African American race and uninsurance are associated with undertreatment and poor survival in solid tumor cancers. This relationship has not been examined in acute myeloid leukemia (AML) where absence of treatment or treatment delays can result in death within weeks or months. Induction followed by consolidation treatment, in contrast, has a high probability for remission or cure. We examined the relationship between race and health insurance and inpatient chemotherapy and survival in AML patients between the ages of 21 and 64 years. We also examined inpatient costs associated with inpatient treatment.
We used population-based data from the Virginia Cancer Registry and the Virginia Health Information discharge data for patients diagnosed with AML between 1999 and 2006 (n = 523). Adjusted logistic regression was used to measure the relationship between the independent variables and chemotherapy. We used the Cox proportional hazards method to estimate survival.
Uninsured patients were more likely to be untreated than their privately insured counterparts (odds ratio, 4.40; 95% confidence interval, 1.85-10.49) and had a higher likelihood of death (hazard ratio, 1.29; 95% confidence interval, 1.02-1.84). Once treatment was adjusted in the survival analyses, differences between insurance groups were not statistically significant. The median 1-year cost of inpatient care following diagnosis for patients who received chemotherapy exceeded $100,000.
Acute myeloid leukemia (AML) is a hematological malignancy that accounts for approximately 13,300 of all new cancer cases per year.1 The incidence of AML increases with age, but often the elderly are unable to tolerate high-dose cytotoxic regimens of chemotherapy.2 Poor performance status and comorbid conditions make these patients poor candidates for treatment,3 and even when tolerated, chemotherapy may not benefit patients ≥70 years of age.4 Younger patients, in contrast, are usually able to tolerate intensive chemotherapeutic treatment regimens. Recommended treatment includes both inpatient induction and consolidation therapy.3 These treatments are given in an inpatient setting and can require long stays in the hospital. However, if successful, 70%-80% of patients <60 years of age enter complete remission.5 With induction and consolidation therapy, 40%-45% of patients will be cured.5 In contrast to many solid tumor cancers (with the exception of those in late stage), without treatment, AML usually results in death within a few months after diagnosis.5
Published studies have not examined the influence of health insurance source and race on AML treatment and survival in adults <65 years of age. Across many solid tumor cancer sites, African Americans are more likely to be diagnosed at later stages, have poorer survival,6 and are less likely to receive adequate treatment.7-11 Likewise, lack of health insurance is associated with lack of cancer-directed treatment,12, 13 although this relationship has not been demonstrated in patients with AML, in whom disease progression is certain and in the absence of treatment, prognosis is almost always death. Limited population-based data are available on younger and uninsured patients because they are not eligible for Medicare and are thus not included in large medical claims databases such as SEER-Medicare.14 Using a statewide, population-based dataset, we examined the probability that young adult (ages 21-64 years) AML patients are treated with chemotherapy. We contrasted the experience of privately insured patients with that of uninsured and publicly insured patients (defined as Medicare or Medicaid), controlling for socioeconomic status along with other demographic characteristics. We also examined the influence of race on the probability that a patient receives chemotherapy. We then compared the probability of survival for patients across different categories of insurance, including being uninsured. Finally, we estimated the costs of inpatient care 1 year following AML diagnosis. Costs were examined to inform hypotheses about the relationship between the cost of care and the probability of treatment, particularly for uninsured patients. These hypotheses could not be tested in this study, but are left for future research.
MATERIALS AND METHODS
Patient data came from 2 statewide sources: the Virginia Cancer Registry (VCR) and the Virginia Health Information (VHI) discharge data. The VCR, which is population-based and North American Association of Central Cancer Registries accredited, was the source for the AML sample. The VCR contained data on patient demographic characteristics, cancer site, diagnosis date, first course treatment, primary health insurer, and patient address, including geocoded coordinates that could be linked to census tracts. The study was approved by the institutional review boards at Virginia Commonwealth University and the Virginia Department of Health.
