Cancer, Medicaid enrollment, and survival disparities

Authors


Abstract

BACKGROUND

The current article examined survival for adults < 65 years old diagnosed with breast, colorectal, or lung carcinoma who were either Medicaid insured at the time of diagnosis, Medicaid insured after diagnosis, or non-Medicaid insured.

METHODS

The authors hypothesized that subjects enrolling in Medicaid after they were diagnosed with cancer would explain disparate survival outcomes between Medicaid and non–Medicaid-insured subjects. The authors used the Michigan Tumor Registry, a population-based cancer registry covering the State of Michigan, to identify subjects who were diagnosed with the cancer sites of interest (n = 13,740). The primary outcome was all cause mortality over an 8-year time period.

RESULTS

Subjects who enrolled in Medicaid after diagnosis with cancer had much lower 8-year survival rates relative to Medicaid-enrolled and non-Medicaid subjects. These reductions in survival were partly due to a high proportion of lung carcinoma and late-stage cancers within the sample of subjects who enrolled in Medicaid after diagnosis. The likelihood of death was two to three times greater for subjects enrolled in Medicaid relative to subjects who were not enrolled in Medicaid once the analysis was stratified by cancer site and stage.

CONCLUSIONS

Disparities in cancer survival were apparent between subjects enrolled in Medicaid and subjects not enrolled in Medicaid. From a policy perspective, cancer survival in the Medicaid population cannot be improved as long as 40% of the population enrolls in Medicaid after diagnosis with late-stage disease. Cancer 2005. © 2005 American Cancer Society.

A substantial body of research provides evidence that disparities in cancer survival exist for low-income individuals.1–8 The lack of health insurance has been identified as an important liability for persons of low socioeconomic status (SES) with cancer.9 Nevertheless, the presence of health insurance, namely, Medicaid, alone may not be sufficient at narrowing the survival gap between upper-income and lower-income persons with cancer. Ayanian et al.,8 for example, found that both uninsured women and those insured by Medicaid presented with more advanced-stage breast carcinoma and had poorer survival than privately insured women. Medicaid-insured and uninsured women had approximately the same risk of death. To date, research has not clarified how the mix of personal, insurance, and policy factors combine to explain why low-income, but insured, persons experience poor survival once diagnosed with cancer.

The current article examines how the timing of Medicaid enrollment relative to a cancer diagnosis affects survival rates for adults age < 65 years diagnosed with disease at 3 common sites of cancer. Specifically, adults enrolled in Medicaid before diagnosis with cancer are compared with adults who enrolled in Medicaid the same month of or after a cancer diagnosis. These Medicaid-insured adults are the target of this investigation. These two groups of Medicaid recipients are likely to be similar to each other in SES and patterns of health-seeking behaviors, and both groups are likely to have been uninsured or underinsured before Medicaid enrollment. Thus, comparisons in rates of survival between these groups will partly adjust for behavioral and access issues so that the Medicaid survival affect on those enrolled before diagnosis can be observed. We also compare the rates of survival for these two groups of Medicaid recipients who differ by the timing of their enrollment with a third group of subjects with the same sites of cancer, but who were not enrolled in Medicaid.

One plausible explanation for the previously observed survival differences between Medicaid patients and non–Medicaid-insured patients is “lead-time bias.” Lead-time bias occurs when cancer is detected earlier, and therefore patients appear to have longer survival times, even in the absence of any real survival benefit.10, 11 For example, if the non-Medicaid population has cancer detected through screening, the cancer is likely to be in the early stage. Therefore, the non-Medicaid population has an inherent survival advantage. In contrast, cancer detected through the presentation of symptoms is likely to be in an advanced stage. Building on earlier work demonstrating that those enrolling in Medicaid after diagnosis had later-stage cancers relative to those enrolled in Medicaid before diagnosis and those who were not insured by Medicaid,12 we hypothesized that subjects enrolling in Medicaid after they were diagnosed with cancer would explain disparate survival outcomes between Medicaid and non–Medicaid-insured subjects.

Unlike patients enrolled in Medicaid after a cancer diagnosis, those enrolled in Medicaid before their diagnosis have had an opportunity to benefit from covered services (e.g., screening and treatment) and thus should have longer cancer survival more closely approximating rates observed from patients with no evidence of Medicaid enrollment, assuming that their cancer was detected early and treated promptly and appropriately. Here, we make the simplifying assumption that the main barrier between low-income individuals and cancer screening, treatment, and prolonged survival is the lack of health insurance rather than treatment inequalities, compromised health status, or other reasons contributing to high mortality rates. Under these circumstances, policies that provide cancer screening and subsequent health insurance based on diagnosis (such as Medicaid coverage for women diagnosed under the provisions of the Breast and Cervical Cancer Prevention and Treatment Act, P.L. 106-354) can be effective at reducing disparities in cancer survival for low-income individuals.

