Access to health care is currently a subject of considerable concern in the United States. Approximately 34 million Americans lack health insurance,1 and many of them may delay or forgo necessary medical care because of its cost2. People without health insurance receive fewer inpatient and outpatient medical services than those with insurance,3-7 and the quality of hospital care may also be lower for uninsured patients8. The Medicaid program has improved access for poor people who qualify,9 but Medicaid enrollees are less likely than privately insured patients to have a personal physician10 and less likely to undergo certain major procedures4,6. Medicaid enrollees often have limited access to office-based physicians, in part because Medicaid's payment rates are lower than those of other insurance programs11,12.
Although there is substantial evidence that the receipt of health services varies according to insurance coverage, only a few studies have demonstrated a link between insurance coverage and health outcomes13-17. If insurance coverage is associated with differences in outcome, the relation is most likely to be evident for potentially severe illnesses that can be diagnosed and treated effectively early in their course. A plausible example is breast cancer. This disease causes the loss of more years of potential life among women under 65 years of age than any other nontraumatic condition in the United States,18 yet it is curable if detected early. Screening with breast examinations and mammography improves survival for women 50 years of age and older, and possibly for women 40 to 49 years old as well19,20.
National surveys have demonstrated that women without private insurance are less likely than privately insured women to receive cancer-screening services21-23. Hospitals that care for large numbers of uninsured patients and Medicaid enrollees may undertake less thorough staging of breast cancer,24 and patients without private health insurance may be treated less vigorously after cancer is diagnosed25. In this study we posed two questions arising from these differences in care: Do uninsured patients and those covered by Medicaid have more advanced breast cancer than privately insured patients when the disease is first diagnosed? And do these patients die sooner, on average, than privately insured patients during the 7.5 years after breast cancer is diagnosed?
METHODS
Data Sources
The New Jersey State Cancer Registry was established within the New Jersey Department of Health as a population-based incidence registry for all new cases of cancer among New Jersey residents, beginning in October 197826. State regulations require that hospitals, physicians, dentists, and clinical laboratories report cases of cancer within six months of diagnosis. Residents whose cancer is diagnosed in hospitals outside the state are identified through agreements with neighboring states. Survival data are obtained by registry staff through regular contact with reporting hospitals and physicians and by reviews of state death, motor-vehicle, and income-tax records. In addition, data on patients with cancer are regularly matched to the National Death Index, as was done most recently in July 1990 for deaths during 1987 and 1988. New Jersey also requires that hospital-discharge abstracts be reported for all hospitalizations for acute care in the state. Therefore, by linking registry records to discharge abstracts, we were able to analyze clinical outcomes for individual patients according to their insurance coverage.
The study protocol was approved by the institutional review board of the New Jersey Department of Health. Data obtained from the registry included age; race (white, black, or other, including American Indian, Asian, and unknown); marital status (married or unmarried, including single, widowed, divorced, and separated); census tract and ZIP Code of residence; dates of diagnosis, last contact, and death, if applicable, through June 1992; the reporting facility; and the summary disease stage at diagnosis. The summary stage was defined as “local” if the tumor was confined to the breast; “regional” if disease had spread to axillary or internal mammary lymph nodes, the chest wall, subcutaneous tissue, or overlying skin; and “distant” if disease had spread beyond these sites27.
From 1984 through 1988, 4 of the 21 counties in New Jersey (Essex, Hudson, Passaic, and Union) were included in the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute. For residents of these counties, more detailed data were collected about the primary tumor and metastases, allowing more precise classification of disease stage (I, IIA, IIB, IIIA, IIIB, or IV)28. Stage I disease includes tumors 2 cm or less in the longest dimension, without metastasis. Stage IIA includes tumors 2 cm or less with metastasis to movable axillary lymph nodes and tumors larger than 2 cm but no more than 5 cm without metastasis. Stage IIB includes tumors larger than 2 cm but no more than 5 cm with metastasis to movable axillary nodes and tumors larger than 5 cm without metastasis. Stage IIIA includes tumors of any size with metastasis to fixed axillary nodes and tumors greater than 5 cm with metastasis to movable axillary nodes. Stage IIIB includes all tumors with direct extension to the chest wall or skin or with metastasis to ipsilateral internal mammary nodes, but no distant metastasis. Stage IV includes all tumors with distant metastasis.