Inpatient treatment information was extracted from the VHI discharge database, which contained discharge abstracts on all Virginia hospital admissions that exceeded 23 hours. Discharge abstracts included patient information, International Classification of Diseases version 9 (ICD-9) diagnosis and procedure codes, payer information, dates of admission and discharge, and charge information. Information was available on all types of discharges—not just those related to the treatment of cancer. However, because these patients were young and most likely healthy, few had inpatient admissions prior to diagnosis.
We linked the VCR and VHI data using deterministic and probabilistic matching techniques. Both datasets contained Social Security Number (SSN), date of birth, sex, and ZIP code. To evaluate and assess the possible matches returned from these efforts, we compared the level of agreement for each data element involved in matching. For example, for each possible matched VHI record, we determined the number of SSN digits that matched (from 0 to 9) the difference in month, day, and year components of the date of birth and whether there was a match on sex and ZIP code to those elements on the VCR record. Among the matched records for AML patients, 92% matched exactly on SSN, date of birth, and sex.
The number of newly diagnosed AML patients aged 21-64 years, diagnosed between January 1, 1999, and December 31, 2006, and who had no prior history of cancer was 539. To assign health insurance to a patient, we used the payer listed in the VHI hospital discharge data for inpatient chemotherapy treatment; for patients who did not receive inpatient chemotherapy, we used the payer listed in the VCR. Payers were categorized into the following groups: Medicaid, Medicare, private or military, other government (eg, local county plan), uninsured, and unknown. We removed patients with unknown, Veterans Administration or other government insurance from the sample (n = 16 [3.1%]) for a total sample size of 523. Race was categorized as white, African American, or Other using data from the cancer registry.
The outcomes of interest were absence of inpatient chemotherapy, survival, and inpatient costs of treatment 1 year after diagnosis. Inpatient chemotherapy was primarily determined from VHI claims (n = 416 [86%]). An additional 69 (14%) patients were identified from the VCR, but did not have a corresponding claim in the VHI database. We combined information from both sources to minimize possible undercounting of patients who may have gone out of state for treatment and thus would not have a claim in the VHI database. Nonetheless, we believe these patients were treated in an inpatient setting, because it is unlikely that many patients in this younger age group would receive chemotherapy in an outpatient setting. Low-dose outpatient chemotherapy is almost exclusively reserved for older patients and those whose acute leukemia followed myelodysplasia or is secondary to previous chemotherapy.15, 16
We verified patients' dates of death with the Social Security Death Index through September 30, 2010. Therefore, we had nearly 4 years of survival data, at a minimum, on all patients in the sample. Death was measured as all-cause mortality.
Inpatient costs were estimated using inpatient charges reported in the VHI discharge data. All patients were included in this analysis, including those that died within 1 year of diagnosis (n = 146). To calculate the cost/charge ratio for each hospital, we divided each hospital's total inpatient and outpatient operating expenses by its inpatient and outpatient charges. Separate hospital-specific cost/charge ratios were estimated for each year during the study period. We converted charges to costs by multiplying charges by the hospital-specific cost/charge ratio for the year in which the inpatient visit occurred. Costs were then adjusted for inflation using the Medical Consumer Price Index for the Washington–Baltimore, Maryland, District of Columbia, Virginia, West Virginia17 area over the period 1999 to 2006 to adjust the costs to the year 2006.
We controlled for patient sex and age in all models. Age at the time of diagnosis was entered into the model as a continuous variable. In the analysis of survival time from diagnosis to death, we added a dichotomous variable indicating whether the patient had received inpatient chemotherapy.
