[Skip to Content]
Access to paid content on this site is currently suspended due to excessive activity being detected from your IP address 54.156.39.245. Please contact the publisher to request reinstatement.
[Skip to Content Landing]
Table 1.  
Methods for Defining Service Use and Eligible Cohort for Each Nonrecommended Servicea
Methods for Defining Service Use and Eligible Cohort for Each Nonrecommended Servicea
Table 2.  
Cohort Characteristics
Cohort Characteristics
Table 3.  
Results From Models to Assess Variation in Physician Service Use
Results From Models to Assess Variation in Physician Service Use
Table 4.  
Imaging for Staging in Breast Cancer by Perceived Level of Appropriateness of Service Received by the Prior Patient
Imaging for Staging in Breast Cancer by Perceived Level of Appropriateness of Service Received by the Prior Patient
Table 5.  
Sub-Group Analysis Modeling Likelihood of a Patient Receiving Nonrecommended Service Given Physician Ownership or Setting of Care Deliverya
Sub-Group Analysis Modeling Likelihood of a Patient Receiving Nonrecommended Service Given Physician Ownership or Setting of Care Deliverya
1.
Brody  H.  Medicine’s ethical responsibility for health care reform—the top five list.  N Engl J Med. 2010;362(4):283-285.PubMedGoogle ScholarCrossref
2.
Orszag  PR. The Overuse, Underuse, and Misuse of Health Care. Testimony before the Committee on Finance, United States Senate. Washington, DC: Congressional Budget Office; July 17, 2008.
3.
Emanuel  E, Tanden  N, Altman  S,  et al.  A systemic approach to containing health care spending.  N Engl J Med. 2012;367(10):949-954.PubMedGoogle ScholarCrossref
4.
ABIM Foundation. Choosing Wisely. 2012; http://www.choosingwisely.org. Accessed April, 2012.
5.
Fisher  ES, Wennberg  DE, Stukel  TA, Gottlieb  DJ, Lucas  FL, Pinder  EL.  The implications of regional variations in Medicare spending. Part 1: the content, quality, and accessibility of care.  Ann Intern Med. 2003;138(4):273-287.PubMedGoogle ScholarCrossref
6.
Feinstein  AJ, Soulos  PR, Long  JB,  et al.  Variation in receipt of radiation therapy after breast-conserving surgery: assessing the impact of physicians and geographic regions.  Med Care. 2013;51(4):330-338.PubMedGoogle ScholarCrossref
7.
Fung  V, Schmittdiel  JA, Fireman  B,  et al.  Meaningful variation in performance.  Med Care. 2010;48(2):140-148.PubMedGoogle ScholarCrossref
8.
Wennberg  JE, Freeman  JL, Shelton  RM, Bubolz  TA.  Hospital use and mortality among Medicare beneficiaries in Boston and New Haven.  N Engl J Med. 1989;321(17):1168-1173.PubMedGoogle ScholarCrossref
9.
Makarov  DV, Soulos  PR, Gold  HT,  et al.  Regional-level correlations in inappropriate imaging rates for prostate and breast cancers: potential implications for the Choosing Wisely campaign.  JAMA Oncol. 2015;1(2):185-194.PubMedGoogle ScholarCrossref
10.
Makarov  DV, Desai  RA, Yu  JB,  et al.  The population level prevalence and correlates of appropriate and inappropriate imaging to stage incident prostate cancer in the medicare population.  J Urol. 2012;187(1):97-102.PubMedGoogle ScholarCrossref
11.
Newhouse  JP, Garber  AM, Graham  RP, McCoy  MA, Mancher  M, Kibria  A. Variation in Health Care Spending: Target Decision Making, Not Geography. Washington, DC: National Academy Press; Institute of Medicine; 2013.
12.
Obermeyer  Z, Powers  BW, Makar  M, Keating  NL, Cutler  DM.  Physician characteristics strongly predict patient enrollment in hospice.  Health Aff (Millwood). 2015;34(6):993-1000.PubMedGoogle ScholarCrossref
13.
Schnipper  LE, Smith  TJ, Raghavan  D,  et al.  American Society of Clinical Oncology identifies five key opportunities to improve care and reduce costs.  J Clin Oncol. 2012;30(14):1715-1724.PubMedGoogle ScholarCrossref
14.
Prasad  SM, Gu  X, Lipsitz  SR, Nguyen  PL, Hu  JC.  Inappropriate utilization of radiographic imaging in men with newly diagnosed prostate cancer in the United States.  Cancer. 2012;118(5):1260-1267.PubMedGoogle ScholarCrossref
15.
Choi  WW, Williams  SB, Gu  X, Lipsitz  SR, Nguyen  PL, Hu  JC.  Overuse of imaging for staging low risk prostate cancer.  J Urol. 2011;185(5):1645-1649.PubMedGoogle ScholarCrossref
16.
Crivello  ML, Ruth  K, Sigurdson  ER,  et al.  Advanced imaging modalities in early stage breast cancer: preoperative use in the United States Medicare population.  Ann Surg Oncol. 2013;20(1):102-110.PubMedGoogle ScholarCrossref
17.
Keating  NL, Landrum  MB, Guadagnoli  E, Winer  EP, Ayanian  JZ.  Surveillance testing among survivors of early-stage breast cancer.  J Clin Oncol. 2007;25(9):1074-1081.PubMedGoogle ScholarCrossref