For adults who are not pregnant or caring for young children and seeking Medicaid enrollment, Medicaid requires that these individuals have a disabling condition expecting to last ≥ 1 year and that they meet asset and income requirements. If cancer is a Medicaid enrollee's qualifying disabling condition, then the cancer, by definition, has to be advanced. Treatments, even if they are state of the art, for late-stage cancers are less likely to offer prolonged survival benefits. Thus, the enrollment criteria a priori burden Medicaid with a population that is likely to have poorer survival possibilities relative to other insurers. A similar situation has been illustrated for patients with human immunodeficiency virus infection enrolled in Medicaid programs.13

MATERIALS AND METHODS

We used the cancer registry data from the Michigan Cancer Surveillance Program (MCSP) to identify subjects who were diagnosed in 1996 and 1997 with either female breast, colorectal, or lung carcinoma. All in situ and invasive malignancies are reportable to the cancer registry, except basal, squamous, or papillary carcinomas of nongenital skin. Reports are required of all Michigan hospitals and laboratories. In addition, supplemental information on cancer cases is received from physicians, nursing homes, hospices, freestanding radiotherapy centers, and other facilities. Information on resident cases diagnosed or treated outside of Michigan is obtained through interstate exchange and the cancer registry is a member of the North American Association of Central Cancer Registries (NAACCR). The Michigan cancer registry is certified by NAACCR as meeting all NAACCR standards for quality, completeness, timeliness, and unresolved duplicate records. The completeness of the MCSP registry ranges from 95–99%.

Personnel at the Michigan Department of Community Health matched subjects from the cancer registry to the 1996 and 1997 Medicaid enrollment files. The procedures used for matching records between the files are described elsewhere.14 Vital status information (e.g., death, date, and cause of death, if applicable) was available through December 2003.

From the linked dataset, we selected subjects for whom the first, primary site of diagnosis was either the female breast, colon/rectum, or lung and whose age was < 65 years old at the time of diagnosis. Subjects' race/ethnicity was defined as white, African American, and other/unknown. We excluded subjects classified as other/unknown because the racial and ethnic groups in this category vary widely, and we were concerned that combining such disparate groups together or with whites or African Americans may lead to biased results. Subjects diagnosed with cancer during the same month as death were excluded from the analysis. The exact date of death was not available to researchers to preserve subject confidentiality. We also excluded 789 subjects whose cancer stage was unknown. The remaining sample size was 13,740. Medicaid insured 14.4% (n = 1972) of the subjects at some point during the 1996 and 1997 study period.

The Medicaid enrollment file contained subjects' complete historical enrollment information from before the 1996 and 1997 study period. We caution the reader that enrollment in the same month of diagnosis, in many cases, indicates that once the beneficiary was determined to meet enrollment criteria, which likely occurred some months after a cancer diagnosis, Medicaid enrollment was made retroactive to the date of diagnosis. If a subject was enrolled in Medicaid during the same or later month and year as the month and year of diagnosis, we searched the Medicaid enrollment files to ensure that the subject was not enrolled in Medicaid during the months before diagnosis and if not, we labeled this subject as late enrolled. Approximately 42% of Medicaid subjects in our sample were late-enrolled.

If the subject was enrolled in Medicaid ≥ 1 month before the date of cancer diagnosis, the subject was considered to be Medicaid enrolled. Approximately 58% of Medicaid-insured subjects in our sample fit this description. Subjects were enrolled in Medicaid an average of only 9 months before diagnosis. Subjects were considered to be non-Medicaid enrolled if they were not enrolled in Medicaid at any time before or during the study period. A few subjects classified as non-Medicaid may have enrolled in Medicaid after 1997. The presence of these individuals would reduce the differences observed between the non-Medicaid and late-enrolled groups. However, we believe the effect of these individuals on the analysis to be insignificant.

