Retention of Enrollees Following a Cancer Diagnosis Within Health Maintenance Organizations in the Cancer Research Network

Transcription

1 Retention of Enrollees Following a Cancer Diagnosis Within Health Maintenance Organizations in the Cancer Research Network Terry S. Field, Jackie Cernieux, Diana Buist, Ann Geiger, Lois Lamerato, Gene Hart, Don Bachman, Rick Krajenta, Sarah Greene, Mark C. Hornbrook, Gary Ansell, Lisa Herrinton, George Reed Population laboratories with complete clinical information on episodes of care are needed to support research on the quality of care delivered to cancer patients. Data resources within the Cancer Research Network (CRN) may overcome many of the limitations of existing cancer databases, but their potential clinical value depends on the stability of the enrolled population. To assess this issue, we studied the retention rates among survivors of the patients diagnosed with cancer from January 1, 1993, through December 31, 1998, who were enrolled at five health maintenance organization sites participating in the CRN. Enrollees were followed from cancer diagnosis through death, disenrollment, or the end of follow-up (i.e., December 31, 1999). The retention rate among survivors for all cancers combined at 1 and 5 years after cancer diagnosis was 96.0% (95% confidence interval [CI] 95.9% to 96.1%) and 83.9% (95% CI 83.4% to 84.3%), respectively. The proportion of enrollees diagnosed with cancer who remained enrolled and available for evaluation suggests that the CRN is well-suited for studies of the quality of care for cancer patients, survivorship, and longterm outcomes. [J Natl Cancer Inst 2004;96:148 52] Studies suggest that scientific breakthroughs in cancer treatment are not fully realized in the day-to-day delivery of care (1). To obtain a better understanding of factors related to the quality of cancer care, researchers require access to population laboratories (2) that have substantial numbers of cancer patients, information about all episodes of care in the full range of clinical settings, and adequate follow-up to identify longterm outcomes. However, none of the existing databases of cancer patients in the United States includes these three components (2). In 1998, the National Cancer Institute (NCI) funded a consortium of 10 research groups, based within not-forprofit health maintenance organizations (HMOs), to conduct collaborative research in cancer prevention and control. This consortium, called the Cancer Research Network (CRN), now includes the research programs, enrollee populations, and databases of 11 members of the HMO research network. The CRN HMOs provide care for nearly 9 million enrollees, who represent a large and ethnically diverse population living across the United States. These HMOs provide a continuum of cancer care, from screening through diagnosis and treatment. Treatment provided to patients outside the HMOs is captured in electronic administrative claims at each site. In addition, most of these HMOs also maintain electronic files on pharmacy use, and all have access to electronic medical records or paper charts for inpatient and outpatient care. Hence, the available information allows researchers to take into account patients comorbidities, health status, age, and prior health care use. This array of information resources supports research efforts into the factors associated with optimal cancer care, the long-term impact of cancer treatments, and the cost of cancer care. One potential issue in regard to the value of the CRN as a population laboratory is the stability of the enrolled population. The 11 HMOs that make up the CRN have overall estimated 1-year enrollee retention rates ranging from 84% to 90% and estimated 5-year retention rates ranging from 59% to 76%. However, despite the good overall retention rates for these HMOs, retention rates specifically for enrollees diagnosed with cancer have not been estimated. To address this issue, we conducted a study to determine 1) the percentage of cancer survivors who remained continuously enrolled at 1, 2, 3, 4, and 5 years after cancer diagnosis and 2) the characteristics of enrollees with higher rates of disenrollment during the initial treatment period (defined as the first year following diagnosis). The five CRN sites participating in this study Henry Ford Health System (based in Detroit, MI), Group Health Cooperative (based in Seattle, WA), and Kaiser Permanente in three regions (Northwest [Oregon and Washington State], Northern California, and Southern California) were predominantly staff-model HMOs that had complete cancer registries linked to patient enrollment histories. Together, these five HMOs provided care to more than enrollees in These five CRN sites were a convenience sample in that they were selected because of the interest of local investigators in participating in this study. The estimated 1-year retention rates of the total enrolled populations at these five sites were similar to and matched the range of retention rates across all sites in the CRN, and we have no reason to believe that the retention rates among the cancer patients in these five sites would not be representative