Other researchers have noted that socioeconomic characteristics can confound the correlation between uninsurance and public insurance and poor cancer outcomes.6 In addition, socioeconomic status was shown to be associated with poor AML survival in a cohort of patients in Sweden18 and in the United Kingdom,19 where health insurance is nearly universal. Therefore, we included ecological measures of socioeconomic status in our analysis. Using a method developed by Diez Roux et al,20 we constructed a summary measure of socioeconomic status for each census tract code in Virginia using data on income, education, and occupation from the 2000 US Census and linked this information to patients' census tract code of residence. Six census variables were used, 3 measures of household wealth and income (median household income; median value of housing units; proportion of households with interest, dividend, or rental income), 2 education variables (proportion of adult residents completing high school; proportion of adult residents completing college), and 1 variable regarding occupational status (proportion of employed residents with management, professional, and related occupations).
The mean and standard deviations were calculated for each of the 6 variables in all Virginia census tracts.21 A z-score was constructed for each census tract code by subtracting the mean of all Virginia census tracts and dividing by the standard deviation. The summary measure was the summation of the 6 z-scores and ranged from −12.2 to 17.5. To facilitate interpretation, we standardized summary scores by converting them into a 0-100 scale (0, greatest socioeconomic disadvantage; 100, greatest socioeconomic advantage).
We described the characteristics of all AML patients by health insurance source and used chi-square and Fisher exact tests to determine statistical differences between the samples. We used t tests to determine the statistical significance between means in continuous variables. Kruskal-Wallis tests were used to determine statistical significance between medians in continuous variables. Adjusted logistic regression was used to measure the relationship between the independent variables and the absence of inpatient chemotherapy. We estimated 2 models—one without insurance status and another with insurance status—to demonstrate how insurance coverage alters the effect of race on treatment. We reported odds ratios (ORs) and 95% confidence intervals (CIs) and P values. Because the absence of chemotherapy has a low prevalence in our sample, the OR closely approximated the relative risks (results not shown). P values were derived from likelihood ratio tests and are 2-sided. We estimated the median length of stay and inpatient costs descriptively.
For survival analyses involving adjustments for confounding factors, we used the Cox proportional hazards method. We estimated proportional hazards models with and without chemotherapy. The proportional hazard assumption was tested by including time-dependent covariates (created as the interactions of the main effects with the logarithmic transformation of time) in the Cox regression model. This method is equivalent to testing for a non-zero slope in a generalized linear regression of the scaled Schoenfeld residuals on functions of time. All analyses were conducted using SAS/STAT version 9.2 for Windows.
Table 1 reports the sample characteristics by health insurance source and for the overall sample. There were fewer privately insured African Americans (13%) relative to other insurance groups. Slightly more females were insured by Medicare or Medicaid (56%) relative to other groups. Patient age was comparable across insurance groups, although uninsured patients were slightly younger, but not statistically significantly so, relative to other patients. The average socioeconomic status score for census tracts where Medicare- and Medicaid-insured patients resided was statistically significantly lower relative to the score for privately insured patients (37.61 vs 44.55). The average socioeconomic status score for census tracts where uninsured patients resided (41.05) was also lower than privately insured patients and was statistically significant (P < .05).
Nearly all privately insured patients received inpatient chemotherapy (94%). In contrast, 19% of uninsured patients did not receive inpatient chemotherapy and 10% of publicly insured patients did not receive inpatient chemotherapy. Patients insured by Medicaid or Medicare had fewer survival months than uninsured and privately insured patients (11.10 months, 16.37 months, 21.23 months, respectively). We noted 2 extraordinary outliers in the uninsured group. When we excluded these patients from the sample, the upper CI for uninsured patients decreased from 57 months to 37 months.
Chemotherapy and Survival
Table 2 reports estimates predicting the likelihood of chemotherapy and the likelihood of death. In estimations without insurance source, African Americans were more likely to forgo inpatient chemotherapy, but the OR was only marginally statistically significant (OR, 2.33; 95% CI, 0.97-5.57; P = .06). In the full model that includes insurance status, uninsured patients were 4 times more likely to forgo inpatient chemotherapy as privately insured patients (OR, 4.40; 95% CI, 1.85-10.49).
