Abstract
Background: The relationship between socioeconomic status and health care disparities in the incidence of brain tumors is unclear.
Objective: To identify the associations between age, sex, and Medicaid enrollment and the incidence of primary malignant brain tumors in Michigan in 1996 and 1997.
Methods: Records were obtained from the Michigan Cancer Surveillance Program on the 1,006 incident cases during this period and cross-checked with Medicaid enrollment files.
Results: Persons enrolled in Medicaid were more likely than non-enrolled persons to develop a malignant brain tumor of any type, a glioblastoma multiforme, and an astrocytoma for certain subgroups. In addition, incidence rates for malignant brain tumors in persons enrolled in Medicaid peaked at a younger age.
Conclusion: Sociodemographic status may be associated with cerebral malignancy and should be considered when targeting treatment and educational interventions at persons at risk.
Although primary malignant brain tumors (PMBT) occur less frequently than other malignancies,1 the severe morbidity and mortality associated with the diagnosis of PMBT make identifying factors associated with their incidence an important area of neuroscientific research. Results of studies on risk factors of PMBT have been inconsistent. Incidence rates of PMBT vary according to age and sex,2 occupational factors,3,4⇓ head trauma,5 and diet.6 The relationship between low socioeconomic status (SES), often accompanied by disparities in access to health care, and the incidence of PMBT is unclear. Although the discovery of sociodemographic and socioeconomic risk factors for developing a brain tumor is not tantamount to discovering a cause for cerebral malignancy, the existence of disparities can determine causal factors, while allowing for more effective targeting of patient groups that need treatment and support. Laying the groundwork for epidemiologic efforts in identifying factors that place persons at differential risk for cerebral malignancy remains an important objective for neuro-oncology research, as emphasized by the National Cancer Institute’s Brain Tumor Progress Review Group.7
This study was conducted to explore potential disparities in the incidence of brain tumors between Medicaid enrollees and other population members. Specifically, the project was designed to answer the following questions: 1) Controlling for age and sex, does Medicaid enrollment predict variations in the incidence rates of PMBT in Michigan during 1996 and 1997? 2) Are these variations in incidence rates the same across histologically distinct subgroups?
Despite the association between the incidence of brain tumors and age8-11⇓⇓⇓ and sex,2 the extent to which SES is related to the incidence of PMBT remains unclear. Positive links between SES and incidence rates of cerebral malignancies have been reported. Researchers evaluating brain tumors in white men in Washington over a 10-year period found a consistent pattern of increased risk for all brain tumors as SES increased.12 In Australia13 and New Zealand,14 and other countries,15 cerebral malignancy was found to be more common among those in affluent areas and those in higher social classes.
Not all investigators have been able to replicate the association between higher SES and an increased incidence of cerebral malignancy. Researchers failed to find an association between incidence and SES in naval military men.16 Preston-Martin et al. did not identify a relationship between SES and the incidence of primary brain tumors in Australia.11 The mixed results may be due to the use of different proxies for SES. For example, most researchers have used occupation and zip code (or postal code) to estimate SES. Particular ecological measures, such as zip codes, are not necessarily strong indicators of individual SES, and occupation may not be a good direct measure of the effect of SES on disparities in health care, leading to difficulty in establishing the validity of indicator variables as proxies for SES.
To clarify the role of individual SES in the incidence of PMBT, we conducted this study. As a proxy for SES, we focused on Medicaid enrollment among Michigan residents diagnosed with PMBT during 1996 or 1997. Enrollment in Medicaid is an individual-level indicator, which can serve as a proxy for very low income. The 1996 (1997) poverty thresholds were $7,995 ($8,183) for a single person household, increasing to $10,233 ($10,473) for a two-person household, with similar increases for additional household members.17 In Michigan, to be eligible for enrollment in Medicaid, a person would have to live in a household with income below 150% of the poverty threshold. Spend-down provisions are such that it would be very difficult for middle-class individuals to become eligible for Medicaid within a few months. A person may also be eligible for Medicaid as a result of physical or functional disability that is secondary to the tumor or its treatment. However, physical disability resulting from the tumor would have to be severe enough and last long enough (at least 12 months) that the patient could qualify for Medicaid. Thus, Medicaid enrollment, and enrollment prior to diagnosis in particular, would appear to be an adequate proxy for SES. With a focus on Medicaid enrollment, it is possible to examine the health care disparities related to very low income and its possible relation to the incidence of PMBT.