18.
Panageas  KS, Sima  CS, Liberman  L, Schrag  D.  Use of high technology imaging for surveillance of early stage breast cancer.  Breast Cancer Res Treat. 2012;131(2):663-670.PubMedGoogle ScholarCrossref
19.
Jagsi  R, Falchook  AD, Hendrix  LH, Curry  H, Chen  RC.  Adoption of hypofractionated radiation therapy for breast cancer after publication of randomized trials.  Int J Radiat Oncol Biol Phys. 2014;90(5):1001-1009.PubMedGoogle ScholarCrossref
20.
Hahn  C, Kavanagh  B, Bhatnagar  A,  et al.  Choosing wisely: the American Society for Radiation Oncology’s top 5 list.  Pract Radiat Oncol. 2014;4(6):349-355.PubMedGoogle ScholarCrossref
21.
Bekelman  JE, Epstein  AJ, Emanuel  EJ.  Single- vs multiple-fraction radiotherapy for bone metastases from prostate cancer.  JAMA. 2013;310(14):1501-1502.PubMedGoogle ScholarCrossref
22.
Warren  JL, Klabunde  CN, Schrag  D, Bach  PB, Riley  GF.  Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population.  Med Care. 2002;40(8)(suppl):IV-3-IV-18.PubMedGoogle Scholar
23.
Klabunde  CN, Potosky  AL, Legler  JM, Warren  JL.  Development of a comorbidity index using physician claims data.  J Clin Epidemiol. 2000;53(12):1258-1267.PubMedGoogle ScholarCrossref
24.
Romano  PS, Roos  LL, Jollis  JG.  Adapting a clinical comorbidity index for use with ICD-9-CM administrative data.  J Clin Epidemiol. 1993;46(10):1075-1079.PubMedGoogle ScholarCrossref
25.
Normand  SLT, Glickman  ME, Gatsonis  CA.  Statistical methods for profiling providers of medical care: issues and applications.  J Am Stat Assoc. 1997;92(439):803-814.Google ScholarCrossref
26.
Shahian  DM, Normand  SL, Torchiana  DF,  et al.  Cardiac surgery report cards.  Ann Thorac Surg. 2001;72(6):2155-2168.PubMedGoogle ScholarCrossref
27.
Goldstein  H, Spiegelhalter  DJ.  League tables and their limitations.  J R Stat Soc Ser A Stat Soc. 1996;159:385-409.Google ScholarCrossref
28.
Bratzler  DW, Normand  SL, Wang  Y,  et al.  An administrative claims model for profiling hospital 30-day mortality rates for pneumonia patients.  PLoS One. 2011;6(4):e17401.PubMedGoogle ScholarCrossref
29.
Snijders  TAB, Bosker  RJ.  Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. London, England: Sage Publications; 1999.
30.
Smith  TJ, Hillner  BE.  Bending the cost curve in cancer care.  N Engl J Med. 2011;364(21):2060-2065.PubMedGoogle ScholarCrossref
31.
Mariotto  AB, Yabroff  KR, Shao  Y, Feuer  EJ, Brown  ML.  Projections of the cost of cancer care in the United States: 2010-2020.  J Natl Cancer Inst. 2011;103(2):117-128.PubMedGoogle ScholarCrossref
32.
Institute of Medicine.  Delivering High-Quality Cancer Care: Charting a New Course for a System in Crisis. Washington, DC: The National Academies Press; 2013.
33.
Bach  PB.  Limits on Medicare’s ability to control rising spending on cancer drugs.  N Engl J Med. 2009;360(6):626-633.PubMedGoogle ScholarCrossref
34.
Elkin  EB, Bach  PB.  Cancer’s next frontier.  JAMA. 2010;303(11):1086-1087.PubMedGoogle ScholarCrossref
35.
Quill  CM, Ratcliffe  SJ, Harhay  MO, Halpern  SD.  Variation in decisions to forgo life-sustaining therapies in US ICUs.  Chest. 2014;146(3):573-582.PubMedGoogle ScholarCrossref
36.
Barnato  AE, Herndon  MB, Anthony  DL,  et al.  Are regional variations in end-of-life care intensity explained by patient preferences?  Med Care. 2007;45(5):386-393.PubMedGoogle ScholarCrossref
37.
Lipitz-Snyderman  A, Elkin  EB, Atoria  CL,  et al.  Provider differences in use of implanted ports in older adults with cancer.  Med Care. 2015;53(7):646-652.PubMedGoogle ScholarCrossref
38.
Godin  G, Bélanger-Gravel  A, Eccles  M, Grimshaw  J.  Healthcare professionals’ intentions and behaviours: a systematic review of studies based on social cognitive theories.  Implement Sci. 2008;3:36.PubMedGoogle Scholar
39.
Michie  S, Johnston  M, Abraham  C, Lawton  R, Parker  D, Walker  A; “Psychological Theory” Group.  Making psychological theory useful for implementing evidence based practice: a consensus approach.  Qual Saf Health Care. 2005;14(1):26-33.PubMedGoogle ScholarCrossref
40.
Eisenberg  JM.  Physician utilization: the state of research about physicians’ practice patterns.  Med Care. 1985;23(5):461-483.PubMedGoogle ScholarCrossref
41.
Ross  I, Womble  P, Ye  J,  et al.  MUSIC: patterns of care in the radiographic staging of men with newly diagnosed low risk prostate cancer.  J Urol. 2015;193(4):1159-1162.PubMedGoogle ScholarCrossref
Original Investigation
Less Is More
October 2016