Abstracts of hospital discharges are maintained by Medix Management systems, a subsidiary of Blue Cross-Blue Shield of New Jersey. These abstracts include up to nine diagnoses from the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM). We obtained abstracts for all hospital discharges with a principal diagnosis of cancer from 1984 through 1988. From these abstracts we identified insurance coverage (private insurance, including Blue Cross, commercial plans, and health maintenance organizations; no insurance, a category that included indigent and self-paying patients; or Medicaid) and the number of coexisting diagnoses, excluding those related to the breasts (ICD-9-CM codes 174.0 through 174.9, 217, 233.0, 238.3, 239.3, 610.0 through 611.9, and V10.3), lymph nodes (codes 196.0 through 196.9 and 785.6), and metastatic cancer (codes 197.0 through 199.1).
Neither registry records nor discharge abstracts included patients' incomes, so we obtained data on median household income according to census tract and ZIP Code for New Jersey from the 1980 U.S. Census. These data were provided by the Princeton University Computing Center.
Patient Population
We identified women residing in New Jersey who were 35 to 64 years of age, in whom invasive carcinoma of the breast was diagnosed during 1985 through 1987, whose cases were reported to the New Jersey State Cancer Registry. We chose 35 years of age as a lower limit because no major organizations in the United States recommend routine screening with mammography before this age29. We chose the age of 64 as an upper limit because almost all women qualify for Medicare coverage at age 65, so relatively few older women remain uninsured or covered solely by Medicaid.
A total of 7290 women in this age group were given a diagnosis of invasive breast cancer from 1985 through 1987. Because we were interested in the outcomes according to disease stage at diagnosis and in subsequent survival, we excluded 607 women whose disease stage was not reported to the registry. We also excluded 657 women reported to the registry by hospitals in other states. These hospitals do not send discharge abstracts to New Jersey, so the type of insurance coverage was unknown for these patients. Therefore, 6026 women were eligible for the study. The women who were excluded were slightly younger than the eligible women (mean age, 51.8 vs. 52.8 years; P<0.001 by t-test), but similar in race (86.9 percent vs. 85.8 percent white, P>0.10 by the chi-square test) and marital status (62.7 percent vs. 65.0 percent married, P>0.10). Women whose cases were reported by hospitals in other states were more likely to have local disease than women whose cases were reported by hospitals in New Jersey (60.6 percent vs. 53.4 percent, P<0.001).
Data Linkage
Registry records were linked to hospital-discharge abstracts at the New Jersey Department of Health, according to the patient's surname, first initial, and date of birth. To minimize the effect of spelling variations, surnames were characterized phonetically with use of a version of the New York State Identification and Intelligence System previously modified for the SEER Program. Matches were characterized as definite if the names and dates of birth were identical in both records or probable if the names were the same and the dates of birth were within two years. The patients' names were deleted from all records before the data were analyzed.
A definite or probable match between the registry record and at least one discharge abstract was obtained for 5120 eligible women (85.0 percent). The women whose registry records were not matched to discharge abstracts were similar to the women with matched records in terms of age (mean, 53.2 vs. 52.7 years; P = 0.08 by t-test), race (85.8 percent vs. 86.2 percent white, P>0.10 by the chi-square test), and marital status (62.8 percent vs. 65.4 percent married, P>0.10).
If a woman had more than one discharge abstract matched to her registry record, we selected the abstract for the admission occurring closest (within 90 days) to the date of diagnosis listed in the registry. For 160 women, the nearest matched abstract was for an admission more than 90 days before or after diagnosis. We did not analyze the data on these women because the primary payer at the time of the admission may not have reflected their insurance status at the time of diagnosis. We also did not analyze data on 285 women with primary insurance other than the three types of interest, such as Medicare or CHAMPUS.
The final cohort was composed of 4675 women, including 4413 with a definite match between registry and discharge records and 262 with a probable match. The study population included 277 uninsured patients (5.9 percent) and 115 patients covered by Medicaid (2.5 percent), comparable to the overall proportions of 6.6 percent and 3.6 percent, respectively, among adults 45 to 64 years of age in the Northeast1. For these 4675 women, we linked 1980 Census data on median household income to registry records for the smallest residential area in which they could be categorized; this was the census tract for 76.0 percent of patients and the ZIP Code for all but 8 of the remaining patients. The patient group was divided in thirds according to annual household income, defined as low ($4,175 to $18,478), intermediate ($18,485 to $24,992), or high ($25,000 to $59,390). Because all patients were assigned the median income of their communities, a few Medicaid patients living in wealthier communities appeared to be relatively affluent.
