Evaluation of Reliability and Correlations of Quality Measures in Cancer Care | Breast Cancer | JAMA Network Open | JAMA Network
[Skip to Navigation]
Sign In
Figure 1.  Spearman Correlation Coefficients Across Summary Measures at the Practice Level for Each Cancer Type
Spearman Correlation Coefficients Across Summary Measures at the Practice Level for Each Cancer Type

Correlation coefficients for lung cancer were observed among 191 practices with summary process and utilization measures, 191 practices with summary process and end-of-life (EOL) care measures, 191 practices with summary process and survival measures, 375 practices with summary utilization and EOL care measures, 410 practices with summary utilization and survival measures, and 397 practices with summary EOL care and survival measures. Correlations for colorectal cancer were observed among 208 practices with summary process and utilization measures, 87 practices with summary process and EOL care measures, 227 practices with summary process and survival measures, 87 practices with summary utilization and EOL care measures, 251 practices with summary utilization and survival measures, and 87 practices with summary EOL care and survival measures. Correlations for breast cancer were observed among 434 practices with summary process and utilization measures, 23 practices with summary process and EOL care measures, and 23 practices with summary utilization and EOL care measures. NA indicates not applicable.

Figure 2.  Spearman Correlation Coefficients Across Cancer Types at the Practice Level for Each Summary Measure
Spearman Correlation Coefficients Across Cancer Types at the Practice Level for Each Summary Measure

Correlation coefficients among process measures were observed among 171 practices with at least 20 patients with lung cancer and colorectal cancer (CRC), 191 practices with at least 20 patients with lung cancer and breast cancer, and 226 practices with at least 20 patients with breast cancer and CRC. Correlations among utilization measures were observed among 259 practices with at least 20 patients with lung cancer and CRC, 382 practices with at least 20 patients with lung cancer and breast cancer, and 259 practices with at least 20 patients with breast cancer and CRC. Correlations among end-of-life care measures were observed among 87 practices with at least 20 patients with lung cancer and CRC, 23 practices with at least 20 patients with lung cancer and breast cancer, and 23 practices with at least 20 patients with breast cancer and CRC. Correlations among survival measures were observed among 342 practices with at least 20 patients with lung cancer and CRC. NA indicates not applicable.