Physician-Driven Variation in Nonrecommended Services Among Older Adults Diagnosed With Cancer

Author Affiliations
  • 1Center for Health Policy and Outcomes, Memorial Sloan Kettering Cancer Center, New York, New York
  • 2Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York
  • 3Genentech, California
  • 4Department of Surgery, Memorial Sloan Kettering Cancer Center, New York, New York
  • 5Department of Urology, Columbia University, New York, New York
  • 6Department of Medicine, Memorial Sloan Kettering Cancer Center, New York, New York
  • 7Department of Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, New York
  • 8Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, New York
 

Copyright 2016 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.

JAMA Intern Med. 2016;176(10):1541-1548. doi:10.1001/jamainternmed.2016.4426
Abstract

Importance  Interventions to address overuse of health care services may help reduce costs and improve care. Understanding physician-level variation and behavior patterns can inform such interventions.

Objective  To assess patterns of physician ordering of services that tend to be overused in the treatment of patients with cancer. We hypothesized that physicians exhibit consistent behavior.

Design, Setting, and Participants  Retrospective study of patients 66 years and older diagnosed with cancer between 2004 and 2011, using population-based Surveillance, Epidemiology, and End Results (SEER)-Medicare data to assess physician-level variation in 5 nonrecommended services. Services included imaging for staging and surveillance in low-risk disease, intensity-modulated radiation therapy (IMRT) after breast-conserving surgery, and extended fractionation schemes for palliation of bone metastases.

Main Outcome and Measures  To assess variation in service use between physicians, we used a random effects model and a logistic regression model with a lag variable to assess whether a physician’s use of a service for a prior patient predicts subsequent service use.

Results  Cohorts ranged from 3464 to 89 006 patients. The total proportion of patients receiving each service varied from 14% for imaging in staging early breast cancer to 41% in early prostate cancer. From the random effects analysis, we found significant unexplained variation in service use between physicians (P < .001 for each service; ICC, 0.04-0.59). Controlling for case mix, whether a physician ordered a service for the prior patient was highly predictive of service use, with adjusted odds ratios (aORs) ranging from 1.12 (95% CI, 1.07-1.18) for surveillance imaging for patients with breast cancer (28% service use if prior patient had imaging vs 25% if not), to 24.91 (95% CI, 22.86-27.15) for IMRT for whole breast radiotherapy (69% vs 7%, respectively).

Conclusions and Relevance  Physicians’ utilization of nonrecommended services that tend to be overused exhibit patterns that suggest consistent behavior more than personalized patient care decisions. Reducing overuse may require understanding cognitive drivers of repetitive inappropriate decisions.

Introduction

There is widespread interest in reducing the overuse of harmful and costly health care services.1-3 Recent initiatives, including the American Board of Internal Medicine (ABIM) Foundation’s Choosing Wisely campaign, have compiled lists of scenarios in which health care services are often overused and should be questioned by patients and their physicians.4 The recommendations largely reflect established evidence or clinical guidelines. Beyond publicizing these lists, effective approaches to reducing overuse have remained elusive but are needed to help improve the overall quality of care.

Cancer care is a prime target for interventions to reduce overuse because it is highly complex and involves an intense period of potentially harmful and costly medical care. Prior studies5-10 have demonstrated wide utilization of nonrecommended services, including for cancer care, and extensive variation by geographic region, practice setting, and individual physician. At least 1 prior study found that, at the regional level, rates of nonrecommended imaging for disease staging were correlated between early stage prostate cancer and early stage breast cancer. This suggests that nonpatient factors influence physician ordering behavior.9 As described in the recent Institute of Medicine report11 on variation in health care spending, there is likely to be considerable variation within the geographic region. Studies11 have found variation by practice setting and even between individual physicians in the same practice. These levels of organization may be more amenable to interventions than regions.

Given the extensive use of many nonrecommended services at the population level, attention to physician-level practice may inform interventions to address overuse. If physicians’ use of a service under specific circumstances is consistent and independent of patient factors, it suggests a focus on physician-specific interventions to reduce overuse. Supporting this notion, Obermeyer and colleagues12 demonstrated differences in hospice utilization at the physician level. The proportion of a physician’s patient panel that was not enrolled in hospice prior to death strongly predicted whether or not his or her patient would receive hospice care. Alternatively, if physicians’ ordering behavior is random across patients, implementing interventions that focus on reducing the use of specific services across all physicians may be more effective.

Our objective was to assess whether physicians exhibit consistent patterns of behavior regarding health care services that are known to be overused in cancer care. We examined physicians’ behavior for 5 services among physicians treating patients with different diseases.

Box Section Ref ID

Key Points

  • Question What is the extent of physician variation for health care services that tend to be overused in cancer care?