Methods.
Population.
Data for this analysis were obtained from a larger project, which consisted of identifying all incident cancer cases in Michigan during the calendar years 1996 and 1997.18 As part of this project, information on all persons in Michigan diagnosed with a malignant CNS neoplasm during 1996 and 1997 was retrieved from the Michigan Cancer Surveillance Program (MCSP). This registry receives reports from all hospitals and laboratories as well as supplemental information from physicians, hospices, nursing homes, and other facilities. Periodic on-site reviews are conducted by the MCSP with institutions submitting data to verify that at least 95% of the cancer incidences are reported. Institutions that do not meet this standard are subject to further, ongoing review until the MCSP is confident that a 95% accuracy rate is maintained.19
Based on this data set, 1,290 cases of malignant CNS neoplasms were diagnosed during this 2-year period. (Because the MCSP only collects data on malignant tumors, benign brain tumors were not included in the analysis.) Only newly diagnosed and primary brain tumors were considered for analysis to avoid counting persons with recurrent tumors as additional primary cases. Newly diagnosed was defined as a patient identified in 1996 or 1997 with no prior record of brain malignancy from 1992 through 1995 (with the exception of patients diagnosed with a malignant meningioma, in which there may have a previous diagnosis of a benign meningioma; n = 20 or 2% of the sample). In addition, all primary sites that could potentially include spinal tumors were identified via ICD-0-II code and were excluded from the analysis. Finally, persons younger than 25 years and those older than 84 years were excluded from the analysis because of extremely low incidence rates in these age groups and because Medicaid eligibility criteria for children are different from those for adults.
The application of these criteria resulted in a data set containing information about 1,006 individual cases of PMBT diagnosed in 1996 and 1997. The Michigan Department of Community Health matched these cases to Medicaid enrollment files using social security number; first and last name; month, day, and year of birth; sex; and address. A match was considered valid if both data sets contained the same name, date of birth, sex, and either the first five digits or the last four digits of the social security numbers. (A more complete description of the procedures involved in obtaining case data is presented by Bradley et al.18). For the majority of analyses, Medicaid enrollment was defined as patients who had filed for, qualified for, and received Medicaid assistance for at least 1 month between 1996 and 1997 (thus analyses included persons who were enrolled in Medicaid both prior to and after diagnosis). Separate analyses were also performed using only persons who were enrolled in Medicaid prior to diagnosis to address spend down bias, and are described later in Results.
Statistical analysis.
Two-year averaged incidence rates for brain tumors were obtained by dividing the number of newly diagnosed brain tumors for a particular population subgroup by the total subgroup population at risk. For example, the incidence rate for 75- to 79-year-old persons was calculated as the number of brain tumors that occurred in Michigan residents of this age during 1996 and 1997 (n = 108) by the number of person-years of exposure for residents in this age group during 1996 and 1997 (i.e., the number of residents living in Michigan aged 75 to 79 in 1996 and 1997; n = 496,268), multiplied by 100,000 (table 1). Population data for 1996 and 1997, stratified by age, sex, and Medicaid enrollment, were obtained from the Office of the State Demographer, Michigan Department of Management and Budget, and, following the usual conventions,20 mid-year estimates were used as proxy for person-years.
