Medical Expenditures of Adult Cancer Survivors under Age 65 in the U.S.

Abstract

Background

This is the first study to provide national estimates of medical expenditures for all adult cancer survivors under age 65. Most studies of expenditures for cancer survivors in this age group have been based on the Medical Expenditure Panel Survey and limited to “affected survivors.”

Methods

MEPS expenditure data for 2001 to 2007 were linked to data identifying all survivors from the National Health Interview Survey, the MEPS sampling frame. The sample was adults 25–64 years old. Propensity-score matching was used to estimate the effects of cancer on average total and out-of-pocket expenditures for all services, and prescriptions separately. Probit models were used to estimate effects on the probability of exceeding different expenditure thresholds.

Results

Mean annual expenditures on all services in 2007 were $16,910 ± $3911 for survivors newly diagnosed with cancer, $7992 ± $972 for survivors diagnosed in previous years, and $3303 ± $103 for other adults. Fifty-three percent of survivors were not identified in MEPS, but only by linking to NHIS. Expenditures for all survivors averaged about $9300, compared to $13,600 for “affected survivors.” For previously diagnosed survivors, the increase in mean expenditures attributable to cancer was about $4000-$5000 annually. On average, relatively little of the increase was paid out of pocket, but cancer nearly doubled the risk of high out-of-pocket expenditures.

Conclusions

Previous MEPS analyses overstated average expenditures for all survivors. Nevertheless, the increase in expenditures attributable to cancer is substantial, even for longer term survivors. Cancer increases the relative risk of high out-of-pocket expenditures.

INTRODUCTION

This study provides national estimates of medical expenditures for cancer survivors in the United States in keeping with the definition of “cancer survivor” adopted by the National Cancer Institute (NCI) and others. Estimates are also provided of the cancer-related increase in medical expenditures per survivor, thus defined. According to NCI, “An individual is considered a cancer survivor from the time of diagnosis through the balance of his or her life.”¹ This definition implies that the population of survivors includes all living individuals ever diagnosed with cancer, and exactly corresponds to the epidemiological concept of prevalence. Cancer prevalence in the U.S. has increased from 6 million to 12 million in the last twenty years.² Now that more people are living longer with a history of cancer, clinicians, public health officials, cancer organizations, and survivors are asking more questions about the long term and late effects of cancer survivorship.^3–5

These questions are all the more urgent because some gains in survival have been accomplished with potentially more damaging treatments—involving higher dosages and combinations of surgical, radiation, chemotherapy, and hormonal therapies.^6,7 Research shows that cancer treatment can have a variety of long-term health effects, including impaired physical and organ function, changes in appearance, sexual dysfunction, incontinence, lymphedema, hormonal imbalances, cognitive difficulties, and fatigue.³ Cancer survivors are subject to psychological stresses and are at increased risk of mental illness.^8,9 One of the greatest risks of survivorship is the possibility of recurrence and heightened risk of second cancers.

Given the potential for long-term health effects, as well as the need for heightened vigilance and monitoring, cancer survivors are likely to incur more medical expenses than other people. However, because of data limitations, information about this aspect of survivorship is limited. In particular, for the population under age 65, there are no published national estimates of average annual medical expenditures for everyone ever diagnosed with cancer. For the population 65 and older, Medicare claims have been used to estimate annual medical expenditures for all survivors, identified by searching for cancer claims over many prior years or linking to cancer registries.^10,11 Most estimates of cancer-related expenditures before age 65 are based on the Medical Expenditure Panel Survey (MEPS), an ongoing national survey conducted by the Agency for Healthcare Research and Quality.^3,2–15 While MEPS has many advantages for studying national medical expenditures, including data for non-Medicare payers, until recently the questionnaire did not systematically identify all prevalent cancer cases.

In this study, we focus on medical expenditures for survivors under age 65. With Medicare’s nearly universal coverage after age 65, medical care is financed much differently before and after age 65. Considering the financing differences, and increased health care needs and expenditures at older ages, it is more meaningful to analyze expenditures for younger and older survivors separately than together. Given space limitations, the availability of Medicare data for older survivors, and the potential usefulness of expenditure estimates for younger survivors to inform the implementation of national health reforms, we limit attention here to the expenditures of younger survivors.