Table 2. Absence of Inpatient Chemotherapy and Likelihood of Death (n=506)a
Adjusted Logistic Regression, Insurance not Included, OR (95% CI)
CI indicates confidence interval; NA, not available; OR, odds ratio; RR, relative risk; SES, socioeconomic status.
Census tract was missing for 17 observations.
Asian, American Indian, other
Census tract SES score
Insurance source was a statistically significant predictor of the likelihood of death in models that did not include the receipt of chemotherapy. Uninsured patients had a higher hazard ratio (HR) than privately insured patients (HR, 1.29; 95% CI, 1.02-1.84). Publicly insured patients were also more likely to die during the study period relative to privately insured patients (HR, 1.39; 95% CI, 1.02-1.88). Older age was associated with a slightly higher likelihood of death (HR, 1.03; 95% CI, 1.02-1.04). African Americans had a higher likelihood of death relative to white patients (HR, 1.43; 95% CI, 1.05-1.94).
In the model that included receipt of chemotherapy, patients who received chemotherapy had a much lower likelihood of death (HR, 0.36; 95% CI, 0.24-0.54), and the HRs for health insurance and race were no longer statistically significant.
Cost of Care
If patients are faced with large out-of-pocket expenses, they may be less inclined to seek treatment or more inclined to delay treatment, particularly if the treatment involves long and costly inpatient stays. In our sample, the greatest median costs for inpatient treatment occurred during the first 2 quarters following initial diagnosis (approximately $80,000 and $20,000 for all patients in the first and second quarters, respectively). For patients that did not survive the year, median costs spiked in the third quarter to approximately $25,000. During the final quarter of the year after diagnosis, inpatient costs decreased substantially and approach zero (Figure 1).
In an analysis stratified by insurance type, the median inpatient cost of care 1 year after diagnosis was approximately $123,000 for privately insured patients and approximately $130,000 for publicly insured patients (Table 3). Uninsured patients, in contrast, had inpatient median costs that were considerably lower (approximately $100,000). These patients had shorter lengths of stay in the hospital (3 and 10 days fewer relative to privately and publicly insured patients, respectively). They also had fewer readmissions within 30 days of discharge and fewer retreatments defined as a readmission for chemotherapy within 120 days after the initial chemotherapy treatment discharge from the hospital.
Table 3. Inpatient Costs 1 Year After Diagnosis by Patient Insurance Status (n=407)
Retreatment within 120 days after inpatient chemotherapy discharge, n (%)
The prognosis of patients with untreated AML is certain: death usually occurs within a few months. If left untreated, AML typically leads to death within weeks to months of its clinical presentation.22 With treatment, particularly in younger adult patients, the prognosis includes a high likelihood of remission and a chance for cure. Uninsured patients were less likely to receive chemotherapy than their privately insured counterparts. Racial differences were present in survival analyses that excluded chemotherapy. In these analyses, African Americans had a greater likelihood of death. Once chemotherapy was added to the estimation, both racial and insurance differences were no longer statistically significant. These findings point to the critical role of health insurance in AML, a life-threatening disease that is expensive to treat. In a separate analysis, approximately one-third of uninsured AML patients died within 2 months after diagnosis (results not shown), a finding that underscores the importance of health insurance in preventing death in patients with a potentially curable disease. Publicly insured patients—a group that was demographically similar, but lived in lower income areas—were not statistically different from privately insured patients in their probability of receiving chemotherapy, although they experienced a slightly higher and statistically significant probability of death (HR, 1.03; P < .05).