Statistical Analysis
Our analysis had three parts. First, we compared the stage of disease at the time of diagnosis among uninsured and Medicaid patients separately with the stage of disease among privately insured patients, using the Wilcoxon rank-sum test. Second, we constructed stage-specific survival curves according to insurance coverage, by the Kaplan-Meier method. The length of time from diagnosis to death or the last follow-up contact was measured in months. The duration of potential follow-up for any individual patient ranged from 54 to 89 months, depending on the date of diagnosis, and follow-up was shorter for some patients who were not known to have died. Mortality from all causes was incorporated into the survival estimates. We compared survival curves with the Mantel-Cox log-rank test. To present the data consistently for each stage, we have displayed survival curves through 72 months after the diagnosis in our figures.
In the third part of our analysis, we estimated the adjusted risk of death among uninsured and Medicaid patients as compared with privately insured patients, using Cox proportional-hazards regression analysis. We used continuous variables for age and median household income and dummy variables for insurance coverage, race, marital status, the number of coexisting diagnoses (1, 2, 3, 4, 5, or ≥ 6), and the stage of disease at diagnosis. Because of potential age-related differences in the biology of breast cancer, we also performed stratified analyses of disease stage and adjusted risk of death among women 35 to 49 years old and 50 to 64 years old.
To determine whether more precise staging might alter the findings of the primary analysis, we performed a secondary analysis of women in the four counties covered by the SEER Program (n = 1464). Among these women, more detailed data on the extent of disease were available for 1086 (74.2 percent). In this group we analyzed the stage of disease and the adjusted risk of death according to insurance coverage by the methods described above. Because few patients were categorized as having stage IIIA or IIIB disease at presentation, we combined these two groups in the analysis (stage III).
All analyses were conducted with SAS statistical software30,31. We report 95 percent confidence intervals for the adjusted relative risks of death derived from the proportional-hazards regression analyses and two-tailed P values for all other tests.
RESULTS
The characteristics of the women in the three insurance categories are presented in Table 1
TABLE 1
Characteristics of Women in New Jersey, 35 through 64 Years of Age, in Whom Invasive Breast Cancer Was Diagnosed in 1985 through 1987, According to Insurance Coverage.. As compared with women with private insurance, uninsured women and women covered by Medicaid were younger, less likely to be white, less likely to be married, and more likely to be living in poor communities; both groups also had more coexisting diagnoses than privately insured women during the initial hospitalization.
Uninsured women and women covered by Medicaid had significantly more advanced disease than privately insured women when their disease was initially diagnosed, as indicated by the summary stage of disease for all women in the state (Table 2
TABLE 2
Summary Stage of Breast Cancer at Diagnosis, According to Insurance Coverage, in Women 35 through 64 Years of Age, in All New Jersey Counties, 1985 through 1987.). When we stratified the population according to age, significantly more advanced disease was present among uninsured women 50 to 64 years of age (P<0.001) and among women covered by Medicaid who were 35 to 49 years of age (P = 0.02) than among women with private insurance in the same age groups. Nonsignificant trends toward more advanced disease were noted among uninsured women 35 to 49 years of age (P = 0.16) and among women covered by Medicaid who were 50 to 64 years of age (P = 0.21). In the secondary analysis of the four counties in the SEER Program, in which more precise staging was recorded, uninsured women and women covered by Medicaid also had significantly more advanced disease than privately insured women (Table 3
TABLE 3
Stage of Breast Cancer at Diagnosis, According to Insurance Coverage, in Women 35 through 64 Years of Age, in Essex, Hudson, Passaic, and Union Counties, 1985 through 1987.).