Table 1.  Reliability of Measures Across Practices
Reliability of Measures Across Practices
Table 2.  Number of Process Measures With an Estimated Reliability of 0.75 or Greater
Number of Process Measures With an Estimated Reliability of 0.75 or Greater
1.
Centers for Medicare & Medicaid Services. Oncology care model. Last updated June 22, 2020. Accessed July 7, 2020. https://innovation.cms.gov/innovation-models/oncology-care
2.
Nardi  EA, McCanney  J, Winckworth-Prejsnar  K,  et al.  Redefining quality measurement in cancer care.   J Natl Compr Canc Netw. 2018;16(5):473-478. doi:10.6004/jnccn.2018.7028 PubMedGoogle ScholarCrossref
3.
Hassett  MJ.  Quality improvement in the era of big data.   J Clin Oncol. 2017;35(28):3178-3180. doi:10.1200/JCO.2017.74.1181 PubMedGoogle ScholarCrossref
4.
Centers for Medicare & Medicaid Services. OCM performance-based payment methodology, version 5.1. Accessed July 7, 2020. https://innovation.cms.gov/files/x/ocm-pp3beyond-pymmeth.pdf
5.
Centers for Medicare & Medicaid Services. Quality payment program: MIPS overview. Accessed July 31, 2020. https://qpp.cms.gov/mips/overview
6.
American Society of Clinical Oncology. QOPI-related measures. Accessed July 7, 2020. https://practice.asco.org/quality-improvement/quality-programs/quality-oncology-practice-initiative/qopi-related-measures
7.
American College of Surgeons. CoC quality of care measures 2020 surveys. Last updated June 1, 2020. Accessed July 7, 2020. https://www.facs.org/quality-programs/cancer/ncdb/qualitymeasurescocweb
8.
National Quality Forum. Measuring performance. Accessed July 7, 2020. https://www.qualityforum.org/Measuring_Performance/Measuring_Performance.aspx
9.
Centers for Medicare & Medicaid Services. Consensus core set: medical oncology measures, version 1.0. Accessed July 31, 2020. https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/QualityMeasures/Downloads/Medical-Oncology-Measures.pdf
10.
Potosky  AL, Riley  GF, Lubitz  JD, Mentnech  RM, Kessler  LG.  Potential for cancer related health services research using a linked Medicare-tumor registry database.   Med Care. 1993;31(8):732-748. doi:10.1097/00005650-199308000-00006 PubMedGoogle ScholarCrossref
11.
National Cancer Institute. SEER data & software. Accessed July 7, 2020. https://seer.cancer.gov/data-software/
12.
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. doi:10.1016/S0895-4356(00)00256-0 PubMedGoogle ScholarCrossref
13.
Hofer  TP, Hayward  RA, Greenfield  S, Wagner  EH, Kaplan  SH, Manning  WG.  The unreliability of individual physician “report cards” for assessing the costs and quality of care of a chronic disease.   JAMA. 1999;281(22):2098-2105. doi:10.1001/jama.281.22.2098 PubMedGoogle ScholarCrossref
14.
Carmines  EG, Zeller  RA. Reliability and validity assessment. In: Lewis-Beck MS, ed. Series: Quantitative Applications in the Social Sciences. Sage Publications; 1979.
15.
Bravo  G, Potvin  L.  Estimating the reliability of continuous measures with Cronbach’s alpha or the intraclass correlation coefficient: toward the integration of two traditions.   J Clin Epidemiol. 1991;44(4-5):381-390. doi:10.1016/0895-4356(91)90076-L PubMedGoogle ScholarCrossref
16.
Zaslavsky  AM.  Statistical issues in reporting quality data: small samples and casemix variation.   Int J Qual Health Care. 2001;13(6):481-488. doi:10.1093/intqhc/13.6.481 PubMedGoogle ScholarCrossref
17.
Adams  JL. The reliability of provider profiling: a tutorial. Prepared for the National Committee for Quality Assurance. RAND Corporation Technical Report Series 2009. Accessed July 7, 2020. https://www.rand.org/content/dam/rand/pubs/technical_reports/2009/RAND_TR653.pdf
18.
Siegel  RL, Miller  KD, Fuchs  HE, Jemal  A.  Cancer Statistics, 2021.   CA Cancer J Clin. 2021;71(1):7-33. doi:10.3322/caac.21654 PubMedGoogle ScholarCrossref
19.
Gondi  S, Wright  AA, Landrum  MB, Zubizarreta  J, Chernew  ME, Keating  NL.  Multimodality cancer care and implications for episode-based payments in cancer.   Am J Manag Care. 2019;25(11):537-538.PubMedGoogle Scholar
20.
Shulman  LN, Palis  BE, McCabe  R,  et al.  Survival as a quality metric of cancer care: use of the National Cancer Data Base to assess hospital performance.   J Oncol Pract. 2018;14(1):e59-e72. doi:10.1200/JOP.2016.020446 PubMedGoogle ScholarCrossref
21.
Panattoni  L, Fedorenko  C, Kreizenbeck  K,  et al.  Lessons from reporting national performance measures in a regional setting: Washington state community cancer care report.   J Oncol Pract. 2018;14(12):e801-e814. doi:10.1200/JOP.18.00410 PubMedGoogle ScholarCrossref
22.
Hutchinson Institute for Cancer Outcomes Research. Community cancer care in Washington state: quality and cost report 2019. Accessed July 7, 2020. https://www.fredhutch.org/content/dam/www/research/institute-networks-ircs/hicor/HICOR-Community-Cancer-Care-Report-2019.pdf
23.
ASCO Practice Central. Quality oncology practice initiative. Accessed July 7, 2020. https://practice.asco.org/quality-improvement/quality-programs/quality-oncology-practice-initiative
24.
mCODE. mCODE: minimal common oncology data elements: the initiative to create a core cancer model and foundational EHR data elements. Accessed July 7, 2020. https://mcodeinitiative.org/
25.
Liu  JB, Huffman  KM, Palis  BE,  et al.  Reliability of the American College of Surgeons Commission on Cancer’s quality of care measures for hospital and surgeon profiling.   J Am Coll Surg. 2017;224(2):180-190.e8. doi:10.1016/j.jamcollsurg.2016.10.053PubMedGoogle ScholarCrossref
26.
Enright  KA, Taback  N, Powis  ML,  et al.  Setting quality improvement priorities for women receiving systemic therapy for early-stage breast cancer by using population-level administrative data.   J Clin Oncol. 2017;35(28):3207-3214. doi:10.1200/JCO.2016.70.7950 PubMedGoogle ScholarCrossref
27.
Weeks  JC, Uno  H, Taback  N,  et al.  Interinstitutional variation in management decisions for treatment of 4 common types of cancer: a multi-institutional cohort study.   Ann Intern Med. 2014;161(1):20-30. doi:10.7326/M13-2231 PubMedGoogle ScholarCrossref
28.
Vos  EL, Koppert  LB, Jager  A, Vrancken Peeters  MTFD, Siesling  S, Lingsma  HF.  From multiple quality indicators of breast cancer care toward hospital variation of a summary measure.   Value Health. 2020;23(9):1200-1209. doi:10.1016/j.jval.2020.05.011 PubMedGoogle ScholarCrossref
29.
Neuss  MN, Malin  JL, Chan  S,  et al.  Measuring the improving quality of outpatient care in medical oncology practices in the United States.   J Clin Oncol. 2013;31(11):1471-1477. doi:10.1200/JCO.2012.43.3300 PubMedGoogle ScholarCrossref
30.
Chernew  ME, Landrum  MB.  Targeted supplemental data collection—addressing the quality-measurement conundrum.   N Engl J Med. 2018;378(11):979-981. doi:10.1056/NEJMp1713834 PubMedGoogle ScholarCrossref
31.
McWilliams  JM. Professionalism revealed: rethinking quality improvement in the wake of a pandemic. Accessed July 30, 2020. https://catalyst.nejm.org/doi/full/10.1056/CAT.20.0226
32.
Centers for Medicare & Medicaid Services. Medicare 2020 Part C & D star ratings technical notes. Last updated October 1, 2019. Accessed July 7, 2020. https://www.cms.gov/Medicare/Prescription-Drug-Coverage/PrescriptionDrugCovGenIn/Downloads/Star-Ratings-Technical-Notes-Oct-10-2019.pdf
33.
Silber  JH, Rosenbaum  PR, Ross  RN,  et al.  Template matching for auditing hospital cost and quality.   Health Serv Res. 2014;49(5):1446-1474. doi:10.1111/1475-6773.12156 PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Original Investigation
    Oncology
    March 22, 2021

    Evaluation of Reliability and Correlations of Quality Measures in Cancer Care

    Author Affiliations
    • 1Department of Health Care Policy, Harvard Medical School, Boston, Massachusetts
    • 2Division of General Internal Medicine, Brigham and Women’s Hospital, Boston, Massachusetts
    • 3Department of Informatics and Analytics, Dana-Farber Cancer Institute, Boston, Massachusetts
    • 4Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts
    • 5Division of Population Sciences, Dana-Farber Cancer Institute, Boston, Massachusetts
    • 6Section of Medical Oncology, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire
    • 7Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts
    • 8Department of Statistics, Harvard Faculty of Arts and Sciences, Cambridge, Massachusetts
    JAMA Netw Open. 2021;4(3):e212474. doi:10.1001/jamanetworkopen.2021.2474
    Key Points

    Question  Can the quality of care be assessed reliably across medical oncology practices, and what are the correlations of different types of practice-level performance across quality measure and cancer types?

