff (Millwood). 2017;36(5):855-864. doi:10.1377/hlthaff.2016.1266
8. Unruh MA, Zhang Y, Jung HY, et al. Physician prices and the cost and quality of care for commercially insured patients. Health Aff (Millwood). 2020;39(5):800-808. doi:10.1377/hlthaff.2019.00237
9. Van Parys J, Skinner J. Physician practice style variation—implications for policy. JAMA Intern Med. 2016;176(10):1549-1550. doi:10.1001/jamainternmed.2016.4433
10. Schwartz AL, Landon BE, Elshaug AG, Chernew ME, McWilliams JM. Measuring low-value care in Medicare. JAMA Intern Med. 2014;174(7):1067-1076. doi:10.1001/jamainternmed.2014.1541
11. Colla CH, Morden NE, Sequist TD, Mainor AJ, Li Z, Rosenthal MB. Payer type and low-value care: comparing Choosing Wisely services across commercial and Medicare populations. Health Serv Res. 2018;53(2):730-746. doi:10.1111/1475-6773.12665
12. 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
13. Mafi JN, Wee CC, Davis RB, Landon BE. Association of primary care practice location and ownership with the provision of low-value care in the United States. JAMA Intern Med. 2017;177(6):838-845. doi:10.1001/jamainternmed.2017.0410
14. Schwartz AL, Zaslavsky AM, Landon BE, Chernew ME, McWilliams JM. Low-value service use in provider organizations. Health Serv Res. 2018;53(1):87-119. doi:10.1111/1475-6773.12597
15. Two-step attribution for measures included in the value modifier. CMS. August 2015. Accessed December 11, 2019. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/PhysicianFeedbackProgram/Downloads/Attribution-Fact-Sheet.pdf
16. Kautter J, Pope GC, Ingber M, et al. The HHS-HCC risk adjustment model for individual and small group markets under the Affordable Care Act. Medicare Medicaid Res Rev. 2014;4(3):mmrr2014-004-03-a03. doi:10.5600/mmrr2014-004-03-a03
17. Williams R. Using the margins command to estimate and interpret adjusted predictions and marginal effects. Stata J. 2012;12(2):308-331. doi:10.1177/1536867X1201200209
18. Mcguire TG. Chapter 9: physician agency. In: Culyer AJ, Newhouse JP, eds. Handbook of Health Economics. Elsevier; 2000:461-536. doi:10.1016/S1574-0064(00)80168-7
19. Mehrotra A, Huckfeldt PJ, Haviland AM, Gascue L, Sood N. Patients who choose primary care physicians based on low office visit price can realize broader savings. Health Aff (Millwood). 2016;35(12):2319-2326. doi:10.1377/hlthaff.2016.0408
20. Chernew M, Cooper Z, Larsen-Hallock E, Morton FS. Are health care services shoppable? evidence from the consumption of lower-limb MRI scans. National Bureau of Economic Research working paper No. 24869. July 2018. Accessed July 26, 2021. https://www.nber.org/papers/w24869