More than 1.3 million new cancer cases are expected in 2003,

Transcription

1 Insurance & Cancer Health Insurance And Spending Among Cancer Patients Nonelderly cancer patients without insurance are at risk for receiving inadequate cancer care, especially if they are Hispanic, this study finds. by Kenneth E. Thorpe and David Howard ABSTRACT: Over 1.3 million new cancer cases are diagnosed each year. While most cancer patients are older and covered by Medicare, our analysis indicates that 10 percent of cancer patients under age sixty-five are uninsured and that 20 percent of Hispanic cancer patients under age sixty-five are uninsured. We find substantial differences in cancer spending by insurance status; uninsured patients under age sixty-five spent 57 percent as much over a six-month period as privately insured patients spent for their cancer care. We present evidence to show that spending differences are due in part or completely to differences in use, which suggests that raising coverage rates will improve cancer treatment. More than 1.3 million new cancer cases are expected in 2003, and nine million Americans with a history of cancer were alive in Cancer is the second leading cause of death (556,000 expected deaths in 2003); it accounts for nearly one of every four deaths in the United States, second only to heart disease. 1 A growing body of outcomes research has focused on policy interventions to prevent cancer, increase cancer screening, and encourage early medical interventions. Moreover, the data also reveal racial and ethic differences in cancer incidence and cancer mortality. 2 Five-year survival rates after cancer diagnosis are lower for blacks than for whites. 3 Recent research has explored the role that cancer biology, treatment, socioeconomic status, and stage of diagnosis play in explaining these racial variations. Studies have found that survival rates vary little for black and white cancer patients receiving similar medical treatment and that survival differences are reduced after insurance type and income are controlled for. 4 Thus, genetic differences appear to account for a small portion of the lower cancer mortality rates among whites. These findings have refocused attention toward the role of health insurance and of race and ethnicity in cancer screening, treatment, and outcome. A substantial body of research has focused on socioeconomic status and cancer Ken Thorpe is the Robert W. Woodruff Professor and chair of the Department of Health Policy and Management, Rollins School of Public Health, Emory University, in Atlanta. David Howard is an assistant professor in that department. HEALTH AFFAIRS ~ Web Exclusive W Project HOPE The People-to-People Health Foundation, Inc.

2 DataWatch survival, but little research has focused on the health insurance status and the use of health care services of cancer patients who are not eligible for Medicare. The few studies that have been conducted on the subject find that uninsured patients are less likely to receive guideline therapy and have worse outcomes. 5 Less is known about the relationships among insurance, race and ethnicity, and cancer treatment. Research to date has focused largely on the importance of health insurance in increasing cancer screening and access to health care. 6 However, little is known about the health insurance coverage of cancer patients and health care spending among patients with and without health insurance. This paper addresses several questions concerning access to and use of health care among insured and uninsured cancer patients: (1) What is the distribution of health insurance among cancer patients under age sixty-five? How many are uninsured, and how many have health insurance? (2) Are Hispanic and African American cancer patients more likely to be uninsured than are other ethnic or racial groups? (3) Do differences in health insurance coverage result in different levels of health care spending among cancer patients? If they exist, are they traced to varying use of inpatient hospital services, outpatient services, physician services, outof-pocket spending, or all of the above? Data And Methods Data for our analysis are derived from the household component of the Medical Expenditure Panel Survey (MEPS) for MEPS is a nationally representative sample of the civilian, noninstitutionalized population, with oversampling of minorities. It collects information on insurance status and use of and spending on health care services (derived from a reference person in the household) on all members of surveyed households during five interviews over a two-year period. Summarizing response rates is somewhat difficult, because there are four panels, each drawn from respondents to the National Health Interview Survey (NHIS), and multiple rounds within each panel. Taking account for the 1995 NHIS response rate, 83 percent of MEPS-eligible households provided data for the first round of the 1996 MEPS. 7 The corresponding rate for the second panel, which began in 1997, was 78 percent. If uninsured people are less likely to participate, then we may underestimate the number of uninsured cancer patients. From the MEPS data, we extracted records for people with at least one health care event (an inpatient or outpatient hospital visit or a visit to a physician s office) with a condition code for cancer (numbers 11 43). Condition codes are groups of similar International Classification of Diseases, Ninth Revision (ICD-9) codes and are used in MEPS to make it easier to identify patients with certain diseases. 8 For each person, we noted the first health care event (in the data) with a cancer condition code and then eliminated people whose first cancer event fell within six months of their scheduled exit from the MEPS survey panel. We were left with a sample of 1,383 people, 81 of whom were uninsured. We recoded the insurance W April 2003