    Findings  In this cross-sectional study, registry and claims-based measures of care processes, utilization, end-of-life care, and survival for newly diagnosed fee-for-service Medicare beneficiaries with cancer were limited by small numbers of patients across oncology practices, even after pooling 5 years of data. Measures of care quality had low reliability and had limited to no correlation across measure and cancer types.

    Meaning  Results of this study suggest that additional research is needed to identify reliable quality measures for practice-level assessments of quality.

    Abstract

    Importance  Measurement of the quality of care is important for alternative payment models in oncology, yet the ability to distinguish high-quality from low-quality care across oncology practices remains uncertain.

    Objective  To assess the reliability of cancer care quality measures across oncology practices using registry and claims-based measures of process, utilization, end-of-life (EOL) care, and survival, and to assess the correlations of practice-level performance across measure and cancer types.

    Design, Setting, and Participants  This cross-sectional study used the Surveillance, Epidemiology, and End Results (SEER) Program registry linked to Medicare administrative data to identify individuals with lung cancer, breast cancer, or colorectal cancer (CRC) that was newly diagnosed between January 1, 2011, and December 31, 2015, and who were treated in oncology practices with 20 or more patients. Data were analyzed from January 2018 to December 2020.

    Main Outcomes and Measures  Receipt of guideline-recommended treatment and surveillance, hospitalizations or emergency department visits during 6-month chemotherapy episodes, care intensity in the last month of life, and 12-month survival were measured. Summary measures for each domain in each cohort were calculated. Practice-level rates for each measure were estimated from hierarchical linear models with practice-level random effects; practice-level reliability (reproducibility) for each measure based on the between-measure variance, within-measure variance, and distribution of patients treated in each practice; and correlations of measures across measure and cancer types.

    Results  In this study of SEER registry data linked to Medicare administrative data from 49 715 patients with lung cancer treated in 502 oncology practices, 21 692 with CRC treated in 347 practices, and 52 901 with breast cancer treated in 492 practices, few practices had 20 or more patients who were eligible for most process measures during the 5-year study period. Patients were 65 years or older; approximately 50% of the patients with lung cancer and CRC and all of the patients with breast cancer were women. Most measures had limited variability across practices. Among process measures, 0 of 6 for lung cancer, 0 of 6 for CRC, and 3 of 11 for breast cancer had a practice-level reliability of 0.75 or higher for the median-sized practice. No utilization, EOL care, or survival measure had reliability across practices of 0.75 or higher. Correlations across measure types were low (r ≤ 0.20 for all) except for a correlation between the CRC process and 1-year survival summary measures (r = 0.35; P < .001). Summary process measures had limited or no correlation across lung cancer, breast cancer, and CRC (r ≤ 0.16 for all).

    Conclusions and Relevance  This study found that quality measures were limited by the small numbers of Medicare patients with newly diagnosed cancer treated in oncology practices, even after pooling 5 years of data. Measures had low reliability and had limited to no correlation across measure and cancer types, suggesting the need for research to identify reliable quality measures for practice-level quality assessments.

    Introduction

    Value-based payment models aim to improve care quality, enhance patient experience, and lower spending on care. Payers are developing more models for specialty care; for example, the Centers for Medicare & Medicaid Services (CMS) Oncology Care Model (OCM) focuses on care for patients with cancer who are undergoing chemotherapy.1 The OCM and other value-based payment models typically measure spending and care quality at the practice level. However, in addition to challenges in identifying high-quality oncology care,2,3 relatively little is understood about identifying high-quality care at the practice level. The OCM currently incorporates 6 measures of quality for calculating performance-based payments, including claims-based measures of emergency department (ED) visits and hospice use.4 For oncology practices that do not participate in the OCM or another advanced alternative payment model, the Medicare Access and CHIP Reauthorization Act requires quality reporting through the Merit-based Incentive Payment System.5 National organizations, such as the American Society of Clinical Oncology (ASCO), the American College of Surgeons Commission on Cancer, the National Quality Forum, and CMS, have developed and/or compiled measure sets to assess oncology care quality,6-9 although the performance of these measure sets has not typically been assessed at the practice level.

    In this cross-sectional study, we characterized practice-level quality of oncology care using registry and claims-based measures of process, utilization, end-of-life (EOL) care, and survival for older patients with newly diagnosed lung cancer, colorectal cancer (CRC), or breast cancer. We then assessed the reliability of cancer care quality measures across oncology practices. We also assessed the correlations of practice-level performance across types of measures and cancers to ascertain whether practices with better performance in some types of measures or cancers had similarly better performance in other types.

    Methods

    This cross-sectional study used the Surveillance, Epidemiology, and End Results (SEER) Program cancer registry linked to Medicare administrative data from January 1, 2010, to December 31, 2016 (the most recent dates in which data were available), to obtain data on cancers diagnosed between January 1, 2011, and December 31, 2015.10 The population-based SEER registry includes demographic and clinical information for all patients with newly diagnosed cancer who were living in areas that covered 28% of the US population during the study period.11 Medicare data included inpatient, outpatient, carrier (including fee-for-service claims submitted by those that provide services, such as physicians, physician assistants, clinical social workers, nurse practitioners, and organizations), durable medical equipment, and hospice and home health files. This study was approved by the Harvard Faculty of Medicine Institutional Review Board, which granted a waiver of informed consent because the study was a secondary analysis of previously collected data. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

    Study Population and Quality Measures

    We included individuals with lung cancer, CRC, or breast cancer (all histologic types) that was diagnosed between January 1, 2011, and December 31, 2015. Individuals were 65 years or older, were continuously enrolled in Parts A and B of fee-for-service Medicare (non–health maintenance organization), and were alive through at least 6 months after diagnosis, although some measures required longer continuous enrollment (eTable 1 in the Supplement). Measures of adjuvant endocrine therapy for breast cancer required Part D enrollment. See eTable 1 in the Supplement for cohort identification.