MATERIALS AND METHODS

Sample

The sample comes from two large, nationally representative surveys, the Household Component of the Medical Expenditure Panel Survey (MEPS-HC) and the National Health Interview Survey (NHIS). MEPS-HC provides detailed information about annual medical expenditures, by type of service and source of payment, for each person in the sample. NHIS collects information about health, access to and use of health services, health insurance, and health behaviors. Because NHIS is the sampling frame for MEPS, MEPS data can be linked to data previously collected on the same individuals in NHIS. In this study, questions systematically identifying all cancer survivors that are asked in NHIS (but not in MEPS) are linked to expenditure data in MEPS to make national expenditure estimates for cancer survivors. Penn State’s Human Subjects Protection Committee determined that this analysis of public use data was not human participant research.

The questions identifying cancer survivors are in a section of the NHIS questionnaire that is administered to one randomly sampled adult per household. The sequence begins, “Has a doctor or other health provider EVER told you that you have a cancer or malignancy of any kind?” Anyone answering “yes” is shown a flashcard listing 29 cancer sites and asked, “What kind of cancer was that?” Respondents can report up to three different cancers, and analysts can calculate time since diagnosis from questions about the respondent’s age at each diagnosis.

Until the 2007 panel was fielded, MEPS did not include similar questions to identify all cancer survivors. Consequently, all studies to date of expenditures of cancer survivors based on MEPS have been limited to “affected prevalence,” meaning survey respondents who reported cancer-related utilization or restricted activity days in the survey year. Whenever care is reported in MEPS, respondents are asked, “Was this [visit] for any specific health condition or were any conditions discovered during this visit? … What condition was that?” Similar questions are asked about restricted activity days. For all conditions identified in this way, the MEPS condition files provide “Clinical Classification” codes developed by AHRQ and assigned from text survey responses coded by trained professionals into the International Classification of Diseases, Ninth Revision (ICD-9). We used clinical classification codes 11 to 46, excluding code 23 (non-melanoma skin cancer) and code 44 (unclassified neoplasm) to identify subjects with cancer from the MEPS condition files. Although we mainly identified cancer survivors from NHIS, we also used the MEPS condition files to identify survivors who were newly diagnosed after being interviewed in NHIS.

Data

Each year a new panel of MEPS households is selected from those participating in the previous year’s NHIS. Then data are collected from the panel for two calendar years. As a result, there are two years of data for most individuals in MEPS (e.g., the 2006 MEPS panel has 2006 and 2007 data for a sample drawn from the 2005 NHIS). By the same token, there are data from two panels for each calendar year (e.g., MEPS data for 2007 are from samples drawn from the 2005 and 2006 NHIS).¹⁶ This study pools annual data from 2001 to 2007 to obtain a large enough sample of cancer survivors to make reliable estimates for survivor subgroups and to characterize entire expenditure distributions for survivors.

Considering the cancer data from NHIS, along with cancer reported in the MEPS condition files, we classified each subject in each survey year as (1) a newly diagnosed cancer survivor (in that calendar year), (2) a previously diagnosed survivor (in a prior calendar year), or (3) not a survivor. All newly diagnosed survivors were identified from cancer-related utilization or restricted activity days in MEPS. Skin cancers other than melanoma were ignored. Selecting years when subjects were under age 65 yielded a final sample of 361 observations for newly diagnosed survivors, 2119 for previously diagnosed survivors, and 47,690 for adults who were not survivors. There were two years of data for most unique individuals in the data set.

To produce national estimates for the civilian, non-institutionalized adult population of the United States from the linked sample, we modified the annual person-level survey weights on the MEPS public use files. The adult population was defined as age 25 or older in MEPS. Because college students living in student housing in NHIS are not followed in MEPS, the college students sampled in an alternative way in MEPS are missing NHIS cancer data.¹⁶ The first step in the reweighting procedure was to adjust the original weights for differences in the probability of selection for NHIS sampled adults in different sized families. For each sampled adult with a positive MEPS weight, an adjustment factor was defined as the inverse of the weighted proportion of household adults represented by the sampled adult. To avoid extreme adjustments for some individuals, the factor was capped at 4.25 (the 99 percentile of unconstrained values).