Survival during the 54 to 89 months after diagnosis was significantly worse for uninsured patients and those covered by Medicaid than for privately insured patients with local disease (P<0.001 for both comparisons) (Figure 1
FIGURE 1
Kaplan-Meier Survival Estimates According to Insurance Coverage for Women 35 through 64 Years of Age in Whom Local-Stage Breast Cancer Was Diagnosed in New Jersey, 1985 through 1987.). Survival was also significantly worse for uninsured patients and those covered by Medicaid who had regional disease (P<0.001 for both comparisons) (Figure 2
FIGURE 2
Kaplan-Meier Survival Estimates According to Insurance Coverage for Women 35 through 64 Years of Age in Whom Regional-Stage Breast Cancer Was Diagnosed in New Jersey, 1985 through 1987.). Survival did not differ significantly according to insurance coverage in patients with distant metastases at presentation (P>0.10 for all comparisons) (Figure 3
FIGURE 3
Kaplan-Meier Survival Estimates According to Insurance Coverage for Women 35 through 64 Years of Age in Whom Distant-Stage Breast Cancer Was Diagnosed in New Jersey, 1985 through 1987.). Survival also did not differ significantly between uninsured and Medicaid patients with any of the three stages of disease (P ≥ 0.10 for all comparisons).
In the primary multivariate analysis of all patients, both uninsured patients and patients covered by Medicaid had a higher adjusted risk of death than privately insured patients (Table 4
TABLE 4
Adjusted Relative Risk of Death among Uninsured Women and Women Covered by Medicaid, as Compared with Privately Insured Women.). When we stratified this analysis according to age, using privately insured patients as the reference group, the adjusted risk of death was significantly greater for uninsured patients 35 to 49 years of age (relative risk, 1.57; 95 percent confidence interval, 1.11 to 2.24), uninsured patients 50 to 64 years of age (relative risk, 1.43; 95 percent confidence interval, 1.10 to 1.86), and patients with Medicaid coverage 35 to 49 years of age (relative risk, 1.59; 95 percent confidence interval, 1.02 to 2.49), but not for Medicaid enrollees 50 to 64 years of age (relative risk, 1.28; 95 percent confidence interval, 0.84 to 1.94). When more precise staging was used in the secondary analysis of patients in the SEER Program, the adjusted risk of death was significantly greater for both uninsured patients and patients with Medicaid coverage than for privately insured patients (Table 4).
DISCUSSION
In this study women without health insurance and those covered by Medicaid had more advanced breast cancer than women with private health insurance when the disease was initially diagnosed. The survival of those women was also worse than that of privately insured women with local and regional disease, and the association of insurance coverage with survival persisted in a multivariate analysis in which adjustment was made for numerous potential confounders.
Reduced access to care may be an important reason for the more advanced disease of uninsured patients and patients covered by Medicaid. Such patients are less likely than privately insured patients to have a primary care physician who can ensure that they are screened for cancer10. Having a primary care physician is strongly associated with the use of mammography,32,33 and a physician's recommendation is the most important factor prompting women to undergo mammographic screening34,35.
Less complete staging among uninsured and Medicaid patients could produce the appearance of worse stage-specific survival, even in the absence of true differences. In one study, for example, hospitals that cared for large numbers of uninsured patients and patients covered by Medicaid were less likely than others to evaluate hormone receptors in cancerous breast tissue24. However, if uninsured patients and those covered by Medicaid undergo less complete staging, then we have underestimated their more advanced disease stage at diagnosis. Incomplete staging may also lead to inadequate therapy.
Differences in treatment may contribute to the worse survival of patients without private health insurance. In a study of lung cancer, patients without private insurance were less likely than privately insured patients to undergo surgery, chemotherapy, or radiation therapy25. Although the subsequent survival of the two groups did not differ, treatment differences could be more consequential for conditions such as breast cancer that are more frequently detected at an early stage.
Lead-time bias is a possible explanation for the better survival of privately insured patients. If their disease is diagnosed sooner (i.e., with a longer lead time) within each stage than that of uninsured patients and patients with Medicaid coverage, their subsequent survival could appear to be longer, even if earlier treatment is not prolonging their lives. Length-time bias is another possible explanation. Higher screening rates among privately insured women may lead to the detection of cancers that spread more slowly (i.e., take longer to become symptomatic) and are less likely to be lethal within a given stage36. These potential biases can be evaluated in randomized trials,19,20,37 but we cannot evaluate their effects with our observational data.
No comments:
Post a Comment