    Quality measures (Table 1) included care processes, ED visits, and hospital use for patients who were receiving chemotherapy as well as EOL care and survival (lung cancer and CRC only, given high breast cancer–specific survival) that could be assessed using registry and administrative claims data. Quality measures were adapted from those endorsed by the National Quality Forum,8 American College of Surgeons,7 ASCO Quality Oncology Practice Initiative,6 OCM,4 CMS Consensus Core Set,9 or National Comprehensive Cancer Network or ASCO guidelines that were applicable during the study period and for which we anticipated sufficient sample sizes. eAppendix 3 and eTable 5 in the Supplement detail the measure specifications.

    We studied the quality of care delivered by medical oncology practices. Patients were attributed to practices on the basis of outpatient evaluation and management visit claims with a cancer diagnosis and a CMS specialty code of medical oncology, hematology/oncology, hematology, or gynecologic oncology (eAppendix 1 and eTable 2 in the Supplement). We attributed patients to the practice (identified using tax identification numbers) with the most evaluation and management visits or with the greatest payments in the event of a tie. Patients were attributed to practices according to care in the 6 months after diagnosis for process and survival measures. Attribution for ED visits and hospitalizations focused on 6-month episodes triggered by chemotherapy, as was done for the OCM attribution.4 Attribution for EOL measures focused on care in the last 6 months of life. For each measure and cancer type, we identified the number of practices with at least 20 patients over the 5-year study period. For each measure, we excluded practices with fewer than 20 patients during the 5-year study period (eAppendix 2 and eTable 3 in the Supplement have additional details).

    Statistical Analysis

    For each measure for which a patient was eligible, we assessed whether the measure was met. We also created summary process measures by summing the number of measures met divided by the number of measures for which patients were eligible. Utilization summary measures assessed if patients had an ED visit or hospitalization. Adjusted practice-level quality measures were estimated using multilevel hierarchical linear models with practice-level random effects to compare rates for each measure, holding patient case mix constant (eAppendix 4 in the Supplement). Models adjusted for age (65-74 years, 75-84 years, or ≥85 years), sex, race/ethnicity (White; Black; Hispanic; or other, including Asian/Pacific Islander, American Indian/Alaskan Native, and unknown), marital status (unmarried, married, or unknown), census-tract median household income (quartiles), census-level proportion of residents without a high school education (quartiles), Charlson Comorbidity Index12 (0, 1, 2, or ≥3), and year of diagnosis (eTable 4 in the Supplement). We adjusted for cancer stage (I, II, III, IV, or unknown) as defined by the American Joint Commission on Cancer Cancer Staging Manual, 7th edition, when patients with more than 1 stage classification were included in a measure. Data on race/ethnicity were missing for less than 1% of patients; these patients were classified as part of the other or unknown race/ethnicity category. Marital status was missing for approximately 5% of patients, and cancer stage was missing for 2.6% to 3.8% (depending on cancer type) of patients, who were categorized separately.

    Using estimated random effects from the hierarchical linear model, we estimated the adjusted practice-level rates of each measure for each practice with at least 20 eligible patients for that measure across all study years. We assessed the reliability of measures13-15 to understand how well each measure distinguished the differences in quality across practices attributable to true variation rather than chance. Reliability represents a measure’s reproducibility and is a function of the measure’s within and between-practice variation and the sample size. Reliability was calculated as the between-practice variance divided by (between-practice variance + within-practice variance ÷ n). For each individual and summary measure, we identified reliability for the practice with the median and 25th and 75th percentile number of patients. We considered reliability lower than 0.75 (which meant that >25% of variation in a practice’s performance was associated with chance instead of true differences in quality) to be inadequate.16,17

    Reliability calculations were based on sample sizes for newly diagnosed fee-for-service Medicare beneficiaries aged 65 years or older who were treated in the oncology practices over a 5-year period. We recalculated reliability after estimating the total number of newly diagnosed patients with cancer that a practice would be expected to treat, assuming that data on quality could be extracted from a comprehensive electronic medical record. We estimated the number of patients per practice expected if we also had data for patients with Medicare Advantage (based on SEER-Medicare data) and for individuals younger than 65 years (based on the age distribution for each cancer type [eAppendix 5 and eTable 6 in the Supplement]). In addition, we estimated reliability on the basis of the expected number of patients if 2 years of data instead of 5 years were pooled given that programs assessing quality would be interested in more recent data. These calculations assumed that between-practice performance variation would not change. However, it is possible that such variation would be larger in a more diverse population, and thus we may be underestimating reliability.

    Next, using Spearman correlation coefficients, we assessed the practice-level correlations of different types of summary measures (process, utilization, EOL care, and survival) within cancer types for practices with 20 or more eligible patients for each pair of summary measures. Similarly, we assessed the correlations of summary measure types (process, utilization, EOL care, and survival) across cancer types for practices with 20 or more patients for each pair of cancer type.

    The analyses were descriptive. Spearman correlation coefficients were reported with 2-sided P values, and P < .05 was considered statistically significant; however, even with statistically significant correlations, the focus was on the correlation coefficient, which was considered negligible if lower than 0.20, weak if 0.20 to 0.39, moderate if 0.40 to 0.69, and strong if 0.70 or higher. Data analyses were conducted from January 2018 to December 2020. We used SAS, version 9.2 (SAS Institute Inc), and R, version 3.5.2 (R Foundation for Statistical Computing).