Survival after a cancer diagnosis is the primary study outcome. Survival time was measured in months from the month of diagnosis. Subjects were diagnosed with cancer between January 1996 and December 1997. Survival data were available on these subjects from January 1996 to December 2003 for a maximum of 96 months (8 years). During this time, 5428 subjects died. We examined all cause mortality although 92% of subjects had cancer listed as their cause of death. Survival was studied as a function of Medicaid enrollment and timing of Medicaid enrollment relative to a cancer diagnosis.

Control Variables

Important covariates associated with cancer survival include age at diagnosis, race, and cancer site and stage. Age at diagnosis was entered into equations as a continuous variable. Race was entered as a dichotomous variable for African American or white.

Research has shown that Medicaid-insured individuals are more likely to be diagnosed with late-stage disease relative to individuals who are not insured by Medicaid.8, 12, 14 Lead-time bias can confound survival, if subjects in one of the three samples were systematically diagnosed at an earlier stage in the course of their disease. A common way to avoid or reduce the effects of lead-time bias on cancer survival is to control for cancer stage at diagnosis.15 Richards et al.15 found that a delay in diagnosis was associated with poor survival when cancer stage was excluded from the analysis, but not when stage was included. The Surveillance, Epidemiology, and End Results summary stage designating in situ, local, regional, and distant cancer stages was available for most subjects. More detailed staging criteria such as the American Joint Committee on Cancer TNM classification system applied to pathologic data were not available in the dataset. We categorized cancer stage as early if patients were diagnosed with in situ or local cancer. Subjects with regional or distant cancers were considered to have late-stage cancers. Ideally, we would use all cancer stages in our model, but there were too few Medicaid-insured subjects with in situ cancers (3% of all late-enrolled subjects and 7% of all enrolled subjects had in situ cancers) to maintain adequate cell sizes within each site of cancer. We also controlled for year of diagnosis (1996 or 1997).

STATISTICAL ANALYSIS

Kaplan–Meier survival curves were estimated by each Medicaid enrollment category. This nonparametric depiction of survival does not account for lead-time bias and, thus, effectively demonstrates why past research may have shown survival differences to be so disparate between Medicaid and non-Medicaid subjects. In the current analysis, the effects of characteristics such as more severe cancer sites and stages that may be inherent to a population that is associated with disability and low-income are observed. This analysis shows the survival for 3 samples with fairly uniform demographic characteristics (e.g., everyone age < 65 years, white, or African American), but with different health insurance coverage and SES.

To account for survival differences that may be due to a disproportionate distribution of severe disease (e.g., lung carcinoma or late-stage cancers regardless of site), multivariable analysis was conducted using Cox regression with stratification by cancer site and gender for a total of five strata (female breast, colorectal-female, colorectal-male, lung-female, and lung-male). We assessed the significance of Medicaid enrollment category, age at diagnosis, race, and year of diagnosis on survival. We then estimated a multivariable model that incorporated all of these independent variables and we assessed the significance of all pairwise interactions between these variables and interactions with the strata, retaining interaction effects that were significant at 1%. A stepwise regression model was also applied to the data with a 5% entry and a 1% stay criterion, which led to exactly the same model as found in the stratified analysis. The final model included Medicaid enrollment status, cancer stage, age, race, and interactions between enrollment status and cancer stage and enrollment status and stratum. Parameter estimates from this final model were used to estimate hazards ratios (HR) and 95% confidence intervals (95% CI) across Medicaid enrollment categories and patient characteristics. (Interim analyses are not shown, but are available from the authors upon request.)

RESULTS

Table 1 contains descriptive statistics for the sample by Medicaid enrollment status. Age was relatively uniform across the three samples. A statistically significantly higher percentage of African Americans was found in the 2 Medicaid samples (27% and 39%, respectively) relative to the non-Medicaid sample (11%). A higher percentage of men (37%) was enrolled in Medicaid after diagnosis with cancer relative to the percentage of men in the Medicaid-enrolled and non-Medicaid samples (29% and 26%, respectively). Men tend to qualify for Medicaid through a disability, so it is not surprising that more men were in the late-enrolled sample. Within the non-Medicaid and Medicaid-enrolled samples, there was a higher percentage of female breast carcinoma (56% and 43%, respectively). In contrast, lung carcinoma was the predominant cancer type (nearly 46%) in the sample that enrolled in Medicaid after a cancer diagnosis and accounted for 41% of the cancer cases in the Medicaid-enrolled sample, indicating a severe case mix for the Medicaid enrollees.