Next we used a “raking” procedure to modify the adjusted weights to match population control totals calculated by calendar year from public use weights for the full MEPS sample by age, sex, race/ethnicity, family income as a percent of poverty, region, and metropolitan location (yes or no). The new weights for the 7 pooled annual samples, which summed to about 7 times the U.S. population, were rescaled to sum to the U.S. population in 2007. Applying the rescaled weights to the pooled, linked adult sample yielded a national estimate of 12.2 ± 0.8 million adult cancer survivors at the start of 2007 and 1.8 ± 0.2 million new cancer cases diagnosed in 2007. The former is consistent with national prevalence estimates based on SEER (11.3 million adults at the beginning of 2006),² but the latter is somewhat higher than the American Cancer Society’s incidence estimates (1.44 million in 2007).¹⁷ Notice that survivors who died in 2007 are included in the MEPS sample, so the combined total of 14 million survivors (12.2 + 1.8 million) is the number alive at any time in 2007, not cancer prevalence at the end of the year.

Expenditures for earlier years were inflated to 2007 with the Personal Health Care Expenditure (PHCE) price index developed by the Office of the Actuary, Centers for Medicare and Medicaid Services.¹⁸ Separate components of the PHCE index were applied to hospital, professional, dental, prescription, medical equipment and vision, and home health expenditures.

Analysis

Descriptive statistics were tabulated by cancer status (newly diagnosed survivor, previously diagnosed survivor, not a cancer survivor), without survey weights. Weighted population estimates, mean expenditures by type of service (e.g., inpatient hospital, outpatient hospital, professional visits, prescriptions) and source of payment (out-of-pocket, private insurance, Medicare, Medicaid, other), and expenditure shares were tabulated by survivor status, using the modified survey weights. Confidence intervals and Z-tests of paired comparisons involving weighted national estimates were constructed from standard errors adjusted for the complex survey design, using survey estimation procedures in SAS 9.1 and Stata 11.

The effects of cancer on annual expenditures are quite different for newly diagnosed survivors (who would mostly be in treatment during the calendar year) and previously diagnosed survivors (who would mostly not be in treatment), so we focused on previously diagnosed survivors in estimating increases in expenditures attributable to cancer. While the short-term effects of cancer on expenditures are direct and relatively obvious (that is, services provided during primary treatment are likely to be identified by MEPS respondents or codes on claims as “cancer-related”), the longer term effects on survivors in the years after treatment are less direct and obvious. Furthermore, if newly diagnosed patients who were undergoing treatment usually reported the expense as cancer-related in MEPS, then estimates of cancer-related expenditures published in earlier MEPS studies^12–13,15 should be reasonably accurate for recently diagnosed cases. Linking back to the cancer questions in NHIS mainly improves expenditure estimates for longer term survivors, so we focus on them here. Additionally, the MEPS sample of long-term survivors is much larger.

To estimate increases in mean total and out-of-pocket expenditures both directly and indirectly attributable to cancer for previously diagnosed survivors, we used propensity-score matching. The matching analyses were limited to a subsample 40–64 years of age, because the small number of survivors in the age group from 25 to 40 presented difficulties in constructing a comparable group of young controls. Considering gender differences in service utilization and cancer types, we performed separate matching analyses for males and females. Four expenditure outcomes were considered: total and out-of-pocket expenditures for all services, and total and out-of-pocket expenditures for prescriptions. Physician-administered drugs, such as chemotherapy agents, were not counted as prescription expenses, but were included in the total for all services.

In the matching analyses, we sought to estimate “the average effect of treatment on the treated (ATT),” in this case the average effect of cancer on the expenditures of a population with the characteristics of our cancer-survivor sample. Matching has the advantage of estimating the ATT from averaged differences among matched individuals with similar observed characteristics, without imposing any assumptions about the functional form of expenditures.