    Results

    We assessed care processes for 49 715 patients with lung cancer who were treated in 502 practices with 20 or more patients with lung cancer, 21 692 patients with CRC who were treated in 347 practices with 20 or more patients with CRC, and 52 901 patients with breast cancer who were treated in 492 practices with 20 or more patients with breast cancer (eTable 3 in the Supplement). Patients were 65 years or older, and approximately 50% of the patients with lung cancer and CRC and all of the patients with breast cancer were women (eTable 3 in the Supplement). The number of practices that were eligible for each process measure varied substantially. For lung cancer, among the 502 practices with 20 or more patients with cancer of any stage, only 12 to 111 practices had 20 or more patients who were eligible for each process measure (Table 1). Similarly, small numbers of practices had 20 or more patients who were eligible for various CRC and breast cancer measures (47 to 223 of 347 practices for CRC measures, and 2 to 492 of 492 practices for breast cancer measures) (Table 1).

    Reliability of measures across practices was low (Table 1). Among process measures, 0 of 6 measures for lung cancer, 0 of 6 for CRC, and 3 of 11 for breast cancer (including summary measures) had reliability of 0.75 or higher for the median-sized practice (Table 1), although 2 lung cancer measures had reliability higher than 0.70. Low reliability resulted from both lack of variation in rates of the measures across practices and relatively few patients who were eligible for the measure. For example, the median (interquartile range [IQR]) rate of adjuvant chemotherapy within 60 days of surgery for patients with stage II or IIIA lung cancer was 88% (86%-90%) for the median-sized practice, and the median (IQR) number of patients was only 24 (20-34), even among the 20 practices with 20 or more patients who were eligible for this measure. Both factors contributed to reliability: 0.34 for the median-sized practice, and 0.42 for the practice at the 75th percentile in size (34 eligible patients). Reliability was higher for measures that assessed imaging (eg, reliability of 0.74 [0.69-0.80] for surveillance computed tomography for resected stage I, II, or IIIA non–small cell lung cancer, or reliability of 0.74 [0.66-0.80] for resected stage II-III CRC), which had more variation across practices and larger sample sizes. Reliability also was low for utilization, EOL care, and survival measures, in which sample sizes were generally larger but variability across practices was limited (Table 1).

    When assessing the correlations of summary measures (process, utilization, EOL care, survival) for each cancer type, we found that the correlation across measures was low (r ≤ 0.21 for all), except for the correlation of the process and survival summary measures for CRC (r = 0.35; P < .001) (Figure 1 and eFigure 1 in the Supplement) and that some measures were inversely correlated, although the inverse correlation between process and EOL care measures for breast cancer was based on only 23 practices. We also observed limited correlation of process, utilization, or survival measures across cancer types. For example, summary process measures had limited or no correlation across lung cancer, breast cancer, and CRC (r ≤ 0.16 for all). The EOL care measures were moderately correlated across cancer types (r = 0.35 to 0.64) (Figure 2 and eFigure 2 in the Supplement).

    Reliability estimates were based on newly diagnosed fee-for-service Medicare beneficiaries aged 65 years or older. Because reliability is a function of sample size, we recalculated reliability after estimating the sample sizes for practices, assuming that patients with Medicare Advantage and those younger than 65 years were included, as if we had data from a comprehensive electronic medical record. Table 2 shows the number of measures with an estimated reliability of 0.75 or higher with updated sample sizes to approximate all patients with newly diagnosed cancers in the practices over 5 years. Despite the much larger sample sizes, only 4 of 6 lung cancer, 1 of 6 CRC, and 7 of 11 breast cancer process measures were estimated to have reliability of 0.75 or higher, pooling 5 years of data (Table 2 and eTable 7 in the Supplement). If limited to 2 years of patients with newly diagnosed cancer instead of 5 years, then 0 of 6 lung cancer, 1 of 6 CRC, and 4 of 11 breast cancer process measures would be expected to have reliability of 0.75 or higher. None of the utilization, EOL care, or survival measures would have reliability of 0.75 or higher. eTable 2 in the Supplement shows the minimum number of patients across practices that is necessary to have reliability of 0.75 or higher across all measures.

    Discussion

    In this study of quality measures for oncology care in patients with newly diagnosed cancer across a variety of domains, we identified that most measures, including measures endorsed by national organizations, were not well suited for assessing care quality across medical oncology practices. These measures were limited by low reliability (ability to detect signal instead of noise) because of generally small sample sizes and/or limited variability in measure rates across practices, despite pooling the data for patients with newly diagnosed cancer over 5 years and after excluding practices with the smallest sample sizes. We also observed limited to no correlation of measures across measure and cancer types.

    Cancer is prevalent in the United States, with more than 1.8 million individuals expected to be diagnosed in 2021.18 Interest in improving care for patients with cancer and the emergence of value-based payment models to reimburse that care have increased interest in measuring the quality of cancer care; however, numerous challenges exist in quality measurement. First, cancer consists of not just one but many diseases; even within cancer types, the extent of disease and tumor characteristics necessitate different local and/or systemic treatments.

    Second, cancer care is multidisciplinary; individual patients may receive care from surgeons, medical oncologists, radiation oncologists, and/or palliative care clinicians. Even when guidelines recommend multidisciplinary care, it is often unclear who should be accountable for that care.19 For example, if a patient with early-stage estrogen receptor–positive breast cancer does not receive endocrine therapy after lumpectomy and radiation, it may be unclear whether the lack of treatment should be attributed to the patient’s surgeon, radiation oncologist, and/or medical oncologist.