Table 1 Number and incidence rates of brain tumors by histologic classification, sex, Medicaid status, and age in Michigan among persons 25–84 years old (1996/1997)*
Relative risk ratios to determine the comparative risk in any two subgroups were obtained by dividing the incidence rate for a particular subgroup by the incidence rate of the comparison subgroup. For example, the relative risk of brain tumors associated with Medicaid enrollment among men aged 75 to 79 was obtained by dividing the incidence rate for brain tumors in this group by the incidence rate for brain tumors in men aged 75 to 79 who were not enrolled in Medicaid.
For the estimation of multivariate models, we employed Poisson regression models, in which the predicted outcome (μi) refers to the count of rare events (such as PMBT). In this regression model (μi = E(yi|xi) = exp(xiβ), the number of events or incidences of PMBT has a Poisson distribution, in which outcome events are assumed to occur independently of each other, but with equal probability within each exposure category defined by the vector of independent predictors.20,21⇓ The independent variables employed in the analysis include the two well-established predictors of the incidence of brain tumors, age (12 5-year categories) and sex, as well as Medicaid status (enrolled versus not enrolled), yielding 12 × 2 × 2 = 48 population subgroups. Race was not included as a potential predictor variable, because the overwhelming majority of cases (92%) in this data set were white (7% were African American, and 1% were other/unknown), providing too few cases to produce any meaningful differences.
Several Poisson regression models were run to model the effects of age, sex, and Medicaid enrollment on the incidence of brain tumors. Outcomes included the combined PMBT diagnoses, followed by separate models for the largest histologic subgroups—glioblastoma multiforme (GBM) and astrocytomas (including astrocytoma not otherwise specified [NOS], anaplastic astrocytoma, diffuse astrocytoma, pilocytic astrocytoma, and unique astrocytoma variants). All other types of primary malignant brain tumors occurred too infrequently to obtain reliable rate estimates for the Michigan population during this 2-year period. Only cases for which the histologic classification was confirmed by tissue examination were included in the analysis. In order to accommodate possible nonlinear relationships between age and tumor incidence rates, higher order polynomials (quadratic and cubic age terms) were also included, if shown significant at p < 0.05 (higher order polynomials are useful in determining the general shape of a function when there are more than two inflection points). All two-way interactions among the predictor variables were explored, and coefficients for significant interactions are reported.
Results.
Table 1 shows the occurrences and incidence rates for all PMBT and its subcategories in Michigan during 1996 and 1997, which are similar to results reported by other states across the country.10,22⇓ The incidence rate for all PMBT combined was 8.1 per 100,000. Incidence rates exhibit a monotone increase from a low of 3.1 per 100,000 among 25- to 34-year-old residents (lower age groups were aggregated in table 1 into 10-year intervals) to a high of 21.8 among 75- to 79-year-old residents. However, in the oldest 5-year age group (80 to 84), the rates decline to 18.3 per 100,000. Incidence rates for all PMBT combined are higher among men (9.1 per 100,000) than women (7.2), and higher among residents enrolled in Medicaid (14.2 per 100,000) than among those not enrolled in Medicaid (7.5).
Table 2 presents the results from several Poisson regressions. In all models, the outcome variable is a count of incident PMBT (of various categories) and exposure is defined in terms of the number of mid-year Michigan residents (1996 to 1997) at risk for a PMBT in each of the 48 population groups categorized in terms of their age, sex, and Medicaid status.
Table 2 Poisson regressions: brain tumor incidences on age, sex, Medicaid enrollment*
In Model 1, the outcome is incidences of PMBT of all types combined. A test of the higher-order polynomials resulted in a significant (p = 0.01) quadratic age effect indicating a single inflection point in the curvilinear relationships between age and the incidence rates (figure 1). Two interaction effects were found to be statistically significant: one between the (squared) age variable and Medicaid status (p = 0.01) and the other between sex and Medicaid status (p = 0.01). Overall, the specified Poisson regression model fit the data well as evidenced by the nonsignificant goodness of fit (GoF) p value (GoF χ2 = 56.05; p = 0.06; pseudo R-squared = 069).23 The nonsignificant χ2 indicates that deviations of observed from predicted cases could be due to sampling chance. Thus, the model can be considered an adequate representation of the pattern in the data.