of the overall enrolled CRN population. This study was approved by the institutional review board of the University of Massachusetts Medical School (Worcester, MA) and the respective institutional review boards of the participating CRN sites. Study participants were health-plan members who were diagnosed with incident cancer while they were enrolled in one of the five participating HMO sites from January 1, 1993, through December 31, Inclusion of par- Affiliations of authors: Meyers Primary Care Institute, Fallon Healthcare System (TSF, JC) and the University of Massachusetts Medical School (TSF, JC, GR), Worcester, MA; Center for Health Studies, Group Health Cooperative, Seattle, WA (DB, GH, SG); Department of Research and Evaluation, Southern California Permanente Medical Group, Pasadena, CA (AG); Henry Ford Health System, Detroit, MI (LL, RK); Center for Health Research, Kaiser Permanente Northwest, Portland, OR (DB, MCH, GA); Division of Research, Kaiser Permanente Northern California, Oakland, CA (LH). Correspondence to: Terry S. Field, DSc, Meyers Primary Care Institute, 630 Plantation St., Worcester, MA ( terry.field@meyersprimary.org and terry.field@umassmed.edu). See Notes following References. DOI: /jnci/djh010 Journal of the National Cancer Institute, Vol. 96, No. 2, Oxford University Press 2004, all rights reserved. 148 BRIEF COMMUNICATIONS Journal of the National Cancer Institute, Vol. 96, No. 2, January 21, 2004

2 ticipants into the study was based on the first cancer diagnosis for each enrollee during the study time period that was not described as noninvasive, diagnosed at autopsy, or nonmelanoma skin cancer. Cancer registries define the date of cancer diagnosis as the earliest date that the potential patient was brought to the attention of a health care provider. Therefore, these dates could be earlier than the patient was considered enrolled in the HMO, even though the final confirmation of cancer diagnosis and initial course of treatment were carried out by the HMO. Patients whose cancer diagnosis was within 60 days before the start of enrollment were included in this study to ensure that we did not lose these patients. Patients whose cancer diagnosis was within 45 days following termination of enrollment were also included to ensure capture of patients who disenrolled immediately after their cancer diagnosis. Information on enrollees diagnosed with cancer was drawn from electronic data files at each of the participating CRN sites. Each site has a cancer registry that covers the total enrolled population and contains data elements, such as primary cancer site, morphology, general stage, and behavior, that are consistent with the Surveillance, Epidemiology, and End Results (SEER) Program 1 registries. From these cancer registries, data were collected for all eligible cancer diagnoses, including the date of the diagnosis, cancer stage, primary tumor site, morphology and behavior of the tumor, enrollee s age at the time of cancer diagnosis, sex, race and ethnicity, and date of death. Enrollees identified in the cancer registry data were linked to enrollment files from which a complete enrollment history was extracted, including dates of enrollment and disenrollment and type of insurance for each enrollment period until the end of follow-up (i.e., December 31, 1999). At the time of data extraction, all cancer registry and enrollment data at each site were up-to-date through Rates of successful linkages between cancer registry data and enrollment files across the five sites ranged from 96.3% to 100%. The cancer registries at these five sites used a variety of methods to track deaths among those patients who remained enrolled in their HMO and those who disenrolled. Four sites maintained ongoing linkages with the SEER programs in their catchment areas for follow-up, and the fifth site used automated linkages to state death records and search firms to track deaths. Three sites enhanced their SEER linkages for tracking deaths by performing independent searches of state and national death indices. The 5-year follow-up rates for deaths among those patients no longer enrolled in their HMO ranged from 95.0% to 99.8%. Disenrollment date was defined as the first termination (of coverage) date after cancer diagnosis. Patients were considered continuously enrolled in an HMO until the termination date or until the end of follow-up. Data from the cancer registries and enrollment files were used to assess patient characteristics associated with disenrollment during the first year after cancer diagnosis. Sex was categorized as female or male, with the male category including 16 patients coded as other in the registry data. Race and ethnicity were combined, and patients were categorized as Hispanic if they were identified as having a Spanish surname or origin in the cancer registries databases (3). The categories from the cancer registries race variable were retained for those patients not identified as Hispanic. American Indian patients were combined with patients coded under the unknown race category because there were too few patients identified as American Indian to enable us to individually assess