  • Findings In this study using population-based Surveillance, Epidemiology, and End Results (SEER) Medicare data, significant unexplained variation between physicians was found. Whether a physician ordered a service for a patient was highly predictive of whether the physician would order the same service for the next patient.

  • Meaning Physicians’ utilization of nonrecommended services exhibit patterns that suggest consistent behavior more than personalized patient care decisions.

Methods

We assessed physician-level variation in nonrecommended service use using a population-based sample of older adult patients with cancer.

Services

We selected examples of nonrecommended services that could be readily identified in health insurance claims. We used recommendations from the Choosing Wisely campaign proposed by the American Society of Clinical Oncology (ASCO) and the American Society for Radiation Oncology (ASTRO). Services included: imaging for staging of early prostate or breast cancer at low risk for metastasis; surveillance imaging for asymptomatic individuals treated for low-risk breast cancer,10,13-18 intensity-modulated radiation therapy (IMRT) for whole breast radiotherapy,19,20 and extended fractionation schemes (>10 fractions) for palliation of bone metastases20,21 (Table 1).

Data Source

The population-based Surveillance, Epidemiology, and End Results (SEER)-Medicare database links 2 sources of data: SEER cancer registry files and Medicare enrollment information and claims. The National Cancer Institute (NCI)-sponsored SEER program includes several US states and regions, covering about 28% of the US population.22 Surveillance, Epidemiology, and End Results registries collect information on site and extent of disease, first course of cancer-directed therapy, and sociodemographic characteristics, with active follow-up for date and cause of death. For older adults diagnosed with cancer in a SEER region, SEER records have been linked with Medicare claims. The SEER-Medicare population has a similar age and sex distribution, a slightly higher proportion of people living in urban and high income areas, and a smaller proportion of nonwhite individuals compared with the US older adult population.22

Study Sample and Time Period

We included patients 66 years and older diagnosed with cancer between 2004 and 2011, with follow-up through 2012. We excluded patients enrolled in an health maintenance organization (HMO) or with incomplete Medicare coverage during the study period, and those diagnosed with cancer on death. We also excluded patients whose physician had only 1 patient. We applied additional inclusion and exclusion criteria for each cohort based on measure definitions used in the literature or provided in the Choosing Wisely campaign (Table 1). For example, for imaging in early breast cancer staging, we included women with stage 0 to II breast cancer as the first primary breast cancer diagnosis. They had to have received a mastectomy or breast conserving surgery within 6 months of diagnosis and not received neoadjuvant chemotherapy.

Variables and Outcomes

Patient demographic, disease, and treatment characteristics from SEER and Medicare records were defined as applicable to the service. Demographic characteristics included: age; race; a modified Charlson comorbidity index based on all inpatient, outpatient, and physician claims from 1 year prior to diagnosis23,24; marital status; median income of census tract of residence; geographic region; and metropolitan location. Disease and treatment characteristics included: cancer stage; tumor characteristics; year of diagnosis; surgery; radiation therapy; and chemotherapy.

For each cohort, we identified use of the service within the defined study time period. For the example of imaging for breast cancer staging, we identified claims for positron emission tomography (PET), computed tomography (CT), and bone scans from diagnosis through surgery (Table 1).

Physician Assignment

Eligible patients were assigned to the physician most likely to be the primary decision maker for the service in question, based on an algorithm involving physician specialty and frequency of visits (Table 1). All physicians had to have a Unique Physician Identification Number (UPIN) or a National Provider Identifier (NPI). A crosswalk provided by NCI was used to link physicians over time. In the example of imaging for breast cancer staging, we assigned accountability for each patient to the surgeon associated with her curative treatment. Surgeons’ patients were sorted chronologically by surgery date.

Statistical Analysis

We examined the proportion of eligible patients who received each service by patient, disease, and treatment characteristics. We also assessed the use of these services among physicians.

Our primary objective was to better understand the extent of physician variation in practice. To meet our objective, we assessed the extent of unexplained variation in service use between physicians using a nonlinear random effects model with logit link function. This model estimates the logit probability of service use as a linear function of case mix and a random effect for physician assigned to the service.25-28 It assumes that differences in underlying physician patterns lead to differences in proportions of service use. A test of whether the estimated variance of the random effects differs significantly from zero provides an indication of significant unexplained variation between physicians. We also calculated the intraclass correlation (ICC) which provides information on the magnitude of the correlation.29

If we found evidence of physician variation using the random effects model, we thought we would more finely assess the magnitude of the correlation between patients. Therefore, we also performed a logistic regression analysis with a lag variable. This model would allow us to examine the likelihood that a physician’s patient received a service, given that the physician’s previous patient also received that service. This model does not include physicians’ first patient because we required each patient to have a lag value. We hypothesized that information on the physician’s prior patient would be informative and reflect physician practice, and trend in a similar direction as the findings from the random effects model. The parameter estimate of the lag variable would offer another and potentially more interpretable measure of the magnitude of physician variation.

For both approaches, services were analyzed separately. In adjusted models, we controlled for case mix. The individual patient encounter was the unit of analysis.