3 Insurance & Cancer held by each person in the sample during the month of the first cancer event. Our measure of health care spending includes all expenditures incurred from the time of the person s first cancer visit in MEPS to six months after that first visit. 9 We also calculated, over these six-month periods, spending by payment source and by site of service. We did not differentiate between spending for cancer services and spending for care for other diseases or conditions. Thus, our results are not dependent on accurate coding of those events that are related to cancer care versus those that are not. Our tabulations of the MEPS data revealed that spending for events with cancer condition codes was, on average, 70 percent of the total spent in the six-month window for the sample. Spending totals do not include free care, charity care, or bad debt. Thus, differences in spending levels by insurance do not necessarily imply differences in the receipt of medical care. We address this issue by comparing the average number of visits by site of service between patients with different types of insurance. We used Poisson regression with the controls mentioned below to assess the significance of differences in use by insurance. A person s first cancer event recorded in MEPS may or may not correspond to the person s first visit for cancer treatment or diagnosis. In most cases it does, as indicated by the year the condition started in the MEPS condition files. However, this information is missing for many of the patients in the sample, so we cannot be sure. Our measure of spending should be viewed as a snapshot of medical costs incurred by cancer patients during a six-month window. For most patients, this period will include the costs of initial treatment and diagnosis; for others this will include only the costs of follow-up and posttreatment screening or the costs of end-of-life care (141 patients died during the MEPS observation period). 10 Given that MEPS only follows participants for a two-year period, using a longer window to calculate costs would force us to eliminate too many observations due to censoring. However, our results are similar when we limit the sample to people whose cancer condition started during the MEPS observation period. For each racial/ethnic group we calculated the proportion in the sample with different types of insurance. We then used probit regressions to estimate the likelihood of having each type of insurance as a function of dummy variables for age group (below age eighteen and over age fifty-five), cancer type (lung, skin, breast and cervical, prostate, not defined; other was omitted), and a dummy variable for each racial/ethnic group. All regressions were weighted and adjusted to account for the MEPS complex survey design. The coefficient on the race variable was used to assess whether a member of the group in question was significantly more or less likely to have a type of insurance compared with the rest of the sample. To estimate the impact of insurance on overall spending, we ran an ordinary least squares regression (the regression was weighted, and the standard errors were adjusted for the survey design), where the logarithm of spending was the dependent variable and dummies for age group, cancer type, race, and insurance type HEALTH AFFAIRS ~ Web Exclusive W3-191