Table 1. Descriptive Characteristics of Subjects with Cancer, Age < 65 Years, 1996–1997 (n = 13,740)
VariablesNon-Medicaid (n = 11,768) (%)Enrolled in Medicaid (n = 1137) (%)Enrolled after diagnosis (n = 835) (%)
  • SD: standard deviation.

  • a

    The median survival was not estimated in the non-Medicaid group because their survival experience exceeded 50% in the observed period.

Demographics   
 Mean age (yrs) (SD)51.3 (8.6)52.2 (8.7)52.9 (8.1)
 African American1322 (11.23)445 (39.14)223 (26.71)
 Males2998 (25.48)329 (28.94)307 (36.77)
Cancer site   
 Female breast6621 (56.26)487 (42.83)291 (34.85)
 Colon/rectum2299 (19.54)178 (15.66)162 (19.40)
 Lung2848 (24.20)472 (41.51)382 (45.75)
Stage of disease   
 Early6399 (54.38)456 (40.11)428 (26.71)
 Late5369 (48.94)681 (59.89)612 (73.29)
Diagnosed in 19965759 (48.94)548 (48.20)382 (51.26)
Median survival (mos) (range)N/Aa 38 (32–44) 19 (17–22)

The distribution of cancer stage also varies by Medicaid enrollment. Approximately 54% and 40% of the subjects in the non-Medicaid sample and the Medicaid-enrolled sample, respectively, had early-stage cancers (in situ and local cancers combined). In contrast, only 27% of the subjects who enrolled in Medicaid after diagnosis had early-stage cancers. This is partly due to the preponderance of lung carcinoma, which is almost always diagnosed at late stages, in the late-enrolled sample. Figure 1 summarizes the dramatic differences in the percentage of subjects with early and late-stage cancers by Medicaid enrollment status.

Figure 1.

Proportion of subjects with early and late-stage breast, colorectal, and lung carcinomas at diagnosis by Medicaid enrollment category, 1996–1997. An analysis predicting stage at diagnosis for female breast, colorectal, cervical, and lung carcinoma appeared in Bradley et al.12 Cervical carcinoma was not included in this figure. Gray bar: in situ/local disease; white bar: regional/distant disease.

Figure 2 presents the Kaplan–Meier survival estimates by non-Medicaid and Medicaid enrollment, separating subjects enrolled after diagnosis from subjects enrolled before diagnosis, for all subjects in the study. The median survival period for subjects who enrolled in Medicaid after diagnosis was 19 months (95% CI, 17–22 months), whereas the median survival period was twice as long or 38 months for subjects who enrolled in Medicaid before diagnosis (95% CI, 32–44 months). Median survival was not estimated in the non-Medicaid group because the survival experience of these subjects exceeded 50% in the observed period. The log-rank test confirmed that the difference in survival among the three groups was statistically significant (chi-square = 799.09, P < 0.0001). The Kaplan–Meier survival analysis shows an advantage for subjects who are enrolled in Medicaid before diagnosis. The prevalence of lung carcinoma—a disease associated with late-stage diagnosis and poor prognosis—may partly explain the survival difference between the Medicaid samples and non-Medicaid insured sample, but does not explain the survival advantage of the Medicaid-enrolled sample compared with the late-enrolled sample. Recall that the percentage of patients with lung carcinoma was nearly equivalent in the two Medicaid samples and that breast carcinoma, which is associated with longer survival, is less prevalent in the Medicaid samples relative to the non-Medicaid sample. If the survival advantages observed are attributable to a higher percentage of early-stage cancers or lead-time bias, then survival differences will not be apparent in a multivariable analysis that controls for cancer stage.

Figure 2.

Kaplan–Meier survival curves for breast, colorectal, and lung carcinoma by Medicaid enrollment group, 1996–2003.

Next, we study such a model that stratifies the sample by cancer site and stage and subject gender to determine if survival differences persist between the Medicaid and non-Medicaid samples, once differences inherent to the samples (e.g., cancer site, stage) are controlled in the analysis. Table 2 shows that the HRs for death (range, 1.22–3.10) for subjects enrolled in Medicaid were elevated relative to subjects in the non-Medicaid sample. Over all sites and stages, the likelihood of death for subjects insured by Medicaid is nearly two to three times the likelihood of death for subjects who are not insured by Medicaid. For example, women enrolled in Medicaid when diagnosed with early-stage breast carcinoma had an HR of 3.06 (95% CI, 2.49–3.77) and women enrolling in Medicaid after diagnosis with early-stage breast carcinoma had an HR of 3.10 (95% CI, 2.35–4.10) relative to non–Medicaid-enrolled women. A similar pattern is found for subjects with colorectal and lung carcinoma. Surprisingly, this pattern is also found for patients with late-stage cancers—for whom the prognosis would be expected to be equally poor regardless of insurance coverage—across all sites studied. Within cancer site and stage, which controls for differences in lead-time bias, the risk of death was not statistically significantly different between the 2 groups of Medicaid-enrolled subjects (P > 0.05).