In practice it is rare to match directly on covariates because of the potentially large number of dimensions across which matches would have to be made. Rosenbaum and Rubin showed that the ATT can be consistently estimated by matching on unidimensional propensity scores, rather than a multidimensional set of covariates.¹⁹ In this case, the propensity score was the predicted probability of being a previously diagnosed cancer survivor, estimated from a data set that included all person-years of MEPS data for previously diagnosed survivors and adults with no cancer history. The propensity scores were estimated as Probits. The covariates included indicators for single years of age, race/ethnicity, education, marital status, Census region, metro location, and dummy variables for survey year. One specification also included indicators for specific chronic conditions (arthritis, asthma, diabetes, emphysema, heart disease, hypertension, and stroke). These conditions were excluded from an alternative specification because of the possibility that they sometimes resulted from the person’s cancer history and should be included in the total cancer effect. For example, anthracycline-based chemotherapy sometimes causes heart problems; premature menopause can alter a woman’s health risks in a many ways; chest radiation carries a risk of lung damage.³ From the MEPS questionnaire, we could not reliably distinguish conditions diagnosed before and after a person’s cancer.

Propensity scores and ATT’s were estimated using PSMATCH2 in Stata,²⁰ without survey weights. Standard errors for ATT’s were estimated from 200 bootstrap replications of the entire estimation procedure (propensity scores and matching), clustering at the person level to account for multiple observations on each individual. To ensure that only comparable individuals were compared, we imposed a “trimming” condition that removed 2 percent of the cancer observations with relatively few close matches among controls. We employed kernel matching, which estimates the ATT as a distance-weighted mean of all comparison group observations, with weights inversely related to the propensity-score differences between each non-cancer observation and the cancer observation. After matching, t-tests for equality of means in the cancer and non-cancer groups were performed on every covariate, separately for females and males, using the PSTEST procedure in PSMATCH2. Rejection rates on these tests (0/42 and 2/42, respectively) were within the expected range given a Type I error rate (p-value) of 0.05. To estimate counterfactual expenditure means for cancer survivors in the absence of cancer, weighted national estimates of actual mean expenditures for previously diagnosed survivors 40–64 years old were reduced by the ATT’s from matching.

To characterize the effect of cancer on the distribution of total and out-of-pocket expenditures, we estimated a sequence of Probit models with a survivorship indicator on the right-hand side, as suggested by Angrist.²¹. The dependent variables were indicators for annual expenditures exceeding each of $0, $1000, $5000, $10,000, $20,000, and $40,000 for total expenditures on all services and $0, $500, $1000, $2000, and $5000 for out-of-pocket expenditures on all services and total and out-of-pocket expenditures on prescriptions. The Probit models used the same samples as the matching estimates and were estimated in Stata using the specially-constructed NHIS-MEPS survey weights described earlier, clustering on the person identification number to account for multiple observations on each individual. The same covariates were used in the Probits and propensity score models. Marginal effects of cancer were calculated as the difference between the cancer and comparison groups in the probability of exceeding each expenditure threshold, holding other covariates at their means. To estimate the counterfactual expenditure distribution for cancer survivors in the absence of cancer, weighted national estimates of the actual percentages of survivors exceeding each level of expenditure were generated for previously diagnosed survivors 40–64 years old; then the actual percentages were altered by the Probit marginal effects. Actual and counterfactual distributions were graphed in terms of the percentages of survivors falling into each expenditure interval, calculated for each interval by differencing the percentages exceeding its lower and upper limits.

RESULTS

Unweighted sample characteristics are shown by cancer status in Table 1. The cancer survivors were considerably older than the comparison group. Thirty-eight percent of each survivor subgroup was in the oldest age group considered (55 to 64 years old), compared to 19% of other adults. The survivors were disproportionately female, non-Hispanic whites, unmarried, and publicly insured. The prevalence of various chronic illnesses was significantly higher among cancer survivors, which would be expected in light of the age differences between the cancer and control groups, but could reflect health effects of cancer and its treatment. Fifty-three percent of survivors in the age group from 25 to 64 were not identified in MEPS, but were only identified by linking to the cancer questions in NHIS.

Table 1.