4 DataWatch (uninsured, private insurance, Medicare, Medicaid; other was omitted) were independent variables. We created an average predicted spending estimate by multiplying the estimated coefficients by the data (to create predicted values of the logarithm of spending), transforming the predictions to constant dollars using Duan s smearing estimator, and then computing the mean across the entire sample. 11 To calculate predicted spending levels by insurance status, we repeated this procedure but set the variable indicating membership in the insurance category of interest to one for every person in the data. Thus, we calculated what average spending would be in the six-month window if every person had the same type of insurance. Bootstrapping was used to assess the significance of the difference in the average predicted spending level and the predicted spending level for a specific type of insurance. The same methods were used to assess differences in spending by site of service and in out-of-pocket spending (which consists of spending by patients and their families for medical services), except that we used a two-part model of spending to account for the nontrivial proportion of people with zero values for the relevant spending category. 12 Spending levels were inflated to constant 2001 dollars using the December all-item Consumer Price Index (CPI) for urban consumers. In all cases, the unit of analysis is the individual. Study Results Exhibit 1 shows weighted and unweighted summary statistics of the sample. Differences between uninsured and insured cancer patients largely reflect the age at which various cancers are typically diagnosed. Many breast cancer cases are diagnosed before age sixty-five, and, not surprisingly, rates of uninsurance are EXHIBIT 1 Summary Statistics On Sample Of Cancer Patients From The Medical Expenditure Panel Survey, White a Female Under age 30 Over age 54 Poor Cancer type Lung Breast/cervical Skin Prostate Uninsured Insured Weighted Unweighted Weighted Unweighted 80% % % % Number 851, ,042,870 1,302 SOURCE: Authors analysis of data from the Medical Expenditure Panel Surveys. a Includes all nonblacks and non-hispanics. W April 2003

5 Insurance & Cancer higher among breast and cervical cancer patients and female cancer patients. Conversely, prostate cancer is typically detected after age sixty-five, and only 7 percent of prostate cancer patients lack insurance. Insurance coverage. Exhibit 2 shows the distribution of insurance coverage at the first event in the MEPS data, by race. The results reveal that substantial numbers of cancer patients have no health insurance. Moreover, there are substantial differences in health insurance coverage between Hispanics and other cancer patients. Eleven percent of cancer patients under age sixty-five, and 20 percent of Hispanic cancer patients, were uninsured. The distribution of health insurance reflects, in part, the timing of a typical cancer diagnosis. The National Cancer Institute reports that 77 percent of all cancers are initially diagnosed among those older than age fifty-five. Data from the Surveillance, Epidemiology, and End Results (SEER) program indicate that nearly 60 percent of all new cancer cases are diagnosed among those age sixty-five and older. 13 These cancer facts are also reflected in the distribution of health insurance coverage. Overall, Medicare covered 56 percent of cancer patients, and 5 percent of all cancer patients were uninsured. Rates of uninsurance were higher among minorities; 8 percent of blacks and 12 percent of Hispanics (p <.05) were unin- EXHIBIT 2 Percentage Distribution Of Health Insurance Among Cancer Patients, By Race/ Ethnicity, All cancer patients Uninsured Medicare Medicaid Other public Private Number Weighted annual number Cancer patients under age 65 Uninsured Medicare Medicaid Other public Private Race/ethnicity Overall White a Black Hispanic 5% ,383 4,223,677 11% % ,129 3,784,881 10% 4 6** 8 73** 8% 49 14** ,475 14% 1 8** 25 51* 12%** * ,070 20%** ** Number Weighted annual number 652 1,886, ,641, , ,727 SOURCE: Authors analysis of data from the Medical Expenditure Panel Surveys. NOTE: Statistical significance indicators express significant difference from overall. a Includes all nonblacks and non-hispanics. *p <.10 **p <.05 HEALTH AFFAIRS ~ Web Exclusive W3-193