The reasons why uninsured patients are less likely to receive chemotherapy are many. We hypothesize that 1 reason may, in part, be due to cost and perhaps delay and reluctance to receive medical consultation for AML symptoms. If patients bear the full cost of inpatient treatment, they will pay $100,000 or more; these costs occurred even for those patients who died within a short period. If hospitals and physicians economize on length of stay and subsequent treatment such as consolidation therapy for the uninsured, costs are still exorbitant. Assuming that patients do not bear the full cost of care, the treating hospital must agree to absorb costs of care or in the case of publicly insured patients, accept lower reimbursement rates. The remaining out of pocket payments for patients may still be beyond what patients and their families can pay. Other reasons related to the absence of chemotherapy may be due to patient health status and comorbid conditions. Patients who present late in the disease course or with conditions such as infection or severe bleeding may not be able to tolerate chemotherapy, or if chemotherapy is initiated, they may not survive treatment. Uninsured patients may also have neglected conditions such as diabetes or coronary heart disease that are counterindications for intensive treatment.3 However, the fact that Medicare and Medicaid patients were more likely to receive chemotherapy than uninsured patients casts doubt on arguments regarding differences in health status.
This study has several limitations. First, the data were confined to a single state and may not be generalized to other regions of the country. Nonetheless, the data were population-based, a considerable strength, and there is no apparent reason why chemotherapy and survival rates would differ in Virginia during the 7-year study period relative to rates observed in other states. Second, the data were specific to inpatient treatment. Information on supportive care, outpatient treatment, and physician services was not available. Differences in outpatient treatment may have contributed to or lessened the differences observed in this study. Third, it is possible that some patients were incorrectly misclassified as not having treatment when they in fact received treatment out of state. The effect of misclassification would bias survival analyses toward the null hypothesis, lending greater confidence to the statistically significant results we observed. We suspect that inpatient chemotherapy misclassification is low in our sample, because the patients who are least likely to have chemotherapy—namely the uninsured and publicly insured—are also the least likely to travel out of state for treatment. Lastly, information on patient health status, functional status, cytogenetics, and comorbid conditions from outpatient data were not available. Only a few patients had inpatient admissions before cancer diagnosis, which suggests that the severity of comorbid conditions, if present, may have been low. Each of these factors could be correlated with insurance status and the outcomes of interest.
In spite of these limitations, the study found statistically significant associations between insurance status and receipt of inpatient treatment and survival. The study also filled an important gap in the literature by providing information relevant to adult AML patients <65 years of age. It confirmed that findings from other published studies regarding uninsurance and treatment and survival in other cancer sites also apply to patients with AML, but the consequences associated with the absence of treatment in AML are much more immediate and dire.
We hypothesize that the relationship between insurance and lack of treatment may be related to the cost of treatment. In a study published in 2006,23 researchers found that the mean inpatient charges for the initial hospitalization were $113,118 for patients who received chemotherapy and $43,999 for patients who did not receive chemotherapy. The estimated costs in our study were similar. Although we could not directly test hypotheses related to high cost of care and absence of treatment in the uninsured in this study, we nonetheless concluded that as costs continue to rise, disparities between insurance groups may widen.
This study's findings suggest that there are opportunities to develop policies that can influence improvements in AML survival among adult patients <65 years of age. Although its final form is yet to be defined, implementation of the Affordable Care Act is likely to result in fewer uninsured patients, and these patients may have easier access to expensive inpatient treatments. However, we recognize that these patients may also have conditions that are counterindications for treatment, and that these conditions may be related to uninsurance. Further research (eg, a detailed review of medical records) is needed to understand the complex relationship between health insurance and the receipt of inpatient chemotherapy and survival for AML patients <65 years of age. Unlike findings in other studies of solid tumor patients and their insurance source, this study points to the urgency for health insurance that affords access to care. Without treatment, the outcome of AML is surely death within only a few months; with treatment, the chance for long-term remission or even cure exists. Few diseases are as time-sensitive to treatment as AML, and until more affordable care is available, many uninsured AML patients may not receive the care they need.
Supported by American Cancer Society grant, RSGI-08-301-01, An examination of uninsured and insured cancer patients in Virginia, Cathy J. Bradley, Principal Investigator.