For imaging related to staging breast cancer, some physicians might consider imaging more appropriate for patients with stage II disease than for patients with stage I or 0 disease, even though the Choosing Wisely recommendation does not make this distinction.13 We therefore repeated the analysis, and assigned a perceived level of indication to the physician’s ordering of the test for the prior patient: indicated (ie, prior patient did not receive test); potentially indicated (ie, prior patient received test and had stage II disease); and not indicated (ie, prior patient received test and had stage 0 or 1 disease).

We repeated the imaging analyses accounting for physician ownership of different types of imaging machines. We also conducted separate analyses for CT scan and PET scan. Similarly, for radiation therapy, we repeated the analyses accounting for setting: outpatient hospital facility or standalone facility.

Analyses were performed in SAS statistical software (version 9.2, SAS Institute). A 2-sided P value of less than .05 was considered to be statistically significant. The NCI approved the use of the SEER-Medicare database for this study, which was deemed exempt research by the Memorial Sloan Kettering Cancer Center institutional review board.

Results

The number of patients in each cohort varied from 3464 for the assessment of extended fractionation schemes for palliation of bone metastases to 89 006 for imaging in early breast cancer staging (Table 2). Between 695 and 4285 physicians were included in the analyses. Across services, the median number of patients assigned to each physician ranged from 3 to 11 (Table 2).

Fourteen percent of patients received imaging in breast cancer staging, 18% IMRT for whole breast radiotherapy, 26% posttreatment surveillance in low-risk breast cancer, 35% extended fractionation schemes for palliation of bone metastases, and 41% imaging in early prostate cancer staging (Table 3). The proportions differed by patient characteristics, including age and geographic region (eTable 1 in the Supplement).

From the random effects analysis, there was significant unexplained variation in service use between physicians, accounting for case mix (P < .001 for each service) (Table 3). The ICCs ranged from 0.04 for surveillance imaging for breast cancer to 0.59 for IMRT for whole breast radiotherapy.

Since we found significant physician variation based on the random effects model, we also conducted the logistic regression model using the lag variable. The likelihood of a patient receiving a nonrecommended service differed by whether or not the physician’s prior patient received the service (eTable 1 in the Supplement). For staging early prostate cancer, 60% of patients received imaging if the prior patient also received imaging compared with 26% if the prior patient did not. Respectively, 28% vs 12% for imaging in breast cancer staging, 28% vs 25% for imaging for posttreatment surveillance in breast cancer, 69% vs 7% for IMRT for whole breast radiotherapy, and 41% vs 32% for extended fractionation schemes for bone metastases. Across all services, the odds of an individual patient receiving a service were higher if the physician’s prior patient had received the service than if the physician’s prior patient had not (Table 3). Controlling for case mix did not change the interpretation of our main effect in any of the examples. The adjusted odds of a patient receiving the service ranged from 1.12 (95% CI, 1.07-1.18) for surveillance imaging for patients with breast cancer, to 24.91 (95% CI, 22.86-27.15) for IMRT to deliver whole breast radiotherapy (Table 3; see eTable 2 in the Supplement for full results).

For the example of imaging for breast cancer staging, the adjusted odds of a woman receiving an imaging test were approximately 3 times higher if the physician’s prior patient also had the test (OR, 3.02; 95% CI, 2.88-3.17). The odds of a patient receiving the test increased by the increasing level of inappropriateness of the prior patient’s receipt of the test, from 2.58 (95% CI, 2.41-2.76) if the prior patient received the test and had stage II disease up to 3.40 (95% CI, 3.21-3.61) if the prior patient received the test and had stage 0 or 1 disease (Table 4).

The addition of ownership or setting of care as covariates did not change the interpretation of the main effect in any model. Ownership of a CT scanner was associated with increased likelihood of CT scan use in the context of prostate cancer staging and breast cancer surveillance. Ownership of a PET scanner was associated with use of a PET scan for breast cancer surveillance. Radiation therapy delivered in a standalone facility compared to a hospital outpatient facility was associated with increased use of IMRT, but not with extended fractionation for bone metastases (Table 5).

Discussion

Contemporary paradigms of clinical practice emphasize that patient care should be individualized to the patient; but consistent behavior by physicians would tend not to be. Under this premise, we assessed the extent of physician variation in use of nonrecommended services. We used a random effects model and a logistic regression model with a lag variable to assess whether the likelihood of a patient’s receipt of an overused service was associated with whether the physician had ordered that service for the prior patient. We observed extensive physician-level variation. This observation held across different services, and for physicians treating patients with different diseases. Overutilization in cancer care may be physician-specific.

Our results were robust to patient characteristics that predicted service use, as well as physician ownership and setting of care delivery. For imaging for staging women with low-risk breast cancer, the strength of the relationship between current and prior use increased with the degree of inappropriateness of the service use for the prior patient.

These findings raise questions about the degree to which marginal decisions, such as those that are a focus of Choosing Wisely, are currently influenced by patient preferences and patient-directed decision making. The campaign specifically focuses on educating patients to question their physicians about the need for overused services.4 For our results to reflect the service use preferences of patients rather than physicians, patients would have to select physicians who would satisfy their preferences for the receipt of specific services. The ability of patients to select physicians in this way is unlikely given the types of services we are examining and today’s relatively opaque health care system.