6 DataWatch sured, compared with 5 percent of all cancer patients. The distribution of insurance coverage among cancer patients under age sixtyfive differs greatly from the overall totals. Approximately 40 percent of new cancer cases are diagnosed among patients below age sixty-five. Among these patients, 11 percent were uninsured, and 70 percent had private health insurance. However, 20 percent of Hispanic cancer patients under age sixty-five were uninsured, compared with 11 percent of patients overall (p <.05). Only 51 percent of all Hispanic cancer patients had private health insurance, compared with 73 percent of patients overall (p <.05). To account for potential confounders to race and ethnicity, such as type of cancer, we estimated a probit regression, where the dependent variable was one if the cancer patient was uninsured. Independent variables included the type of cancer (for example, breast or lung), patient characteristics (age and sex), and race and ethnicity. We calculated odds ratios to compare the relative likelihood of lacking insurance by race and ethnicity. Results from the analysis revealed that Hispanics were twice as likely as all cancer patients and cancer patients under age sixty-five were to lack insurance. 14 Health spending among cancer patients. Uninsured cancer patients incurred a little more than half (55 percent) of the health care spending of privately insured patients (Exhibit 3). During the six-month window, cancer patients with private insurance spent an average of $6,550, compared with $3,606 among uninsured cancer patients (p <.05). The largest dollar differences in spending among privately insured and uninsured patients were for inpatient hospital services, but spending EXHIBIT 3 Predicted Health Care Spending Among Cancer Patients, By Age Group And Source Of Coverage At First Cancer Event, All ages Total expenditures Inpatient Outpatient Physician s office Out of pocket Under age 65 Total expenditures Inpatient Outpatient Physician s office Out of pocket Coverage at first cancer event Overall Medicare Medicaid Private Uninsured $6,115 3,646 1,039 1, ,252 5,212 1,342 1, $ 6,080 3, ** 1, ** $10,146 7,486 1,135 1, $5,943 3, * 820* 149 7,805 4, ** 856** 165** $6,550 4,473 1,486** 1,435** 480 8,419 5,643 1,547** 1,515** 549 $3,606** 1,155** 1, ** 1,051** 4,806** 1,454** 1, ** 1,343** SOURCE: Authors analysis of data from the Medical Expenditure Panel Surveys. NOTE: Statistical significance indicators express significant difference from overall. *p <.10 **p <.05 W April 2003

7 Insurance & Cancer by site of care for uninsured patients was lower across the board. Compared with cancer patients covered by Medicaid, uninsured patients spent much less overall (p <.05) but more on outpatient hospital services. Measured differences in spending are not due to outliers; when we eliminate the top 5 percent of spenders, large and statistically significant spending differences between insured and uninsured patients remain. 15 After controlling for insurance status, we did not detect significant differences in spending by race. When we restrict the analysis to patients below age sixty-five, we find similarly large differences in spending (Exhibit 3). Among privately insured patients, average total spending was $8,419, but only $4,806 among uninsured patients. Again, the bulk of the difference is attributable to increased hospital spending among the insured. Overall, predicted spending for the population below age sixty-five was much higher than for the total population, mirroring the raw, unweighted spending means of $5,200 versus $4,800, respectively. One possible explanation for these differences is that, consistent with guidelines, physicians provide less intensive treatments to very old cancer patients. 16 Understanding the source of these differences is an important topic for future work. Despite their lower overall spending, uninsured patients paid more out of pocket than insured patients paid, in both absolute and percentage terms $1,343 for uninsured patients below age sixty-five but only $576 in the entire sample (p <.05) and $165 for patients with Medicaid (p <.05). 17 Spending differences between insured and uninsured cancer patients could be affected by the receipt of free care or care written off as bad debt not measured in the MEPS data. However, MEPS collects data on charges in addition to data on spending. We reestimated spending models using total charges, which are recorded in MEPS separately from payments, as the dependent variable. We find that average total charges in the sample were $10,332. Charges by insurance status were as follows: Medicare, $10,015; Medicaid, $13,250; private, $11,105; and uninsured, $6,816. The difference between spending among uninsured patients and the sample mean is significant at the 10 percent level. 18 We also examined use of health care services among insured and uninsured cancer patients (Exhibit 4). Uninsured patients had fewer overall provider encounters (p <.10), inpatient admissions, and physician office visits (p <.05). They also had fewer hospital outpatient visits and fewer emergency room visits. These results are virtually identical our earlier findings using payment data. Discussion Our results present new information concerning the insurance coverage and use of services among cancer patients under age sixty-five. We find substantial disparities in health insurance coverage between Hispanic and other cancer patients and substantial differences in spending among people with and without health insurance. Overall, 5 percent of all cancer patients, and 11 percent of cancer HEALTH AFFAIRS ~ Web Exclusive W3-195