Table 2. Hazards Ratios and 95% Confidence Intervals Predicting Death By Cancer Site, Stage of Disease, and Subject Gender, 1996–2003
StrataHazards ratioa95% CI
  • 95% CI: 95% confidence interval.

  • a

    Hazards ratio estimates were controlled for subject age and race.

Breast carcinoma  
 Enrolled vs. non-Medicaid, early stage3.062.49–3.77
 Late enrolled vs. non-Medicaid, early stage3.102.35–4.10
 Enrolled vs. non-Medicaid, late stage2.031.68–2.45
 Late enrolled vs. non-Medicaid, late stage2.431.94–3.04
Colorectal carcinoma  
 Enrolled vs. non-Medicaid, female, early stage2.281.61–3.22
 Late enrolled vs. non-Medicaid, female, early stage2.781.87–4.14
 Enrolled vs. non-Medicaid, female, late stage1.511.10–2.07
 Late enrolled vs. non-Medicaid, female, late stage2.181.59–2.98
 Enrolled vs. non-Medicaid, male, early stage2.381.72–3.30
 Late enrolled vs. non-Medicaid, male, early stage1.941.34–2.82
 Enrolled vs. non-Medicaid, male, late stage1.581.18–2.11
 Late enrolled vs. non-Medicaid, male, late stage1.521.15–2.01
Lung carcinoma  
 Enrolled vs. non-Medicaid, female, early stage2.051.65–2.56
 Late enrolled vs. non-Medicaid, female, early stage1.641.19–2.26
 Enrolled vs. non-Medicaid, female, late stage1.361.16–1.60
 Late enrolled vs. non-Medicaid, female, late stage1.281.08–1.52
 Enrolled vs. non-Medicaid, male, early stage1.851.47–2.31
 Late enrolled vs. non-Medicaid, male, early stage1.771.29–2.42
 Enrolled vs. non-Medicaid, male, late stage1.221.05–1.42
 Late enrolled vs. non-Medicaid, male, late stage1.381.19–1.61

DISCUSSION

The current study compared the survival of subjects enrolled in Medicaid, either before or after the month of a breast, colorectal, or lung carcinoma diagnosis, with the survival experiences of a non-Medicaid cancer sample. We hypothesized that survival differences between Medicaid and non-Medicaid subjects may be due to the Medicaid enrollment policy that requires subjects to meet disability requirements, which would place them closer to the end of life. We found that subjects who enrolled in Medicaid after a cancer diagnosis were diagnosed at later cancer stages, had a higher proportion of lung carcinoma, and most likely, were not eligible for Medicaid until they became disabled. Patients enrolled in Medicaid before diagnosis survived twice as long as patients who enrolled in Medicaid after diagnosis, but had inferior survival relative to non-Medicaid subjects. Once cancer site and stage were controlled in the analysis, survival differences persisted between Medicaid and non–Medicaid-insured subjects, but not between Medicaid-enrolled and late-enrolled subjects.

An epidemiologic interpretation of the findings is that the observed differences in survival are confounded by lead-time bias because a higher proportion of the non-Medicaid and enrolled Medicaid samples had early-stage cancer. However, a lead-time bias interpretation alone is misleading and overlooks important policy implications. Because the survival plateaus between the non-Medicaid and Medicaid samples do not converge, lead-time bias is unlikely to fully account for observed survival differences between the non-Medicaid and the two Medicaid samples. Furthermore, once the sample was stratified by site and stage of cancer, Medicaid enrollees had a two to three times greater risk of death relative to the non-Medicaid sample. Medicaid recipients may have more comorbid conditions than the non-Medicaid population and thus, have lower survival possibilities. In addition, Medicaid recipients may not be compliant to the treatment regimens prescribed. Finally, Medicaid recipients may not receive the same level of care as the non-Medicaid population. Explanations such as these warrant further investigation, although it is important to recall that subjects were enrolled in Medicaid an average of 9 months before diagnosis. Therefore, any survival improvement in this sample is laudable.