We observed a wide range of ORs across our service-specific analyses, but all findings were in the hypothesized direction. We might have seen a small effect for surveillance imaging after treatment for breast cancer because there was inaccurate physician assignment given the long follow-up period, symptom-driven imaging was misclassified as inappropriate, or the true effect is small. We saw a very large effect in the context of IMRT for whole-breast radiotherapy. Given that IMRT is widely available across the nation, we can be confident that our observation of IMRT use in breast cancer does not reflect availability of the technology in certain facilities. Possible explanations for our observations include clinical indication not reflected in the current data set, physician preference, patient desire, and financial concerns.

Oncology has been touted as an important area for overuse reduction efforts given its complexity, costs, and the uncertainty surrounding decision making.13,30-34 Accounting for patient and disease characteristics and preferences, several studies12,35,36 have documented unexplained variation in practice, which may be attributable to the physician or practice setting. Using claims, we were able to observe physicians’ consistent interpatient behavior in practice for several services that reflect different types of care and contexts. Our analysis provides evidence of physicians’ consistent behavior with regard to overused services in oncology. We observed similar findings in our prior study37 examining another example of a likely overused therapeutic service, implanted infusion ports in patients receiving chemotherapy. Our results are also consistent with evidence of patterns of hospice referral.12

The magnitude of the effects we observed raises an important question: to what extent is the observed behavior driven by physicians’ intentional choices vs reflexive, ingrained tendencies? We still have an incomplete understanding of the mechanisms driving these decisions and the most effective physician-level strategies to address them. Habit and past behavior might influence physicians’ intentions and resulting consistent behaviors in this context.38 Factors such as beliefs about evidence based practices or guidelines, knowledge, skills, self-efficacy, anticipated consequences of the decision, and motivation and goals might also inform physicians’ decisions.39 A deeper understanding of drivers behind physician decision making is needed to help change professional practice. In any case, our approach offers an intuitive way to investigate and quantify the consistent and repetitive nature of physicians’ behavior at the population level.

Limitations

Findings should be interpreted in the context of its limitations. First, we likely did not include the full patient panel for each physician. However, patients 66 years and older represent a large proportion of patients with the cancer types studied. Second, the decision of whether to order a test may have been made by multiple health care providers. However, assigning the decision to the wrong physician would result in an underestimate of the true association. Third, since our sample size is large, statistically significant results may not imply clinical significance. Finally, based on claims alone, we cannot observe the indications for each service. For surveillance imaging following breast cancer treatment, patient-reported symptoms may prompt testing. It is also relevant for radiation for palliation of bone metastases, since we cannot be certain whether the radiation was intended to be palliative or curative, or the metastases site. Following the recommendations, we expect that the majority of service use was not clinically indicated, and cases for which it was would be randomly distributed and cannot explain our results.

Conclusions

Efforts to address overuse should consider the possibility that it is concentrated among physicians who practice in repetitive ways, relatively uninfluenced by the patient. Ensuring accessible, affordable cancer care was cited as integral to achieving high-quality cancer care by the Institute of Medicine.32 The physician practice patterns we observed support intervening with physicians who tend to be consistent high users, which can lead to a better understanding of why these behaviors persist.40 Likewise, there can be learning opportunities from more measured users. Quality measurement activities such as feedback, profiling, and pay-for-performance may be useful, as demonstrated in initiatives such as the MUSIC collaborative to reduce imaging for staging in low-risk prostate cancer.41 Physician-focused interventions can supplement existing education and dissemination efforts to help reduce overuse in oncology and improve patient outcomes.

Findings provide support that physicians’ use of nonrecommended services in the context of cancer care is consistent and repetitive in nature. The interpatient effects are striking in the scenarios included in this study. Reducing overuse may require understanding the cognitive drivers of repetitive inappropriate decisions.

Back to top
Article Information

Corresponding Author: Allison Lipitz-Snyderman, PhD, Center for Health Policy and Outcomes, Memorial Sloan Kettering Cancer Center, 485 Lexington Ave, Second Floor, New York, NY 10017 (snyderma@mskcc.org).

Accepted for Publication: June 14, 2016.

Published Online: August 15, 2016. doi:10.1001/jamainternmed.2016.4426

Author Contributions: Drs Lipitz-Snyderman and Bach, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Lipitz-Snyderman, Sima, Elkin, Anderson, Blinder, Tsai, Bach.

Acquisition, analysis, or interpretation of data: Lipitz-Snyderman, Sima, Atoria, Elkin, Panageas.

Drafting of the manuscript: Lipitz-Snyderman, Sima, Panageas.

Critical revision of the manuscript for important intellectual content: Sima, Atoria, Elkin, Anderson, Blinder, Tsai, Panageas, Bach.

Statistical analysis: Lipitz-Snyderman, Sima, Atoria, Panageas.

Obtained funding: Lipitz-Snyderman.

Study supervision: Anderson, Bach.

Other: Tsai, Elkin.

