Figure 1. Cohort identification. The first box represents all cases of cancer identified within the data set for each disease site. The second box depicts the cohort for each disease site after general restrictions. The final box represents the cohort for each guideline following guideline-specific criteria. These guideline-specific cohorts are not mutually exclusive and may overlap because more than 1 guideline may be applicable to a given patient. ESRD indicates end-stage renal disease; AXD, axillary dissection; XRT, radiation therapy; PMRT, postmastectomy radiation therapy; TT, total thyroidectomy; and CND, central neck dissection.
Figure 2. Hospital distribution of 100% concordance for each guideline, depicting the percentage of hospitals with 5 or more cases and providing guideline-concordant care to 100% of eligible patients treated. BCS indicates breast-conserving surgery; ER, estrogen receptor; IBC, invasive breast cancer; PMRT, postmastectomy radiation therapy; and XRT, radiation therapy. (See the “Identification of Existing Clinical Practice Guidelines” subsection of the “Methods” section for explanation of the grades.)
Greenberg CC, Lipsitz SR, Neville B, In H, Hevelone N, Porter SA, Weeks C, Jha AK, Gawande AA, Schrag D, Weeks JC. Receipt of Appropriate Surgical Care for Medicare Beneficiaries With Cancer. Arch Surg. 2011;146(10):1128-1134. doi:10.1001/archsurg.2011.141
Author Affiliations: Center for Surgery and Public Health, Department of Surgery, Brigham and Women's Hospital (Drs Greenberg, Lipsitz, In, C. Weeks, and Gawande and Mr Hevelone), Center for Outcomes and Policy Research, Department of Medical Oncology, Dana-Farber Cancer Institute (Drs Greenberg, In, Schrag, and J. C. Weeks and Ms Neville), and Department of Health Policy and Management, Harvard School of Public Health (Drs Jha and Gawande), Boston, Massachusetts. Ms Porter is a medical student at Harvard Medical School, Boston.
Objective To investigate receipt of appropriate surgical care in Medicare beneficiaries with cancer.
Design Retrospective cohort study.
Setting National Surveillance, Epidemiology, and End Results registry linked to Medicare claims data.
Patients Fee-for-service Medicare patients aged 65 years or older who underwent a definitive surgical resection for breast, colon, gastric, rectal, or thyroid cancer diagnosed between January 2000 and December 2005. Claims data were available from January 1999 through December 2007.
Main Outcome Measures Receipt of care concordant with established practice guidelines in surgical oncology in the aggregate and by hospital.
Results Concordance with guidelines was greater than 90% for 7 of 11 measures. All guidelines regarding adjuvant therapy had concordance rates greater than 90%. Only 2 of 5 measures for nodal management had concordance rates greater than 90%. At least 50% of hospitals provided guideline-concordant care to 100% of their patients for 6 of 11 guidelines. Patients receiving appropriate care tended to be younger, healthier, white, and more affluent, to have less advanced disease, and to live in the Midwest.
Conclusions We found a high level of concordance with guidelines in some domains of surgical oncology care but far less so in others, particularly for gastric and colon nodal management. Given the current national focus on improving the quality of health care, surgeons must focus on generating data to define appropriate care and translating those data into everyday practice.
There is currently a major focus on improving the quality of health care in America. Quality health care means delivering the right care to the right patient at the right time. To ensure such care is provided, we must first know what the right or appropriate care is and then be able to determine whether that care was provided. The practice of health care in America varies widely across both institutions and demographic populations; this is true across specialties and disease sites.1,2 This variation may reflect 2 things: (1) a lack of knowledge about the optimal approach to care; or (2) a lack of acceptance regarding currently defined standards of care.
We sought to use practice guidelines to investigate whether appropriate surgical care was provided to Medicare beneficiaries with a new diagnosis of cancer. We first determined the percentage of patients receiving guideline-concordant care at an aggregate patient level. We then analyzed the degree of variation in hospital-level performance to understand whether each guideline has been accepted into routine practice. Finally, we identified patient and institutional factors associated with the likelihood of receiving appropriate care.
We organized our study around the 3 areas in which a surgeon treating cancer must be proficient: (1) surgical management of the primary tumor; (2) evaluation and treatment of regional nodal basins; and (3) appropriate referral for multidisciplinary adjuvant therapy. We first identified existing disease-specific guidelines in each of these areas for 5 common cancers in which surgery plays an important role, including breast, colon, gastric, rectal, and thyroid cancers, based on the following criteria: (1) the guideline must be endorsed by a professional organization or society whose members are considered experts in that disease, addressed in a National Institutes of Health Consensus Statement, or included as a recommendation in the AJCC Cancer Staging Manual3; (2) cancer registry and/or administrative data must be sufficient to define concordance, including both the numerator (patients who received the recommended care) and the denominator (patients meeting inclusion criteria for the recommended care); and (3) all other treatments must be considered inappropriate care.