The policy interpretation, which relies more on the nonparametric analysis, of the findings is quite significant. From a policy perspective, cancer survival in the Medicaid population cannot be improved as long as 40% of the population enrolls in Medicaid after diagnosis with late-stage disease. Nevertheless, once the effects of early-stage diagnosis (i.e., lead-time bias) are reduced in the survival analysis, subjects enrolling in Medicaid after diagnosis have comparable outcomes to the existing Medicaid population. Policies aimed at improving how providers treat Medicaid-insured patients must bear in mind that a substantial portion of the population presents at a stage when treatment offers only palliative benefits. Treatment, even if it is state of the art, provided to patients at such late stages in life is not likely to substantially improve survival. Studies that do not account for Medicaid enrollment relative to the diagnosis of cancer may overly attribute poor survival to the care provided once they are enrolled in Medicaid when, in fact, this care may simply be futile.

Four important limitations to our research are noted. First, it is uncertain if our findings can be generalized to Medicaid recipients in other states. Nevertheless, Medicaid enrollment policies are somewhat similar across the country. Therefore, findings with respect to cancer survival and Medicaid enrollment categories are likely robust—regardless of the state from which the population was drawn. Second, the dataset did not contain information on the insurance status of non-Medicaid enrollees. Therefore, we cannot compare Medicaid enrollees with other insured or uninsured groups of individuals. The presence of uninsured individuals in the non-Medicaid group may diminish observed differences between Medicaid and non-Medicaid enrollees. Although we identified Medicaid recipients who were not enrolled in Medicaid before the month of diagnosis, we cannot assume that these individuals were previously uninsured; they simply were not insured by Medicaid. However, other studies have shown that uninsured persons with cancer are often diagnosed at late stages and then enrolled in Medicaid.16 Therefore, it is likely that some portion of the sample was uninsured previous to Medicaid enrollment. Third, ideally, we would have stratified the sample by more refined categories of stage. The gross categories we use may partly explain the survival differences observed. Fourth, one can conjecture that subjects enrolling in Medicaid after their diagnosis may have qualified for Medicaid before the onset of cancer, but chose not to enroll. Perhaps, this scenario applies to the subjects with early-stage cancers who enrolled in Medicaid after diagnosis because early-stage cancers alone are not typically disabling. These individuals may have met another qualifying condition (e.g., caring for a young child) before the onset of cancer.

The current study partially demonstrates why an insured, low-income population has disparate cancer survival. Subjects enrolled in Medicaid after cancer diagnoses are vulnerable to poor survival because they have more severe disease. These subjects comprise a substantial portion (nearly 42%) of the Medicaid-insured cancer population. Furthermore, those who were insured by Medicaid before diagnosis had only been insured for a period of 9 months. One strategy to improve cancer survival for otherwise, uninsured low-income persons might be an early detection and treatment program for cancer (as is currently done with the Breast and Cervical Cancer Control Program [BCCCP]). In 2001, women whose cancer was detected under the auspices of BCCCP were enrolled in the Michigan Medicaid program. Research is needed to evaluate if programs such as BCCCP improve survival for its recipients.

With the exception of women enrolling in Medicaid through the BCCCP, the Medicaid enrollment policy with respect to cancer appears to be inconsistent with a national agenda to reduce disparities in cancer outcomes because it limits access to screening and follow-up care to otherwise uninsured individuals who are not disabled or attached to another qualifying program (e.g., Aid to Families with Dependent Children). Such a policy disproportionately affects African Americans because they are more likely to be low-income and uninsured relative to whites and disproportionately affects men who are less likely to qualify for Medicaid under other programs. Thus, the enrollment policy may exacerbate known disparities in health outcomes.

The expansion of cancer detection programs and health care coverage to low-income, uninsured individuals may be less expensive, although this is uncertain, rather than enrolling these individuals in Medicaid once their prognosis is so poor. Medicaid claim files suggest that ≥ 90% of those enrolling in Medicaid after a cancer diagnosis received cancer-directed treatment. One can argue that patients with short life spans are less costly to treat than patients with early-stage disease. This argument implicitly and, perhaps, incorrectly assumes that low-income patients with cancer do not support family members who may also become Medicaid dependent once individuals with cancer are disabled. Should reductions occur in either covered services or the number of covered lives, low-income individuals will not receive preventive services17, 18 and may eventually require Medicaid to provide expensive treatments that are not likely to prolong survival and may further strain an already overburdened system of care.

Ancillary