    Third, given the heterogeneity of cancer diagnoses, the number of patients with newly diagnosed cancers of particular type and stage who are treated at even a relatively large practice within a year tends to be small.20 Another effort to assess the quality of oncology care across practices in Washington state similarly found relatively small sample sizes (even after restricting to a few larger practices) and limited variability across measures.21,22 That effort successfully created summary measures that reflected the receipt of recommended treatment across cancer types.21,22 Despite larger sample sizes, reliability of the summary measures in the present study remained limited. Moreover, the lack of correlation of process measures across cancer type for the Medicare patients in the present study suggests that such a strategy is unlikely to be successful in this population.

    Fourth, the available data used to analyze the quality of cancer care have limitations. Administrative data are easily available to payers but lack clinical details, such as stage, histologic type, and tumor markers. These details are critical for risk-adjusting the outcome measures and understanding the appropriateness of care. Such data will be even more important as cancer therapy becomes increasingly targeted. Medical records data provide details about cancer characteristics and care delivered, such as those collected for the ASCO Quality Oncology Practice Initiative.23 However, medical record abstraction is time consuming and costly. Ideally, electronic medical record systems would allow for timely and efficient quality measurement. Unfortunately, the ability to incorporate needed data into structured fields that can be combined across various electronic health record systems has yet to be realized, although this is an area of interest (eg, mCODE and CancerLinQ from ASCO, The MITRE Corporation, and others).24

    Fifth, even among the few measures that have been endorsed by national organizations such as the National Quality Forum and CMS,8,9 the endorsement process has not involved assessing reliability at the practice level. Research assessing hospital and surgeon profiling for 3 Commission on Cancer surgery measures noted acceptable reliability for some but not all measures.25 Previous work has incorporated assessments of institution-level variation in identifying quality improvement targets26-28; such assessments are particularly important for value-based payment models that include financial incentives for practices to improve the care they deliver. Some measures had high rates of performance but little variability, whereas others had opportunities for improvement; however, we observed little variability across practices, reducing the reliability of these measures. A caveat to this analysis is that we focused on patients with newly diagnosed cancer. Nevertheless, other research has shown little change in practice-level quality over time for many measures, including measures with wide performance gaps.29

    The limited correlations we observed across measure type suggest that these different measures provide different information or incomplete pictures of quality. For example, although the process measures reflect care that has been correlated with better long-term survival, such improvements may not be captured in the survival measure at 1 year. Similarly, inadequate risk adjustment could potentially explain limited or negative correlations between the utilization and survival measures (eg, sicker patients have more hospitalizations and worse survival).

    Findings of this study may be disappointing for proponents of value-based payment and those seeking to pay for quality and not volume. Nevertheless, these findings provide a reminder of the need to assess measure reliability, particularly for high-stakes purposes such as public reporting or payment. It is possible that other measures, such as guideline-recommended use of supportive care drugs for patients undergoing chemotherapy (eg, appropriate use of prophylactic growth factors), may perform better than the measures that we examined given that they can be assessed across large populations of patients with different cancer types. Other measures worth exploring could target the avoidance of low-value care, including choice of lower cost but equally effective treatments, such as biosimilar products. However, a key factor in the choice of measures is the intended use. If measures are to be used for value-based payment models, in which incentives for practices to provide more efficient care already exist, then quality measures to assess the underuse of care would be of higher priority than measures of efficiency. Efficiency measures might be more appropriate for use by payers that are looking to identify preferred practices for their patients to receive treatment.

    Additional research is needed to identify reliable quality measures for practice-level alternate payment models. For example, strategies that target supplemental data collection on a subset of practices that perform below certain thresholds on certain core measures hold promise.30 Given the challenges and limitations of measure-focused approaches to improving care, initiatives that aim to improve care by more productively leveraging professionalism31 are also promising.

    Strengths and Limitations

    This study has several strengths. First, it is a comprehensive assessment of a variety of quality measures across 3 cancer types and 4 domains. Second, it analyzed hundreds of practices that treat patients with cancer and assessed measures that have been endorsed by national organizations.

    This study also has some limitations. First, we used SEER linked to Medicare administrative data to identify stage-specific care for patients with a new cancer diagnosis, for whom the receipt of high-quality care may be particularly relevant, but our assessments may not generalize to the care of patients with recurrent disease. In addition, the small sample sizes that we identified may be less relevant to programs such as the OCM, which focuses on patients undergoing chemotherapy. Nevertheless, claims-based measures used in value-based programs, such as the OCM (ED visits and hospice enrollment for ≥3 days), had limited variability across practices, necessitating large sample sizes for adequate reliability, although variation may be greater in cohorts beyond patients who were recently diagnosed. We analyzed data from older adults enrolled in fee-for-service Medicare. Even after extrapolating the sample sizes to consider all patients (of all ages and insurance types) with a new diagnosis, the study results suggested that reliability would remain low, although variation in performance would likely be higher in a more diverse population.

    Second, attributing patients to oncology practices can be challenging. Some patients receive care from more than 1 practice, and for many measures, the recommendations from surgeons, radiation oncologists, and primary care clinicians may be a factor in a patient’s receipt of certain therapies. We studied medical oncologists to compare similar types of practices and used established methods to identify the practices that are most important in a patient’s care.4

    Third, we considered reliability of 0.75 or higher to be good. This decision was based on Consumer Assessment of Healthcare Providers and Systems reporting for Medicare, in which star ratings are not reported if reliability is very low (<0.60) and are reported as “low reliability” if reliability is 0.60 or higher but lower than 0.75.32 Although reliability of 0.90 or higher may be deemed high,16 others have proposed that reliability higher than 0.70 is acceptable for drawing conclusions about groups.17 Nevertheless, most of the measures we analyzed had reliability well below 0.70.

    Fourth, we used a regression modeling approach to adjust the case mix, which relies on extrapolation of data. An alternative approach is template matching,33 which can avoid extrapolation but may exclude practices with patient populations that are not similar to other practices.