To classify the strength of a given guideline, we used the classification system published by the National Comprehensive Cancer Network (NCCN).4 The NCCN clinical practice guidelines are readily available, are widely used, span multiple cancer sites, and classify the strength of both the evidence and panel members' consensus supporting a recommendation. The NCCN identifies 4 guideline categories: grade 1 indicates that the recommendation is based on a high level of evidence and there is uniform NCCN consensus; grade 2A, the recommendation is based on a lower level of evidence and there is uniform NCCN consensus; grade 2B, the recommendation is based on a lower level of evidence and there is nonuniform NCCN consensus (but no major disagreement); and grade 3, the recommendation is based on any level of evidence but reflects major disagreement.
The Surveillance, Epidemiology, and End Results (SEER) program of the National Cancer Institute collects detailed data including stage at diagnosis, tumor morphology, first course of treatment, and demographic variables for persons with cancer living in a SEER region.5 The SEER data have been linked to Medicare enrollment and claims data to facilitate evaluation of treatment and outcomes of care across the cancer continuum.6
Patients diagnosed with breast, colon, gastric, rectal, or thyroid cancer as their first or only cancer between January 2000 and December 2005 were identified and linked to Medicare claims data from January 1999 through December 2007. We required that all patients have at least 1 cancer-specific surgical procedure between 30 days before through 365 days after diagnosis and be associated with a nonmissing hospital identifier. Patients were also required to be continuously enrolled in Medicare and have no health maintenance organization enrollment during this period.
Between 2000 and 2005, there were a total of 181 534 patients with breast cancer, 116 383 with colon cancer, 22 625 with gastric cancer, 41 857 with rectal cancer, and 10 453 with thyroid cancer identified. Criteria used to specify the cohort are shown in Figure 1. We excluded patients younger than 65 years at the time of diagnosis and those with end-stage renal disease or disability as their qualifying condition. Patients with stage IV disease, disease of an unknown stage (except for thyroid cancer), or a discrepancy in the death date reported by SEER and Medicare of greater than 3 months were also excluded.
Additional specific inclusion and exclusion criteria for the denominator and numerator for each guideline were applied and are available on request. For situations where there was a discrepancy or ambiguity between identified guidelines regarding appropriate criteria, we performed sensitivity analyses varying the inclusion criteria. Tumor size criteria for receipt of total thyroidectomy were varied between 1 and 2 cm. As conclusions did not change, results are presented only for a threshold of 1.5 cm. Because this study focused specifically on the surgical practitioner, guidelines that related to the receipt of adjuvant therapy were expanded to include a visit with a medical oncologist or radiation oncologist regardless of whether the patient received treatment. Medical oncologists were identified by an associated Medicare specialty code for medical oncology or hematology/oncology or by having submitted a claim for chemotherapy.7- 9 Radiation oncologists were defined by a Medicare specialty code for radiation oncology or by having submitted a claim for radiation planning or administration.
Concordance was measured as the proportion of patients who met the numerator criteria among all patients who met the denominator criteria for each guideline. The concordance rate for each guideline was determined at the aggregate patient level by considering all patients with the relevant condition and at the hospital level by considering all hospitals that treated at least 5 patients with the relevant condition. This was done to ensure stability in the estimate of the hospital concordance rate. While patients treated at hospitals with fewer than 5 patients were excluded from the hospital-level analysis, they were included in the aggregate patient-level analysis. We report aggregate patient-level concordance rates with 95% confidence intervals. Generalized estimating equations were used to account for clustering at the hospital level when calculating 95% confidence intervals.10 For the hospital-level analysis, we determined the proportion of hospitals for which 100% of treated patients received guideline-concordant care as well as exact binomial 95% confidence intervals.
For each of the guidelines meeting our eligibility criteria, logistic regression models were built to investigate covariates associated with receipt of concordant care. Generalized estimating equations were used to estimate the logistic regression parameters, which allowed us to investigate both patient and known institutional factors while accounting for clustering at the institutional level. Because the logistic regression analyses were exploratory, we did not adjust the type I error rate to account for multiple comparisons; thus, P values should be interpreted cautiously. The number of significant associations (P < .05) in these analyses is greater than expected by chance if there were no associations in the data. All tests were 2-tailed. All data management and statistical analyses were performed using SAS version 9.2 statistical software (SAS Institute, Inc, Cary, North Carolina).