Conflict of Interest Disclosures: Dr Sima reports employment with Genentech. Dr Blinder reports consulting with Pfizer. Dr Bach reports personal fees from Association of Community Cancer Centers, personal fees from America's Health Insurance Plans, personal fees from AIM Specialty Health, personal fees from American College of Chest Physicians, personal fees from American Society of Clinical Oncology, personal fees from BARCLAYS, personal fees from Defined Health, personal fees from Express Scripts, personal fees from Genentech, personal fees from Goldman Sachs, personal fees from McKinsey and Company, personal fees from MPM Capital, personal fees from National Comprehensive Cancer Network, personal fees from Biotechnology Industry Organization, personal fees from The American Journal of Managed Care, personal fees from The Boston Consulting Group, personal fees from Foundation Medicine, personal fees from Anthem Inc, outside the submitted work.

Funding/Support: This study was supported by the National Institute for Health Care Management (NIHCM) Foundation, Cancer Center Support Grant (P30 CA 008748) to Memorial Sloan Kettering Cancer Center.

Role of the Funder/Sponsor: The National Institute for Health Care Management had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

References
1.
Brody  H.  Medicine’s ethical responsibility for health care reform—the top five list.  N Engl J Med. 2010;362(4):283-285.PubMedGoogle ScholarCrossref
2.
Orszag  PR. The Overuse, Underuse, and Misuse of Health Care. Testimony before the Committee on Finance, United States Senate. Washington, DC: Congressional Budget Office; July 17, 2008.
3.
Emanuel  E, Tanden  N, Altman  S,  et al.  A systemic approach to containing health care spending.  N Engl J Med. 2012;367(10):949-954.PubMedGoogle ScholarCrossref
4.
ABIM Foundation. Choosing Wisely. 2012; http://www.choosingwisely.org. Accessed April, 2012.
5.
Fisher  ES, Wennberg  DE, Stukel  TA, Gottlieb  DJ, Lucas  FL, Pinder  EL.  The implications of regional variations in Medicare spending. Part 1: the content, quality, and accessibility of care.  Ann Intern Med. 2003;138(4):273-287.PubMedGoogle ScholarCrossref
6.
Feinstein  AJ, Soulos  PR, Long  JB,  et al.  Variation in receipt of radiation therapy after breast-conserving surgery: assessing the impact of physicians and geographic regions.  Med Care. 2013;51(4):330-338.PubMedGoogle ScholarCrossref
7.
Fung  V, Schmittdiel  JA, Fireman  B,  et al.  Meaningful variation in performance.  Med Care. 2010;48(2):140-148.PubMedGoogle ScholarCrossref
8.
Wennberg  JE, Freeman  JL, Shelton  RM, Bubolz  TA.  Hospital use and mortality among Medicare beneficiaries in Boston and New Haven.  N Engl J Med. 1989;321(17):1168-1173.PubMedGoogle ScholarCrossref
9.
Makarov  DV, Soulos  PR, Gold  HT,  et al.  Regional-level correlations in inappropriate imaging rates for prostate and breast cancers: potential implications for the Choosing Wisely campaign.  JAMA Oncol. 2015;1(2):185-194.PubMedGoogle ScholarCrossref
10.
Makarov  DV, Desai  RA, Yu  JB,  et al.  The population level prevalence and correlates of appropriate and inappropriate imaging to stage incident prostate cancer in the medicare population.  J Urol. 2012;187(1):97-102.PubMedGoogle ScholarCrossref
11.
Newhouse  JP, Garber  AM, Graham  RP, McCoy  MA, Mancher  M, Kibria  A. Variation in Health Care Spending: Target Decision Making, Not Geography. Washington, DC: National Academy Press; Institute of Medicine; 2013.
12.
Obermeyer  Z, Powers  BW, Makar  M, Keating  NL, Cutler  DM.  Physician characteristics strongly predict patient enrollment in hospice.  Health Aff (Millwood). 2015;34(6):993-1000.PubMedGoogle ScholarCrossref
13.
Schnipper  LE, Smith  TJ, Raghavan  D,  et al.  American Society of Clinical Oncology identifies five key opportunities to improve care and reduce costs.  J Clin Oncol. 2012;30(14):1715-1724.PubMedGoogle ScholarCrossref
14.
Prasad  SM, Gu  X, Lipsitz  SR, Nguyen  PL, Hu  JC.  Inappropriate utilization of radiographic imaging in men with newly diagnosed prostate cancer in the United States.  Cancer. 2012;118(5):1260-1267.PubMedGoogle ScholarCrossref
15.
Choi  WW, Williams  SB, Gu  X, Lipsitz  SR, Nguyen  PL, Hu  JC.  Overuse of imaging for staging low risk prostate cancer.  J Urol. 2011;185(5):1645-1649.PubMedGoogle ScholarCrossref
16.
Crivello  ML, Ruth  K, Sigurdson  ER,  et al.  Advanced imaging modalities in early stage breast cancer: preoperative use in the United States Medicare population.  Ann Surg Oncol. 2013;20(1):102-110.PubMedGoogle ScholarCrossref
17.