Unweighted sample characteristics of cancer survivors and other adults 25–64 years of age (NHIS sampled adults linked to MEPS, 2001–2007)

	Cancer survivor		No cancer
	Newly diagnosed	Previously diagnosed
Number	361	2119	47,690
	Percent distribution
Age
25–34	14*	11*	24
35–44	21	20	29
45–54	27	30	27
55–64	38	38	19
Gender
Male	28*	25*	44
Female	72	75	56
Race/ethnicity
Hispanic	13*	10*	20
Black, not Hispanic	15	12	17
White, not Hispanic	69	74	57
Other, not Hispanic	3	5	6
Education
LT high school	16*	18*	21
High school	30	33	31
Some college	25	25	22
College	14	15	15
Post college	15	9	10
Marital status
Never married	18*	16*	20
Married	48	47	55
Divorced/ separated	28	31	22
Widowed	6	6	3
Chronic Conditions
Arthritis	30*	35*	19
Asthma	15*	17*	10
Diabetes	11*	12*	7
Emphysema	4*	3*	1
Heart Disease	11*	14*	7
Hypertension	35*	36*	23
Stroke	6*	4*	2
MSA (yes)	83	77*	81
Health insurance
Any private	70*	66*	68
Public only	20	21	13
Uninsured	10	13	18

^*Cancer survivors in column are significantly different from other adults (p<.05).

Mean annual expenditures on all services for individuals newly diagnosed with cancer in 2007 were $16,910 ± $3911 (Table 2). The mean for survivors diagnosed in previous years was about half as large ($7992 ± $972), but more than twice the mean for adults with no cancer history ($3303 ± $103). Survivors more than 5 years post-diagnosis averaged $7383 compared to $9301 for survivors 1 to 5 years after diagnosis, a difference that was not statistically significant. Although newly diagnosed cases accounted for only 15% of the total number of cancer survivors, they accounted for 28% of total survivor spending. Medicare and Medicaid paid the largest share of expenses for previously diagnosed survivors. Medicare’s share was larger for previously diagnosed survivors than newly diagnosed survivors.

Table 2.

Mean medical expenditures and sources of payment for adult cancer survivors 25–64 years of age (United States, 2007)

	Cancer survivor		No cancer
	Newly diagnosed	Previously diagnosed
Population (millions)	1.0	5.5	147.0
Mean total expenditure*	$16,910*	$7,992*	$3,303
Source of payment (percent of total in parentheses)
Out-of-pocket	$2,159(13%)*	$1,391 (17%)*	$679(21%)
Private health insurance	$11,560(68%)*	$4,325 (54%)*	$1,832(55%)
Medicare	$548 (3%)*	$819(10%)*	$193(6%)
Medicaid	$1,308(8%)*	$763(10%)*	$301 (9%)
Other	$1,335 (8%)*	$693 (9%)*	$298 (9%)

^*Dollar amounts for cancer survivors and other adults are significantly different ( p<0.05).

The combined mean expenditure for all survivors identified from NHIS and MEPS was $9293 ± $1044, about two-thirds the mean for the subgroup identified in MEPS without linking to NHIS ($13,558 ± $2038, not shown in table). Survivors identified only in NHIS averaged $5526 ±$717 per year. Given the higher average expenditures of survivors identified in MEPS, MEPS captured nearly 70% of aggregate national expenditures uncovered by the NHIS link ($47 billion of $68 billion), while counting only half as many survivors as the linked surveys (7 million of 14 million survivors).

Inpatient hospital services accounted for 36% of total expenditures for newly diagnosed survivors (Table 3), closely followed by professional services (32%). Although total expenditures for previously diagnosed survivors were much higher than for adults without cancer, the distributions by type of service were fairly similar. Newly diagnosed survivors paid an average of $2159 out of pocket in 2007 (Table 4), a larger dollar amount but smaller share of total expenditures (13%) compared to previously diagnosed survivors ($1391,17%), who in turn paid more out of pocket but a smaller share of the total compared to adults without cancer ($679, 21%). Prescriptions accounted for the largest share of out-of-pocket expenses for both previously diagnosed survivors (44%) and adults without cancer (39%). However, the average dollar amount paid out-of-pocket for prescriptions was more than twice as large for survivors ($607 compared to $265).

Table 3.