Eleven guidelines were identified and are listed in Table 1, showing the original sources, the year they were proposed, and whether the NCCN grades them as based on a high level of evidence (grade 1) or uniform consensus but a lower level of evidence (grade 2A).4,11- 18 Given that we are using guidelines to represent appropriate care, we did not include any guidelines for which there was nonuniform consensus or major disagreement. The 11 guidelines include 5 for breast cancer, 2 for colon cancer, 1 for gastric cancer, 1 for rectal cancer, and 2 for thyroid cancer. They represent the following domains of surgical oncology care: 1 measure for surgery directed at the primary cancer, 5 measures for nodal management, and 5 measures related to adjuvant therapy. Table 2 provides a description of the cohort for each disease site included in this analysis.
Concordance with guidelines was greater than 90% for 7 of the 11 measures (Table 3). All guidelines regarding evaluation for or receipt of adjuvant therapy had concordance rates greater than 90%. The measures that evaluated breast cancer management had the highest concordance rates. These included radiation therapy or evaluation following breast-conserving surgery (99.2%), chemotherapy or medical oncology evaluation for estrogen receptor–negative breast cancer (98.1%), axillary dissection for node-positive breast cancer (96.7%), postmastectomy radiation therapy or evaluation for patients with more than 4 positive nodes, with positive lymph nodes and a tumor greater than 5 cm, or with stage III cancer (94.9%), and nodal evaluation for invasive breast cancer (91.8%). The lowest concordance rates were seen for central neck dissection for node-positive thyroid papillary cancer (72.7%), colon cancer nodal evaluation of at least 12 nodes (48.5%), and gastric nodal evaluation of at least 15 nodes (32.5%).
There was wide variation in the proportion of hospitals providing uniformly guideline-concordant care, ie, treating 100% of their patients according to the guideline recommendation (Figure 2). Fewer than 1% of institutions met this standard for colon cancer nodal evaluation or central neck dissection, while 93% provided either evaluation for or treatment with radiation for all patients following breast-conserving surgery. More institutions had uniformly concordant care for guidelines concerning adjuvant therapy than for guidelines dealing with surgery directed at the primary cancer or nodal management.
Table 4 summarizes the factors associated with increased or decreased likelihood of receiving concordant care by guideline. For a number of the guidelines, younger age, less aggressive disease, white race, higher income level, being married, and care at a hospital that participates in an oncology group were independently associated with higher guideline concordance. Patients treated in the Midwest were more likely than patients treated in the Northeast to receive guideline-concordant care for all measures; the direction of the effect for the South and West (as compared with the Northeast) varied by measure.
Using practice guidelines to define appropriate care, we found that more than 90% of all patients received recommended care for 7 of the 11 guidelines examined at an aggregate patient level. We also identified 6 measures for which at least half of the institutions were concordant with the guidelines 100% of the time. These high concordance rates suggest that the factors influencing clinical decision making are adequately captured in the current guidelines and that surgeons recognize the importance of these therapies. As a result, we were able to demonstrate that Medicare beneficiaries are highly likely to receive appropriate care and that this finding is consistent across hospitals.
For several of the guidelines related to nodal management, however, concordance rates were low and few hospitals provided appropriate care to all patients. Specifically, we found that published guidelines recommending examination of a minimum number of lymph nodes in colon and gastric cancer have not resulted in routine adoption of this practice in elderly Americans despite the inclusion of these recommendations in the AJCC Cancer Staging Manual. It is possible that despite apparent expert consensus regarding the importance of evaluation of a minimum number of nodes, the lack of definitive evidence supporting a particular threshold has left many health care practitioners (surgeons and/or pathologists) unconvinced. They may not believe that the potential benefits of more extensive lymph node harvests outweigh the added operative risks, perhaps particularly in older patients. The limitations of the evidence base supporting nodal evaluation and the difficulty in defining an appropriate nodal threshold for use in guidelines or quality measurement are well described in 2 recent reviews, a report from the Cochrane Collaboration regarding the extent of lymph node dissection for gastric cancer and a meta-analysis on colon cancer nodal evaluation.19,20
We observed a somewhat different pattern of concordance for central neck dissection for node-positive papillary thyroid cancer. A higher aggregate concordance rate of 71.6% was observed; however, in the hospital-level analysis, no institutions performed neck dissections in all of their patients. Instead, most institutions performed them in approximately 80% of patients. Possible explanations for these findings are that there is agreement that a central neck dissection constitutes appropriate care for most but not all elderly patients with node-positive papillary thyroid cancer and that the factors relevant to selecting patients for the procedure are not adequately captured by the current guideline inclusion criteria, at least in this population of elderly Americans. Such factors might include the presence of a macrometastasis rather than a micrometastasis or nodes detectable preoperatively or at the time of surgery rather than only on postoperative pathology review.