their disenrollment. Cancer type was established by combining the cancer registries primary tumor site and morphology variables. For prognosis, we estimated the 5-year survival rates of patients for specific cancer types on the basis of the 5-year survival rates from SEER ( ; see seer.cancer.gov/csr/1975_2000/results _merged/topic_survival.pdf) and categorized these rates in approximate quartiles by their frequency in our sample. For cancer stage, we used the cancer registries SEER general summary stage variable (see gov/tools/ssm/intro.pdf). Enrollment variables included type of insurance and length of continuous enrollment before cancer diagnosis. We categorized insurance type as Medicare, commercial, self-pay, and Medicaid at the time of cancer diagnosis. Because there were only a few patients eligible for both Medicare and Medicaid and their disenrollment rates matched those for patients with Medicaid alone, we combined these two insurance type categories. Patients whose insurance was obtained through a state s expanded Medicaid program were classified as Medicaid. Length of continuous enrollment before cancer diagnosis was categorized, a priori, as less than 30 days, days, 180 days to less than 1 year, 1 to less than 4 years, 4 to less than 10 years, and more than 10 years. After characterizing the population in terms of demographics (e.g., age, sex, and race), cancer types and stages, insurance types, and length of enrollment before cancer diagnosis, we calculated the number of cancer patients who were alive at 1, 2, 3, 4, and 5 years after diagnosis, censoring patients who died without disenrolling and those who had not died or disenrolled by the end of the follow-up period. The number of patients who could be followed differed for each of these calculations according to the year of their diagnosis and the remaining time available before the end of the follow-up period (i.e., December 31, 1999). For example, the only patients who could be followed for 5 years after their cancer diagnosis were those diagnosed in 1993 and We calculated the percentage of eligible survivors who remained continuously enrolled at 1, 2, 3, 4, and 5 years after diagnosis. We repeated this series of calculations for patients within each of the four most frequently occurring cancer types (i.e., breast, prostate, colorectal, and lung/ bronchus) and within each CRN site individually. We also analyzed patient characteristics associated with disenrollment for reasons other than death during the first year after cancer diagnosis to identify any subgroups of patients with high rates of disenrollment. HMO disenrollment among children (i.e., those aged 0 19 years) is primarily based on their parents access to health insurance, so that analyses of disenrollment would have to be run separately in this age group. Moreover, the number of children with cancer in our study was too small for multivariate analysis, so this aspect of our analysis focused only on adults (i.e., those aged 20 years or older). This series of analyses began with calculations of the rates of disenrollment during the first year after cancer diagnosis for each category of the demographic, cancer type and stage, du- Journal of the National Cancer Institute, Vol. 96, No. 2, January 21, 2004 BRIEF COMMUNICATIONS 149

3 Table 1. Characteristics of patients enrolled in five health maintenance organization sites participating in the Cancer Research Network diagnosed with cancer from January 1, 1993, through December 31, 1998 Characteristics patients (%) Race White (76.0) African American (10.3) Asian or Pacific-Islander 7214 (5.4) Hispanic 9714 (7.3) Other/unknown 1191 (0.9) Sex Female (49.2) Male/other* (50.8) Age at first diagnosis during the time period, y (0.3) (0.8) (9.8) (14.6) (23.6) (29.7) (17.6) (3.6) Cancer type Breast (17.7) Prostate (17.2) Lung/bronchus (12.3) Colorectal (10.6) Melanoma, skin 6219 (4.7) Non-Hodgkin s lymphoma 5147 (3.9) Uterine 4123 (3.1) Oral/pharynx 3441 (2.6) Leukemia 3424 (2.6) Bladder 2986 (2.3) Kidney/renal 2839 (2.1) Ovarian 2788 (2.1) Pancreatic 2680 (2.0) Other (16.8) *Includes 16 patients with sex coded as other in the health maintenance organizations electronic cancer registry files. Includes all other cancer types that occurred in less than 2% of our sample. ration of prior enrollment, and insurance type variables. Multivariate models using Cox proportional hazards models were then developed using Stata Statistical Software, release 6.0 (4) to determine the independent association of enrollee characteristics with disenrollment during the first year after cancer diagnosis. Patients whose insurance type was not given in the electronic enrollment files were not included in the analysis using the multivariate models (n 6807). The assumptions of the Cox proportional hazards models were met through review of the Schoenfeld residuals (5) for each covariate and through graphical assessment of the Cox proportional hazards functions (6). We included all of the proposed variables in each multivariate model and controlled for CRN site and year of diagnosis. We assessed all possible two-way interaction terms among variables and found none to be