Mean total medical expenditures by type of service for cancer survivors and other adults 25–64 years of age (United States, 2007)

	Cancer survivor		No cancer
	Newly diagnosed	Previously diagnosed
Population (millions)	1.0	5.5	147.0
Mean total expenditure*	$16,910*	$7,992	$3,303
Type of service (percent of total in parentheses)
Inpatient hospital	$6,166(36%)*	$1,755(22%)*	$690 (22%)
Outpatient hospital	$2,284(14%)*	$1,114(14%)*	$396(12%)
Professional services	$5,444 (32%)*	$2,743 (34%)*	$1,051 (33%)
Prescriptions	$2,347(14%)*	$1,691 (21%)*	$754 (24%)
Dental	$406 (2%)*	$385 (5%)*	$289 (9%)
Other	$263 (2%)*	$304 (4%)*	$123 (4%)

^*Dollar amounts for cancer survivors and other adults are significantly different (p<0.05).

Table 4.

Mean out-of-pocket medical expenditures by type of service for cancer survivors and other adults 25–64 years of age (United States, 2007)

	Cancer survivor		No cancer
	Newly diagnosed	Previously diagnosed
Population (millions)	1.0	5.5	147.0
Mean OOP expenditure	$2,159*	$1,391*	$679
Type of service (percent of total in parentheses)
Inpatient hospital	$109(5%)*	$67 (5%)	$25 (4%)
Outpatient hospital	$118(5%)*	$84 (6%)*	$35 (5%)
Professional services	$850 (39%)*	$354 (25%)*	$176(26%)
Prescriptions	$808 (37%)*	$607 (44%)*	$265 (39%)
Dental	$194(9%)*	$195(14%)*	$130(19%)
Other	$79 (4%)*	$84 (6%)*	$47 (7%)

^*Dollar amounts for cancer survivors and other adults are significantly different (p <0.05).

For previously diagnosed survivors in the age group from 40 to 64, propensity score matching showed that the increase in average annual expenditures for all services attributable to cancer was $4452±$1188 for females and $5112±$2062 for males (Table 5). Matching additionally on the presence of specific chronic conditions reduced the estimates to $3631±$1083 for females and $3560±$1433 for males (not shown in table). The increase in out-of-pocket expenditures attributable to cancer was considerably smaller than the increase in total expenditures (Table 5). Females paid only 9% of the cancer-related increase in total expenditures out of pocket, while males paid 16% of the cancer-related increase in total expenditures out of pocket. We estimated that cancer added $832 ±$286 to average annual expenditure on prescriptions for female survivors and $1219±$637 for male survivors. Survivors of both genders paid about 30% of the cancer-related increase in prescription expenses out of pocket.

Table 5.

Estimated effects of cancer on mean annual medical expenditures of previously diagnosed cancer survivors 40–64 years of age (United States, 2007)

	Females (N-1227) Population = 3.5m			Males (N=459) Population = 1.6m
	Actual	If no cancer	CA effect (s.e.)	Actual	If no cancer	CA effect (s.e.)
All services
Total	$8,822	$4,370	$4,452 (606)	$8,521	$3,409	$5,112 (1052)
Out of pocket	$1,456	$1,039	$417 (98)	$1,626	$816	$811 (204)
Prescriptions
Total	$1,875	$1,043	$832 (146)	$1,831	$612	$1,219 (325)
Out of pocket	$702	$442	$259 (57)	$640	$290	$350 (99)

Note: The increase in expenditures attributable to cancer (“CA effect”) was estimated from propensity-score kernel matching, with indicators for single years of age, race/ethnicity, education, marital status, census region, metro location, and survey year as covariates.

Estimates from Probit models of the likelihood of exceeding different expenditure thresholds indicated that previously diagnosed survivors in the age group from 40 to 64 were about twice as likely to experience unusually high levels of total spending because of their cancer history (Figure 1). About 9% of cancer survivors of either gender exceeded $20,000 in total annual expenditures for all services (including 4% above $40,000), compared to a projected 5–6% in the absence of cancer (including 2–3% above $40,000). As shown in Figure 2, about 20% of cancer survivors spent more than $2000 out of pocket (including 5% above $5000), compared to a projected 13%–15% in the absence of cancer (including 3% above $5000). The projected percentage exceeding $5000 in total prescription expenses in the absence of cancer (Figure 3) was fairly similar to the actual percentage (8% with cancer vs. 6% without cancer among female survivors, and 7% with cancer vs. 5% without cancer among male survivors). Cancer added about 3 percentage points to the risk of spending more than $2000 out-of-pocket on prescriptions (Figure 4).