Several other explanations could account for the patterns observed in this study. Prior research on the Hospital Quality Alliance measures has suggested an association between the length of time guidelines have been in place and concordance rates.21 We found high concordance rates and minimal variation for the breast cancer measures, guidelines among the first to be developed in cancer care.
It is also possible that because referral for adjuvant therapy is a dichotomous decision, facilitating both compliance and measurement, it is easier for institutions to achieve consistent and high concordance rates on these measures than on guidelines measuring a continuous outcome such as number of lymph nodes examined. Alternatively or in addition, the fact that referrals result in reimbursement for the health care provider or the health care system while lymph node harvests do not may contribute to the higher concordance we observed for guidelines related to adjuvant therapy.
Finally, there is a suggestion in our data that the level of evidence on which a guideline is based is associated with the level of concordance. Six of the 7 guidelines with concordance rates greater than 90% were based on a high level of evidence, and all of the 4 guidelines with concordance rates lower than 90% were based on a lower level of evidence (Table 3). Similarly, 5 of the 6 guidelines for which more than 50% of institutions provided concordant care to 100% of their patients were graded as 1, while 4 of the 5 guidelines for which fewer than half of the institutions provided fully concordant care were graded as 2A (Figure 2). The role that level of evidence may play in acceptance of guidelines deserves further investigation.
This study has the usual limitations associated with analyses of large national databases, including incomplete capture of cases, loss to follow-up, and missing data, as well as the limitations of claims data such as variation in billing practices and coding inaccuracies. Additionally, because Medicare provides the only consistent and comprehensive national data source on medical services delivered, our analysis was limited to patients aged 65 years and older. A focus on older patients did allow us to study a particularly vulnerable population and one that makes up the majority of patients with gastric and colorectal cancers. However, breast and thyroid cancers are common in younger Americans, so our reliance on Medicare data represents a more significant limitation for these diagnoses.
We found a high level of concordance with guidelines in some domains of surgical oncology care but far less so in others, especially those associated with nodal management. Five of the 6 measures with wide acceptance into practice (as evidenced by the majority of hospitals providing recommended care to 100% of their patients) relate to appropriate referral for or receipt of adjuvant therapy. Given the current national focus on quality in health care, there is increasing pressure to develop measures to determine whether patients are getting appropriate care; however, within the surgical disciplines, there is a paucity of data to support what constitutes appropriate care. It is critical that surgeons focus on generation of the data necessary to inform clinical decision making and promote high-quality surgical care.
Correspondence: Caprice C. Greenberg, MD, MPH, Center for Surgery and Public Health, Brigham and Women's Hospital, One Brigham Circle, 1620 Tremont St, 4-020, Boston, MA 02115 (email@example.com).
Accepted for Publication: March 11, 2011.
Published Online: June 20, 2011. doi:10.1001/archsurg.2011.141
Author Contributions: Dr Greenberg had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Greenberg, Lipsitz, Jha, Gawande, and J. C. Weeks. Acquisition of data: Greenberg, Neville, and C. Weeks. Analysis and interpretation of data: Greenberg, Lipsitz, Neville, In, Hevelone, Porter, Jha, Gawande, Schrag, and J. C. Weeks. Drafting of the manuscript: Greenberg, In, Porter, and C. Weeks. Critical revision of the manuscript for important intellectual content: Greenberg, Lipsitz, Neville, Hevelone, Porter, Jha, Gawande, Schrag, and J. C. Weeks. Statistical analysis: Greenberg, Lipsitz, Neville, In, and Hevelone. Obtained funding: Greenberg and J. C. Weeks. Administrative, technical, and material support: Neville, In, Porter, and C. Weeks. Study supervision: Greenberg and J. C. Weeks.
Financial Disclosure: None reported.
Funding/Support: This work was supported in part by a grant from the American Surgical Association Foundation.
Role of the Sponsors: The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript.
Disclaimer: This study used the linked SEER-Medicare database. The interpretation and reporting of these data are the sole responsibility of the authors.
Additional Contributions: We acknowledge the efforts of the Applied Research Program, National Cancer Institute; the Office of Research, Development and Information, Centers for Medicare and Medicaid Services; Information Management Services, Inc; and the SEER program tumor registries in the creation of the SEER-Medicare database. Scott Regenbogen, MD, MPH, contributed to study conception and design, and Katherine Corso, MPH, assisted with manuscript preparation.