statistically significant. Characteristics of the patients who were diagnosed with cancer during the study period are shown in Table 1. This patient population was 24% nonwhite and had a substantial number of patients in each adult age category (i.e., patients aged 20 years or older). Cancer types with the highest frequency were breast (17.7%), prostate (17.2%), lung/bronchus (12.3%), and colorectal (10.6%). Retention rates among 1-year survivors were 95% or higher for all cancers combined and for breast, prostate, colorectal, and lung/bronchus cancers when assessed individually (Table 2). Oneyear retention rates among survivors varied from 93.0% to 97.0% across the five CRN sites, with a 1-year retention rate among survivors for all cancers combined of 96.0% (95% confidence interval [CI] 95.9% to 96.1%). Fiveyear retention rates among survivors across the five CRN sites were 77% or higher for all cancers combined and for breast, prostate, colorectal, and lung/ bronchus cancers when assessed individually, with a 5-year retention rate among survivors for all cancers combined of 83.9% (95% CI 83.4% to 84.3%; Table 2). Patients with the highest rates of disenrollment during the initial treatment period were aged years, were nonwhite, and had a short length of enrollment before their cancer diagnosis (Table 3). Patients insured through Medicaid had the highest disenrollment rate, with an adjusted hazard ratio for disenrolling of 3.5 (95% CI 2.9 to 4.3). Among the patients who died during the initial treatment period, 98.9% remained enrolled until death (data not shown). Retention rates among survivors diagnosed with all cancers combined were high: 96.0% (95% CI 95.9% to 96.1%) among 1-year survivors and 83.9% (95% CI 83.4% to 84.3%) among 5-year survivors. The proportion of patients who remained enrolled and available for complete assessment of the initial treatment period was sufficient for most studies of the quality and cost Table 2. Cancer survivors continuously enrolled in five health maintenance organization sites participating in the Cancer Research Network 1 5 years after cancer diagnosis* Diagnosis years patients with cancer years of follow-up after diagnosis survivors % of survivors still enrolled (95% CI) Range in % of survivors still enrolled across the five sites All cancers combined (95.9 to 96.1) (92.4 to 92.8) (89.4 to 89.9) (86.6 to 87.3) (83.4 to 84.3) Breast cancer (96.8 to 97.3) (93.5 to 94.2) (91.2 to 92.2) (88.7 to 90.0) (85.6 to 87.4) Prostate cancer (96.5 to 97.0) (93.8 to 94.5) (91.5 to 92.4) (89.0 to 90.2) (86.2 to 87.9) Colorectal cancer (96.6 to 97.2) (93.4 to 94.4) (90.7 to 92.1) (88.4 to 90.4) (85.3 to 88.0) Lung/bronchus cancer (94.4 to 95.5) (90.1 to 92.1) (85.9 to 88.9) (81.9 to 86.0) (77.7 to 83.5) *CI confidence interval. 150 BRIEF COMMUNICATIONS Journal of the National Cancer Institute, Vol. 96, No. 2, January 21, 2004

4 Table 3. Disenrollment among adult enrollees with known insurance type during the first year after cancer diagnosis* Variable patients with cancer of patient care during this time period. Because breast and prostate cancer patients have high rates of 5-year survival, they are of particular interest for studies of long-term outcomes. Retention rates among 5-year survivors of these cancers were more than 86%, suggesting that the CRN is also an appropriate base for survivorship research. Our analyses of subgroups of cancer patients with low retention rates during the first year of treatment identified several groups of patients whose initial treatment might be difficult to follow. The highest rates of disenrollment were among individuals aged years, those insured through Medicaid, and those who were recently enrolled. The CRN s information resources may not directly capture the cancer experience of these groups of patients, but they can % of patients who disenrolled Adjusted hazard ratio (95% CI) Total Age, y (1.7 to 2.1) (1.0 to 1.4) (referent) Sex Female (0.92 to 1.0) Male/other (referent) Race Asian (1.1 to 1.5) African American (1.1 to 1.3) Hispanic (0.95 to 1.2) Other/unknown (1.4 to 2.3) White (referent) Stage of cancer at diagnosis Unstaged/unknown (1.3 to 1.7) Distant (0.98 to 1.2) Regional (0.77 to 0.91) Local (referent) Prognosis Quartile (1.4 to 1.8) Quartile (0.99 to 1.2) Quartile (1.1 to 1.3) Quartile (referent) Enrollment prior to diagnosis 30 days (4.1 to 6.1) days (4.2 to 5.3) days (3.6 to 4.7) 1to 4 years (2.7 to 3.2) 4to 10 years (1.5 to 1.9) 10 years (referent) Type of insurance Medicaid (2.9 to 4.3) Commercial (1.4 to 1.6) Self-pay (1.1 to 1.4) Medicare (referent) *CI confidence interval; not applicable. Controlling for all variables listed as well as site and year of diagnosis, censored at date of death or at the end of the first year after diagnosis if still enrolled. Includes 16 patients with sex coded as other in the health maintenance organizations electronic cancer registry files. Prognosis was based on 5-year mortality rates for each type of cancer. provide the data needed to directly contact the individuals for follow-up studies. Among the strengths of this study is the large size of the population, which enabled us to produce precise