Keating  NL, Landrum  MB, Guadagnoli  E, Winer  EP, Ayanian  JZ.  Surveillance testing among survivors of early-stage breast cancer.  J Clin Oncol. 2007;25(9):1074-1081.PubMedGoogle ScholarCrossref
18.
Panageas  KS, Sima  CS, Liberman  L, Schrag  D.  Use of high technology imaging for surveillance of early stage breast cancer.  Breast Cancer Res Treat. 2012;131(2):663-670.PubMedGoogle ScholarCrossref
19.
Jagsi  R, Falchook  AD, Hendrix  LH, Curry  H, Chen  RC.  Adoption of hypofractionated radiation therapy for breast cancer after publication of randomized trials.  Int J Radiat Oncol Biol Phys. 2014;90(5):1001-1009.PubMedGoogle ScholarCrossref
20.
Hahn  C, Kavanagh  B, Bhatnagar  A,  et al.  Choosing wisely: the American Society for Radiation Oncology’s top 5 list.  Pract Radiat Oncol. 2014;4(6):349-355.PubMedGoogle ScholarCrossref
21.
Bekelman  JE, Epstein  AJ, Emanuel  EJ.  Single- vs multiple-fraction radiotherapy for bone metastases from prostate cancer.  JAMA. 2013;310(14):1501-1502.PubMedGoogle ScholarCrossref
22.
Warren  JL, Klabunde  CN, Schrag  D, Bach  PB, Riley  GF.  Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population.  Med Care. 2002;40(8)(suppl):IV-3-IV-18.PubMedGoogle Scholar
23.
Klabunde  CN, Potosky  AL, Legler  JM, Warren  JL.  Development of a comorbidity index using physician claims data.  J Clin Epidemiol. 2000;53(12):1258-1267.PubMedGoogle ScholarCrossref
24.
Romano  PS, Roos  LL, Jollis  JG.  Adapting a clinical comorbidity index for use with ICD-9-CM administrative data.  J Clin Epidemiol. 1993;46(10):1075-1079.PubMedGoogle ScholarCrossref
25.
Normand  SLT, Glickman  ME, Gatsonis  CA.  Statistical methods for profiling providers of medical care: issues and applications.  J Am Stat Assoc. 1997;92(439):803-814.Google ScholarCrossref
26.
Shahian  DM, Normand  SL, Torchiana  DF,  et al.  Cardiac surgery report cards.  Ann Thorac Surg. 2001;72(6):2155-2168.PubMedGoogle ScholarCrossref
27.
Goldstein  H, Spiegelhalter  DJ.  League tables and their limitations.  J R Stat Soc Ser A Stat Soc. 1996;159:385-409.Google ScholarCrossref
28.
Bratzler  DW, Normand  SL, Wang  Y,  et al.  An administrative claims model for profiling hospital 30-day mortality rates for pneumonia patients.  PLoS One. 2011;6(4):e17401.PubMedGoogle ScholarCrossref
29.
Snijders  TAB, Bosker  RJ.  Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. London, England: Sage Publications; 1999.
30.
Smith  TJ, Hillner  BE.  Bending the cost curve in cancer care.  N Engl J Med. 2011;364(21):2060-2065.PubMedGoogle ScholarCrossref
31.
Mariotto  AB, Yabroff  KR, Shao  Y, Feuer  EJ, Brown  ML.  Projections of the cost of cancer care in the United States: 2010-2020.  J Natl Cancer Inst. 2011;103(2):117-128.PubMedGoogle ScholarCrossref
32.
Institute of Medicine.  Delivering High-Quality Cancer Care: Charting a New Course for a System in Crisis. Washington, DC: The National Academies Press; 2013.
33.
Bach  PB.  Limits on Medicare’s ability to control rising spending on cancer drugs.  N Engl J Med. 2009;360(6):626-633.PubMedGoogle ScholarCrossref
34.
Elkin  EB, Bach  PB.  Cancer’s next frontier.  JAMA. 2010;303(11):1086-1087.PubMedGoogle ScholarCrossref
35.
Quill  CM, Ratcliffe  SJ, Harhay  MO, Halpern  SD.  Variation in decisions to forgo life-sustaining therapies in US ICUs.  Chest. 2014;146(3):573-582.PubMedGoogle ScholarCrossref
36.
Barnato  AE, Herndon  MB, Anthony  DL,  et al.  Are regional variations in end-of-life care intensity explained by patient preferences?  Med Care. 2007;45(5):386-393.PubMedGoogle ScholarCrossref
37.
Lipitz-Snyderman  A, Elkin  EB, Atoria  CL,  et al.  Provider differences in use of implanted ports in older adults with cancer.  Med Care. 2015;53(7):646-652.PubMedGoogle ScholarCrossref
38.
Godin  G, Bélanger-Gravel  A, Eccles  M, Grimshaw  J.  Healthcare professionals’ intentions and behaviours: a systematic review of studies based on social cognitive theories.  Implement Sci. 2008;3:36.PubMedGoogle Scholar
39.
Michie  S, Johnston  M, Abraham  C, Lawton  R, Parker  D, Walker  A; “Psychological Theory” Group.  Making psychological theory useful for implementing evidence based practice: a consensus approach.  Qual Saf Health Care. 2005;14(1):26-33.PubMedGoogle ScholarCrossref
40.
Eisenberg  JM.  Physician utilization: the state of research about physicians’ practice patterns.  Med Care. 1985;23(5):461-483.PubMedGoogle ScholarCrossref
41.
Ross  I, Womble  P, Ye  J,  et al.  MUSIC: patterns of care in the radiographic staging of men with newly diagnosed low risk prostate cancer.  J Urol. 2015;193(4):1159-1162.PubMedGoogle ScholarCrossref
×