    Conclusions

    In this cross-sectional study, we found that measures of oncology care process, utilization, EOL care, and survival that were assessed using registry and administrative data were limited by a small number of fee-for-service Medicare patients with newly diagnosed cancer across oncology practices, even after pooling 5 years of data. Most measures had low reliability because of small sample sizes and/or limited variability across practices. Moreover, the measures had limited to no correlation across measure and cancer types. Future research is needed to identify reliable quality measures for practice-level alternate payment models.

    Back to top
    Article Information

    Accepted for Publication: January 29, 2021.

    Published: March 22, 2021. doi:10.1001/jamanetworkopen.2021.2474

    Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Keating NL et al. JAMA Network Open.

    Corresponding Author: Nancy L. Keating, MD, MPH, Department of Health Care Policy, Harvard Medical School, 180 Longwood Ave, Boston, MA 02115 (keating@hcp.med.harvard.edu).

    Author Contributions: Dr Keating and Ms Meneades 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.

    Concept and design: Keating, Brooks, Landrum.

    Acquisition, analysis, or interpretation of data: All authors.

    Drafting of the manuscript: Keating, Cleveland, Meneades.

    Critical revision of the manuscript for important intellectual content: Keating, Wright, Brooks, Riedel, Zubizarreta, Landrum.

    Statistical analysis: Keating, Cleveland, Meneades, Zubizarreta, Landrum.

    Obtained funding: Keating.

    Administrative, technical, or material support: Keating, Meneades, Riedel.

    Supervision: Keating.

    Conflict of Interest Disclosures: Dr Keating reported receiving grants from Arnold Ventures during the conduct of the study and grants from the Centers for Medicare & Medicaid Services (CMS) contract evaluation team for the Oncology Care Model outside the submitted work. Dr Wright reported receiving grants from Arnold Ventures during the conduct of the study. Dr Riedel reported receiving grants from Arnold Ventures during the conduct of the study. Dr Landrum reported receiving grants from Arnold Ventures during the conduct of the study. No other disclosures were reported.

    Funding/Support: This study was supported by Arnold Ventures. The collection of cancer incidence data used in this study was supported by the California Department of Public Health as part of the statewide cancer reporting program mandated by California Health and Safety Code Section 103885; the National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) Program under contract HHSN261201000140C awarded to the Cancer Prevention Institute of California, contract HHSN261201000035C awarded to the University of Southern California, and contract HHSN261201000034C awarded to the Public Health Institute; and the Centers for Disease Control and Prevention (CDC) National Program of Cancer Registries under agreement U58DP003862-01 awarded to the California Department of Public Health.

    Role of the Funder/Sponsor: The funders 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.

    Disclaimer: The views expressed herein are those of the authors and do not reflect the official policy or position of Arnold Ventures, its directors, officers, or staff. The ideas and opinions expressed herein are those of the authors, and endorsement by the California Department of Public Health, the NCI, and the CDC or their contractors and subcontractors is not intended nor should be inferred.