Figure 1. Predicted overall brain tumor rates. Filled circle = IR for women on Medicaid; open circle = IR for women not on Medicaid; filled square = IR for men on Medicaid; open square = IR for men not on Medicaid.
The graph in figure 1 highlights the estimation results presented in table 2. Consistent with the literature,8 incidence rates for PMBT are lower for women than for men, but as indicated by the interaction effect between Medicaid status and sex, this difference is larger among persons enrolled in Medicaid and smaller among persons not on Medicaid. The graph in figure 1 also illustrates that brain tumor incidence rates vary with age, which confirms other reported findings.9 Among the Medicaid population, incidence rates reach a maximum at age 67 to 68; among the non-Medicaid population, they continue to rise beyond the age of 84.
When comparing incidence rates among subgroups, it can be seen that the relative risk of being diagnosed with a PMBT varies with age. When the independent effect of Medicaid enrollment is evaluated among men only, the Medicaid population has a consistently higher incidence rate than the non-Medicaid population. However, due to the interaction effect between Medicaid status and the squared age term (p = 0.01), the differences are greatest at age 27 (IRR = 5.8) and disappear at age 83 (IRR = 1). (At age 66, the lower 95% confidence limits for the Medicaid enrolled men start to overlap with the upper 95% confidence limits for non-Medicaid men, indicating significant differences at least for the age range below age 66.) Among women, brain tumor incidence rates for the Medicaid enrolled population exceeded those for the non-Medicaid population, except for women older than 74. (Significant differences between Medicaid and non-Medicaid women occur only below age 60.) Beyond this age, women’s incidence rates were higher among the non-Medicaid population, with significant differences starting at age 79.
The Medicaid variable employed in the foregoing model defined as on Medicaid if they applied for, qualified for, and were granted Medicaid assistance for at least 1 month during 1996 and 1997. Even though it was mentioned that spend-down provisions for Medicaid are unlikely to apply in the case of the relatively short survival times for the majority of persons with PMBT, the use of Medicaid as a proxy for SES may be questioned due to qualification for Medicaid on the basis of physical/functional disability, whenever Medicaid enrollment occurs after the diagnosis of PMBT. Among the 168 cases identified as Medicaid enrollees, 81 were enrolled prior to diagnosis; 87 enrolled in Medicaid during the month of diagnosis or within maximally 11 months after diagnosis. Based on this information, we repeated the analysis confining the definition of on Medicaid to only those persons who had received at least 1 month of Medicaid assistance prior to their PMBT diagnosis. The results of this analysis are shown as Model 1b in table 2. Most of the coefficients are of similar magnitude, except a decline in the coefficient associated with the Medicaid variable itself, which is because the not on Medicaid group now contains individuals enrolled after their diagnosis, thus reducing the contrast. Given the similarity, and the need to maintain adequate sample sizes for the analyses involving histologic subgroups of PMBT, we continued to employ the broader definition of Medicaid enrollment for the remainder of the analyses.
Regarding variations in incidence rates due to age, sex, and Medicaid enrollment consistent among histologically distinct subgroups, only GBM (n = 507, IR = 4.1 per 100,000) and the combined astrocytoma diagnoses (n = 163, IR = 1.3 per 100,000) were sufficiently frequent events to warrant separate analyses. Model 2 in table 2 and figure 2 shows the results for the separate GBM incidence rates only. GoF indices for the GBM model demonstrated acceptable fit (GoF χ2 = 55.05; p = 0.07; pseudo R-squared = 0.72). Concerning the incidence rates among the Medicaid and non-Medicaid population, figure 2 shows the incidence rates intersecting at age 76 to 77. (However, non-overlapping 95% CI only occur in ages below 67.) Thus, with increasing age, both the sex and Medicaid status differences disappear. The relationship between age, Medicaid status, and incidence rates for GBM is illustrated in the graph in figure 2. The incidence rates are staggered, such that incidence rates peak earliest among men enrolled in Medicaid (at age 61), followed by women on Medicaid (65), men not on Medicaid (73), and women not on Medicaid (79). At the same time, the earlier the incidence rates peak, the higher the maximum rates; incidence rates reach 21 per 100,000 among men enrolled in Medicaid, followed by 15 per 100,000 among women on Medicaid (65), 13 per 100,000 among men not enrolled in Medicaid, and finally 11 per 100,000 among women not enrolled in Medicaid.