8 DataWatch EXHIBIT 4 Cancer Patients Use Of Health Services, Measured As Number Of Annual Admissions Or Visits, By Health Insurance Status, Coverage at first cancer event All health care events a Inpatient admissions Outpatient hospital visits Emergency room visits Physician office visits All Medicare Medicaid Private Uninsured * * * ** SOURCE: Authors analysis of data from the Medical Expenditure Panel Surveys. NOTE: Statistical significance indicators express significant difference from all. a Sum of inpatient admissions, visits to hospital outpatient departments and emergency rooms, and physician office visits. *p <.10 **p <.05 patients under age sixty-five, were uninsured at their initial diagnosis. Applying these data nationally, the results imply that every year about 200,000 of the four million (prevalent cases) or so patients undergoing for treatment for cancer lack insurance. Among cancer patients under age sixty-five, 20 percent of Hispanics were uninsured, compared with 11 percent of patients overall. Much of the difference appears to be linked to the higher rate of private health insurance (70 percent) among all cancer patients compared with Hispanics (51 percent). We find significant differences in spending between uninsured and insured cancer patients. We do not find differences in spending by race or ethnicity after controlling for insurance status. Our results indicate that spending differences by insurance status are attributable in part or completely to differences in use of care and are not solely a function of reimbursement levels or free care (see Exhibit 4). If they were, then we would expect that uninsured patients would have lower out-of-pocket spending in addition to lower total spending and that charges, which are issued even when care is free, would be similar. This is not what we find. Moreover, when we examine the issue of service use directly, we find that uninsured cancer patients have fewer provider encounters than those with insurance have, even after differences in age and cancer type are accounted for. Study limitations. The extent of our analysis is limited by a small sample size. Studies of patients with specific types of cancer would help to shed light on the reasons for spending differences for example, whether they are attributable to differences in the intensity of treatment or use of ancillary services and the implications of those differences for patient outcomes. The inability to identify a definitive start or end to treatment may impart some bias to the estimates. However, if we were able to consistently compare costs by treatment phase, the spending differences between the uninsured and insured would perhaps be even larger. Assuming that uninsured patients are less likely to receive follow-up care and posttreatment screen- W April 2003

9 Insurance & Cancer Extending health insurance to cancer patients without health insurance could result in earlier treatment and improved survival. ing, then the observation period for an uninsured person is more likely to include the initial treatment phase, in which costs are much higher. Another limitation is that results may be subject to response bias (recall that the MEPS response rate is about 70 percent), particularly if uninsured patients are less likely to participate in the NHIS or MEPS. Utilization data are self-reported, which could lead to undercounting of events. Finally, we cannot measure whether differences in service use between uninsured and insured patients are attributable to financial barriers, differences in treatment preferences, or other factors. Policy implications. Our results indicate that 200,000 cancer patients are uninsured. Congress has attempted to address a part of the problem through the Breast and Cervical Cancer Prevention and Treatment Act (BCCPTA) of 2000, P.L This law provides states with the option of providing medical assistance through Medicaid to women screened and diagnosed with breast or cervical cancer. To date, all but three states have implemented this provision or plan to do so; by this measure, the program could be viewed as a success. 19 However, although it is an important first step, breast and cervical cancer account for only 15 percent of newly diagnosed patients with cancer. Extending health insurance to the remaining cancer patients without health insurance could result in earlier treatment and improved survival. The authors thank the Commonwealth Fund for providing support for this work. They also appreciate the helpful comments from two anonymous reviewers. Downloaded from by Health Affairs on October 5, 2017 by HW Team HEALTH AFFAIRS ~ Web Exclusive W3-197