    References
    1.
    Centers for Medicare & Medicaid Services. Oncology care model. Last updated June 22, 2020. Accessed July 7, 2020. https://innovation.cms.gov/innovation-models/oncology-care
    2.
    Nardi  EA, McCanney  J, Winckworth-Prejsnar  K,  et al.  Redefining quality measurement in cancer care.   J Natl Compr Canc Netw. 2018;16(5):473-478. doi:10.6004/jnccn.2018.7028 PubMedGoogle ScholarCrossref
    3.
    Hassett  MJ.  Quality improvement in the era of big data.   J Clin Oncol. 2017;35(28):3178-3180. doi:10.1200/JCO.2017.74.1181 PubMedGoogle ScholarCrossref
    4.
    Centers for Medicare & Medicaid Services. OCM performance-based payment methodology, version 5.1. Accessed July 7, 2020. https://innovation.cms.gov/files/x/ocm-pp3beyond-pymmeth.pdf
    5.
    Centers for Medicare & Medicaid Services. Quality payment program: MIPS overview. Accessed July 31, 2020. https://qpp.cms.gov/mips/overview
    6.
    American Society of Clinical Oncology. QOPI-related measures. Accessed July 7, 2020. https://practice.asco.org/quality-improvement/quality-programs/quality-oncology-practice-initiative/qopi-related-measures
    7.
    American College of Surgeons. CoC quality of care measures 2020 surveys. Last updated June 1, 2020. Accessed July 7, 2020. https://www.facs.org/quality-programs/cancer/ncdb/qualitymeasurescocweb
    8.
    National Quality Forum. Measuring performance. Accessed July 7, 2020. https://www.qualityforum.org/Measuring_Performance/Measuring_Performance.aspx
    9.
    Centers for Medicare & Medicaid Services. Consensus core set: medical oncology measures, version 1.0. Accessed July 31, 2020. https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/QualityMeasures/Downloads/Medical-Oncology-Measures.pdf
    10.
    Potosky  AL, Riley  GF, Lubitz  JD, Mentnech  RM, Kessler  LG.  Potential for cancer related health services research using a linked Medicare-tumor registry database.   Med Care. 1993;31(8):732-748. doi:10.1097/00005650-199308000-00006 PubMedGoogle ScholarCrossref
    11.
    National Cancer Institute. SEER data & software. Accessed July 7, 2020. https://seer.cancer.gov/data-software/
    12.
    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. doi:10.1016/S0895-4356(00)00256-0 PubMedGoogle ScholarCrossref
    13.
    Hofer  TP, Hayward  RA, Greenfield  S, Wagner  EH, Kaplan  SH, Manning  WG.  The unreliability of individual physician “report cards” for assessing the costs and quality of care of a chronic disease.   JAMA. 1999;281(22):2098-2105. doi:10.1001/jama.281.22.2098 PubMedGoogle ScholarCrossref
    14.
    Carmines  EG, Zeller  RA. Reliability and validity assessment. In: Lewis-Beck MS, ed. Series: Quantitative Applications in the Social Sciences. Sage Publications; 1979.
    15.
    Bravo  G, Potvin  L.  Estimating the reliability of continuous measures with Cronbach’s alpha or the intraclass correlation coefficient: toward the integration of two traditions.   J Clin Epidemiol. 1991;44(4-5):381-390. doi:10.1016/0895-4356(91)90076-L PubMedGoogle ScholarCrossref
    16.
    Zaslavsky  AM.  Statistical issues in reporting quality data: small samples and casemix variation.   Int J Qual Health Care. 2001;13(6):481-488. doi:10.1093/intqhc/13.6.481 PubMedGoogle ScholarCrossref
    17.
    Adams  JL. The reliability of provider profiling: a tutorial. Prepared for the National Committee for Quality Assurance. RAND Corporation Technical Report Series 2009. Accessed July 7, 2020. https://www.rand.org/content/dam/rand/pubs/technical_reports/2009/RAND_TR653.pdf
    18.
    Siegel  RL, Miller  KD, Fuchs  HE, Jemal  A.  Cancer Statistics, 2021.   CA Cancer J Clin. 2021;71(1):7-33. doi:10.3322/caac.21654 PubMedGoogle ScholarCrossref
    19.
    Gondi  S, Wright  AA, Landrum  MB, Zubizarreta  J, Chernew  ME, Keating  NL.  Multimodality cancer care and implications for episode-based payments in cancer.   Am J Manag Care. 2019;25(11):537-538.PubMedGoogle Scholar
    20.
    Shulman  LN, Palis  BE, McCabe  R,  et al.  Survival as a quality metric of cancer care: use of the National Cancer Data Base to assess hospital performance.   J Oncol Pract. 2018;14(1):e59-e72. doi:10.1200/JOP.2016.020446 PubMedGoogle ScholarCrossref
    21.
    Panattoni  L, Fedorenko  C, Kreizenbeck  K,  et al.  Lessons from reporting national performance measures in a regional setting: Washington state community cancer care report.   J Oncol Pract. 2018;14(12):e801-e814. doi:10.1200/JOP.18.00410 PubMedGoogle ScholarCrossref
    22.
    Hutchinson Institute for Cancer Outcomes Research. Community cancer care in Washington state: quality and cost report 2019. Accessed July 7, 2020. https://www.fredhutch.org/content/dam/www/research/institute-networks-ircs/hicor/HICOR-Community-Cancer-Care-Report-2019.pdf
    23.
    ASCO Practice Central. Quality oncology practice initiative. Accessed July 7, 2020. https://practice.asco.org/quality-improvement/quality-programs/quality-oncology-practice-initiative
    24.
    mCODE. mCODE: minimal common oncology data elements: the initiative to create a core cancer model and foundational EHR data elements. Accessed July 7, 2020. https://mcodeinitiative.org/
    25.
    Liu  JB, Huffman  KM, Palis  BE,  et al.  Reliability of the American College of Surgeons Commission on Cancer’s quality of care measures for hospital and surgeon profiling.   J Am Coll Surg. 2017;224(2):180-190.e8. doi:10.1016/j.jamcollsurg.2016.10.053PubMedGoogle ScholarCrossref
    26.
    Enright  KA, Taback  N, Powis  ML,  et al.  Setting quality improvement priorities for women receiving systemic therapy for early-stage breast cancer by using population-level administrative data.   J Clin Oncol. 2017;35(28):3207-3214. doi:10.1200/JCO.2016.70.7950 PubMedGoogle ScholarCrossref
    27.
    Weeks  JC, Uno  H, Taback  N,  et al.  Interinstitutional variation in management decisions for treatment of 4 common types of cancer: a multi-institutional cohort study.   Ann Intern Med. 2014;161(1):20-30. doi:10.7326/M13-2231 PubMedGoogle ScholarCrossref
    28.
    Vos  EL, Koppert  LB, Jager  A, Vrancken Peeters  MTFD, Siesling  S, Lingsma  HF.  From multiple quality indicators of breast cancer care toward hospital variation of a summary measure.   Value Health. 2020;23(9):1200-1209. doi:10.1016/j.jval.2020.05.011 PubMedGoogle ScholarCrossref
    29.
    Neuss  MN, Malin  JL, Chan  S,  et al.  Measuring the improving quality of outpatient care in medical oncology practices in the United States.   J Clin Oncol. 2013;31(11):1471-1477. doi:10.1200/JCO.2012.43.3300 PubMedGoogle ScholarCrossref
    30.
    Chernew  ME, Landrum  MB.  Targeted supplemental data collection—addressing the quality-measurement conundrum.   N Engl J Med. 2018;378(11):979-981. doi:10.1056/NEJMp1713834 PubMedGoogle ScholarCrossref
    31.
    McWilliams  JM. Professionalism revealed: rethinking quality improvement in the wake of a pandemic. Accessed July 30, 2020. https://catalyst.nejm.org/doi/full/10.1056/CAT.20.0226
    32.
    Centers for Medicare & Medicaid Services. Medicare 2020 Part C & D star ratings technical notes. Last updated October 1, 2019. Accessed July 7, 2020. https://www.cms.gov/Medicare/Prescription-Drug-Coverage/PrescriptionDrugCovGenIn/Downloads/Star-Ratings-Technical-Notes-Oct-10-2019.pdf
    33.
    Silber  JH, Rosenbaum  PR, Ross  RN,  et al.  Template matching for auditing hospital cost and quality.   Health Serv Res. 2014;49(5):1446-1474. doi:10.1111/1475-6773.12156 PubMedGoogle ScholarCrossref
    ×