Figure 2. Predicted glioblastoma tumor rates. Filled circle = IR for women on Medicaid; open circle = IR for women not on Medicaid; filled square = IR for men on Medicaid; open square = IR for men not on Medicaid.
Incidence rates for other astrocytoma tumors reveal yet another pattern. The GoF indices for this demonstrated acceptable fit (GoF χ2 = 47.85; p = 0.25; pseudo R-squared = 0.21). Omitted from the prediction equation is the sex term, because neither the main effect nor any of the interaction terms involving sex show significant risk differences. In addition, incidence rates for other astrocytoma tumors exhibit an age-Medicaid status interaction. As figure 3 illustrates, among persons enrolled in Medicaid, astrocytoma incidence rates are highest at a young age (25 to 29) at 3.5 per 100,000, then decline to about 1.5 per 100,000 at age 42, stabilize until age 67, only to decline further in higher age groups. By contrast, among the non-Medicaid population, after a small initial decline between ages 25 and 32, incidence rates tend to rise until the age of 74, after which there is a small drop-off. Beyond the age of 58, incidence rates for other astrocytoma tumors in the non-Medicaid population exceed those in the Medicaid population. (Non-overlapping 95% CI occur below the age of 47 and above the age of 67.)
Figure 3. Predicted astrocytoma tumor rates. Filled triangle = IR for people on Medicaid; open triangle = IR for people not on Medicaid.
In general, our analysis demonstrated an overall effect of age and sex on the incidence of PMBT and GBM and an overall effect of age on the incidence of astrocytomas. Disparities in health care and SES, defined for the study as Medicaid status, also demonstrated an effect on overall incidence, incidence of GBM, and incidence of astrocytoma, although these effects varied by subgroup.
Discussion.
While existing studies have related the risk of developing brain tumors to age, sex, and SES, the more complex interactions among these predictors have remained unexplored. Using Michigan data from 1996 and 1997, patterns in incidence rates for malignant brain tumors, including rates for two major subcategories of PMBT, GBM and astrocytomas, were examined. Consistent with other findings,8 our analysis confirmed a generally higher risk of developing a PMBT among older adults, except in the oldest age groups (75+). Furthermore, men had higher incidence rates for most PMBT, but not for astrocytomas. Different from some other reports,12-14⇓⇓ we found that persons with low SES (defined as persons who were enrolled in Medicaid) were generally more likely than those who were not enrolled in Medicaid to develop a malignant brain tumor of any type.
From these Michigan data, it is apparent that socioeconomic disparities in the risk of PMBT diagnosis are greatest among the young. Among men under 44 years of age, those enrolled in Medicaid have incidence rates at least four times higher than those not enrolled in Medicaid. Among women under 44 years of age, the differences are narrower but in the same direction.
Female Michigan residents under 44 years of age enrolled in Medicaid are at least 2.56 times more likely to be diagnosed with PMBT than their peers not enrolled in Medicaid. Yet, for both men and women, these socioeconomic disparities disappear with increasing age. Similarly, while both men and women enrolled in Medicaid have a higher incidence rates of GBM than those not on Medicaid, these differences become smaller with age, with no significant differences in incidence rates of GBM by Medicaid status or sex for persons over 67.