10 DataWatch NOTES 1. American Cancer Society, Cancer Facts and Figures, 2002 (Atlanta: ACS, 2002), V.L Shavers, L.C. Harlan, and J.L. Stevens, Racial/Ethnic Variation in Clinical Presentation, Treatment, and Survival among Breast Cancer Patients under Age 35, Cancer 97, no. 1 (2003): ACS, Cancer Facts and Figures, 2002, See, for example, P. Bach et al., Survival of Blacks and Whites after a Cancer Diagnosis, Journal of the American Medical Association 287, no. 16 (2002): ; and C.J. Bradley, C.W. Given, and C. Roberts, Race, Socioeconomic Status, and Breast Cancer, Journal of the National Cancer Institute 94, no. 7 (2002): See, for example, M.L. Brown, J. Lipscomb, and C. Snyder, The Burden of Illness of Cancer: Economic Cost and Quality of Life, Annual Review of Public Health 22 (2001): ; J.W. Eley et al., Racial Differences in Survival from Breast Cancer: Results of the National Cancer Institute Black/White Cancer Survival Study, Journal of the American Medical Association 272, no. 12 (1994): ; R.G. Roetzheim et al., Effects of Health Insurance and Race on Breast Carcinoma Treatments and Outcomes, Cancer 89, no. 11 (2000): ; D.F. Penson et al., The Association between Socioeconomic Status, Health Insurance Coverage, and Quality of Life in Men with Prostate Cancer, Journal of Clinical Epidemiology 54, no. 4 (2001): ; and J.Z. Ayanian et al., The Relation between Health Insurance Coverage and Clinical Outcomes among Women with Breast Cancer, New England Journal of Medicine 329, no. 5 (1993): J. Hsia et al., The Importance of Health Insurance as a Determinant of Cancer Screening: Evidence from the Women s Health Initiative, Preventive Medicine 31, no. 3 (2000) : S. Cohen, Sample Design of the 1997 Medical Expenditure Panel Survey Household Component, Methodology Report no. 11, Pub. no (Rockville, Md.: Agency for Healthcare Research and Quality, 2000). 8. A. Elixhauser et al., Clinical Classifications for Health Policy Research: Hospital Inpatient Statistics, 1995, HCUP-3 Research Note, Pub. no (Rockville, Md.: AHRQ, 1999). 9. Some expenditures in MEPS are imputed. 10. For an analysis of phase-specific costs, see M.L. Brown et al., Estimating Health Care Costs Related to Cancer Treatment from SEER-Medicare Data, Medical Care 40, no. 8 (2002): IV-104 IV N. Duan, Smearing Estimate: A Nonparametric Retransformation Method, Journal of the American Statistical Association (September 1983): N. Duan and W.G. Manning, A Comparison of Alternative Models for the Demand for Medical Care, Journal of Business and Economics Statistics 1, no. 2 (1983): L. Ries et al., SEER Cancer Statistics Review, , seer.cancer.gov/csr/1973_1999, Tables I-11 and I-12 (14 March 2003). 14. The odds ratio for being uninsured if Hispanic was 2.52 (95 percent confidence interval: 1.45, 4.39). Full regression results are available on request from the authors; send an request to Ken Thorpe, kthorpe@sph.emory.edu. 15. Excluding patients with total spending above the ninety-fifth percentile, we find that average total spending in the sample was $5,018. Spending levels by insurance status were as follows: Medicare, $4,994; Medicaid, $5,434; private, $5,250; and uninsured, $3,051. The difference between spending among uninsured patients and the sample mean is significant at the 5 percent level. As in the main analysis, we also detect significant differences for spending on inpatient and physician services and out-of-pocket spending. 16. See, for example, American Cancer Society and National Comprehensive Cancer Network, Prostate Cancer Guidelines for Patients, Version III, October 2002, index.htm (14 March 2003). 17. For a given event, payments from secondary payers are recorded as private expenditures. The out-ofpocket category includes spending by patients and their families, net of payments by primary and secondary insurers. Events for Medicare beneficiaries in MEPS may be associated with costs to Medicare and private insurers as well as out-of-pocket payments. 18. Complete results are available from the authors on request. 19. Centers for Medicare and Medicaid Services, Breast and Cervical Cancer Prevention and Treatment Activity Map, 31 December 2002, (14 March 2003). W April 2003