estimates of retention rates among subgroups of age and race/ethnicity, cancer type, and insurance type. The availability of complete cancer registries linked to HMO enrollment files ensured that we could thoroughly ascertain cancer diagnoses among enrollees. However, although the cancer registries of these five CRN sites have accurate models for tracking mortality, there is a possibility that the deaths of some patients who had disenrolled from an HMO may not have been captured. If that were the case, our results would, in fact, overestimate survival and, thus, underestimate the retention rate among survivors. Three major sources of cancer data in the United States SEER, the National Cancer Data Base, and the Centers for Disease Control and Prevention s National Program of Cancer Registries have been used for large-scale studies of cancer treatment. However, each source lacks information on some aspects of cancer care and excludes subgroups of the population. SEER registries collect data on treatment during the initial course of treatment (7) but do not include information on comorbidities (8). With linkage to claims data from the Medicare system, SEER s database expands to include ongoing treatment and follow-up care beyond the initial treatment period for patients aged 65 years or older. However, even with this expanded database, SEER registries do not have information on prescription drugs or access to patient records and exclude those patients who are enrolled in managed care. The National Cancer Data Base captures patients of all ages (9 11). However, this database is hospitalbased and is less able to capture information from ambulatory care settings, where cancer care is increasingly provided. The Centers for Disease Control and Prevention s National Program of Cancer Registries includes broad age and geographic ranges for cancer patients but is focused only on the initial course of treatment (12 14). The CRN is, therefore, able to fill some of the information gaps in these data sources. In conclusion, our finding of high enrollee retention rates in this cohort of HMO-enrolled patients during up to 5-years of follow-up suggests that valuable population laboratories can be built from the CRN information source and that it can serve as a base for studies of quality of care, survivorship, and longterm outcomes. REFERENCES (1) Hewitt M, Simone JV, editors. Ensuring quality cancer care. Washington (DC): National Academy Press; p. 2. (2) Hewitt M, Simone JV. Enhancing data systems to improve the quality of care. Washington (DC): National Academy Press; (3) Stewart SL, Swallen KC, Glaser SL, Horn- Ross PL, West DW. Comparison of methods for classifying Hispanic ethnicity in a population-based cancer registry. Am J Epidemiol 1999;149: (4) StatCorp. Stata Statistical Software (release 6.0). College Station (TX): Stata Corporation; Journal of the National Cancer Institute, Vol. 96, No. 2, January 21, 2004 BRIEF COMMUNICATIONS 151

5 (5) Schoenfeld D. Partial residuals for the proportional hazards regression model. Biometrika 1982;69: (6) Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994;81: (7) Hankey BF, Ries LA, Edwards BK. The surveillance, epidemiology, and end results program: a national resource. Cancer Epidemiol Biomarkers Prev 1999;8: (8) 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: IV-3 IV-18. (9) Stewart AK, Bland KI, McGinnis LS Jr, Morrow M, Eyre HG. Clinical highlights from the National Cancer Data Base, CA Cancer J Clin 2000;50: (10) American College of Surgeons National Cancer Data Base (NCDB) What is the NCDB? (Available at: cancer/ncdb/ncdbabout.html). Last accessed December 5, (11) Menck HR, Cunningham MP, Jessup JM, Eyre HJ, Winchester DP, Scott-Conner CE, et al. The growth and maturation of the National Cancer Data Base. Cancer 1997;80: (12) DHHS. Centers for Disease Control and Prevention (CDC) The National Program of Cancer Registries 2003 Program Fact Sheet Cancer Registries: The Foundation for Cancer Prevention and Control. Available at: htm. (Last accessed December 8, 2003.) (13) Bickell NA, Chassin MR. Determining the quality of breast cancer care: do tumor registries measure up? Ann Intern Med 2000; 132: (14) Malin JL, Kahn KL, Adams J, Kwan L, Laouri M, Ganz PA. Validity of cancer registry data for measuring the quality of breast cancer care. J Natl Cancer Inst 2002; 94: NOTES 1 Editor s note: SEER is a set of geographically defined, population-based central cancer registries in the United States, operated by nonprofit organizations under contract to the National Cancer Institute (NCI). Registry data are submitted electronically without personal identifiers to the NCI on a biannual basis, and the NCI makes the data available to the public for scientific research. This study was carried out under the auspices of the Cancer Research Network and was supported by Public Health Service grant CA79689 and an administrative supplement from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services. Manuscript received May 8, 2003; revised October 27, 2003; accepted November 3, BRIEF COMMUNICATIONS Journal of the National Cancer Institute, Vol. 96, No. 2, January 21, 2004