Incidence rates for the combined astrocytomas showed a slightly different pattern in that they did not vary by sex. Yet, with respect to Medicaid enrollment and age, the data confirm the overall pattern of Medicaid status adding to the risk of diagnosis at younger ages (<47), but non-Medicaid rates exceeding Medicaid rates in persons older than 67 years. We can only speculate that the comparatively greater decline in incidence rates with age among the Medicaid population is actually the result of greater mortality at an earlier age: poverty may accelerate the onset of PBMT among those who are biologically predisposed and may thus deplete the ranks of the predisposed before old age. It should be noted, however, that low power to detect interactions, particularly in histologic subgroups, may be responsible for differences in the curves in figures 1 through 3⇑⇑.
In this analysis, we employed the variable Medicaid enrollment as a proxy for low SES and other associated characteristics, rather than as a causal variable. Because the short survival periods for most persons diagnosed with PMBT afford limited spend-down opportunities, Medicaid enrollment as a proxy for low SES appears acceptable, but it is not known which of the many factors associated with Medicaid enrollment are linked to the incidence of PMBT. SES is associated with environmental factors (such as exposure to toxins), quality of nutrition and shelter, education and health-related behaviors, variations in stress or other psychosocial experiences, or even genetic factors.24 Because we did not collect data on levels of income for persons in the non-Medicaid group, we do not know whether the incidence of brain tumors varies according to income after persons with low income are removed from the group. Although we differentiated persons who had extremely low SES, we did not differentiate those who had an extremely high SES. Thus, there may be a U-shaped relationship between SES and incidence of PMBT—those with both very high and very low SES may be at risk for increased incidence of PMBT (similar to other reports in the literature of the association between higher SES and increased incidence of PMBT).
In addition, because of low incidences of brain tumors in general, even population data from a fairly populous state do not provide sufficient numbers to control for more than two to three risk factors or allow for the examination of different trends among various specific tumor subgroups. For instance, while both GBM and astrocytomas have declining incidence rates among elderly persons, this is not the case for all brain tumors combined, which suggests that at least some of the other brain tumor diagnoses follow a different age pattern. Using a sample of this size may also affect significance testing and GoF indices (ours were just over the acceptable levels). Using data from multiple states or for a greater number of years would allow for further significance testing as well as investigation into the role that race or ethnicity might play as a potential confounding factor. With the very small number of incident PBMT cases among minority residents in Michigan in 1996 and 1997, the effects of ethnicity and race could not be separated from socioeconomic effects.
Acknowledgments
Supported by Michigan State University Institute for Health Care Studies, Department of Human Medicine, Department of Family Practice, and the Michigan Department of Community Health–Cancer Control (the project “Linking clinical and claim file data sets for a cohort of Medicaid patients to test policy analyses for screening, quality of care outcomes,” 1999–2001, subproject of “Cancer prevention, outreach and cancer control for patients in Medicaid managed care and community-based programs”), National Institute of Nursing Research, National Institutes of Health (F31 NR08069), American Cancer Society, Walther Cancer Institute, and the Michigan State University College of Nursing.
- Received May 13, 2003.
- Accepted January 16, 2004.
References
Letters: Rapid online correspondence
- Primary malignant brain tumor incidence and Medicaid enrollment
- Reply to Greenberg
REQUIREMENTS
If you are uploading a letter concerning an article:
You must have updated your disclosures within six months: http://submit.neurology.org
Your co-authors must send a completed Publishing Agreement Form to Neurology Staff (not necessary for the lead/corresponding author as the form below will suffice) before you upload your comment.
If you are responding to a comment that was written about an article you originally authored:
You (and co-authors) do not need to fill out forms or check disclosures as author forms are still valid
and apply to letter.
Submission specifications:
- Submissions must be < 200 words with < 5 references. Reference 1 must be the article on which you are commenting.
- Submissions should not have more than 5 authors. (Exception: original author replies can include all original authors of the article)
- Submit only on articles published within 6 months of issue date.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- Submitted comments are subject to editing and editor review prior to posting.
You May Also be Interested in
Hemiplegic Migraine Associated With PRRT2 Variations A Clinical and Genetic Study
Dr. Robert Shapiro and Dr. Amynah Pradhan