Figure 1.

Effects of cancer on percent distribution of survivors by total medical expenditures (United States, 2007)

Figure 2.

Effects of cancer on percent distribution of survivors by out-of-pocket medical expenditures (United States, 2007)

Figure 3.

Effects of cancer on percent distribution of survivors by total prescription expenditures (United States, 2007)

Figure 4.

Effects of cancer on percent distribution of survivors by out-of-pocket prescription expenditures (United States, 2007)

DISCUSSION

A widely accepted definition equates the population of cancer survivors with the prevalence of cancer, and includes everyone who was ever diagnosed with cancer from time of diagnosis until death. By linking to cancer history from the NHIS for individuals in MEPS, we discovered that about half of the cancer survivors encompassed by this definition were ignored in previous MEPS studies of cancer expenditures. Furthermore, because the survivors missing from earlier studies were those least affected by cancer (i.e., they reported no cancer-related utilization or health effects in MEPS), the average expenditure per survivor with the missing survivors included was a third lower than the average without them. Thus, the change made in the MEPS questionnaire in 2007—to ask everyone if they were ever diagnosed with cancer— should greatly improve the picture of cancer survivorship offered by that survey. In the meantime, because the new cancer questions will not support the analysis of expenditures for all survivors until a sufficiently large sample can be accumulated from MEPS public use files over the next several years, this study provides a more accurate description of the expenditures of survivors than would be available otherwise. Although the NHIS link will not be needed to estimate expenditures for cancer survivors from MEPS in future years, the link could be used to study trends for cancer survivors spanning, the questionnaire change and to study prevalent cases of other diseases that are systematically identified in NHIS but not MEPS.

While concluding that estimates from earlier studies overstate average expenditures for the full population of cancer survivors, we still find that cancer has a considerable effect on annual total medical expenditures of adult survivors, even in the years following diagnosis (roughly $4000-$5000). Our estimates of cancer-related increases in expenditures were somewhat sensitive to matching on co-morbidities: estimates assuming that the entire increase in relative risk of co-morbidities in the cancer sample was attributable to cancer (with comorbidities excluded as covariates) differed by $820 for females and $1550 for males from estimates assuming that none of the increased risk was attributable to cancer (with co-morbidities included as covariates). These two extremes, which bracket the potential contribution of comorbidities to cancer-related expenditures, are well within the confidence intervals surrounding both estimates.

Although the increase in expenditures associated with surviving cancer represents an economic burden on society, only a small share of the aggregate burden falls directly on survivors and their families. However, out-of-pocket expenditures are distributed unevenly among cancer survivors, with about 60% of survivors in the age group from 40 to 64 spending $1000 or less in 2007 and 5% spending more than $5000. The relative risk of spending more than $5000 out of pocket in the years following a cancer diagnosis was nearly double the risk without cancer (5% compared to 3%).

Cancer survivors have a considerable stake in the details of prescription benefits that are negotiated as part of the implementation of national health reforms, including formularies and the tiering of copayments for different drugs based on cost. For longer term survivors, prescriptions accounted for 44% of out-of-pocket expenditures, an average of $600 per person in 2007 compared to about $350 for professional services (the next most costly service in terms of out-of-pocket expenses). Depending on gender, 7–8% of previously diagnosed survivors in the age group from 40 to 64 spent more than $2000 out of pocket on prescriptions, including 1–2% who spent more than $5000 out of pocket. The appropriate level of insurance coverage for costly drugs taken by some cancer survivors remains controversial.^22–23 USA Today reported monthly costs for patented cancer drugs that four years ago were as much as $2500 to $4500 for Gleevec, Avastin, and Herceptin and soared to $10,000 or more for Erbitux and Rituxan.²⁴

Although this is the first study to provide national estimates of average medical expenditures for all adult cancer survivors under age 65 in the United States, the estimates have limitations that should be recognized. All of the data identifying individuals diagnosed with, treated for, or affected by cancer are self-reported. The sample of survivors in any single year of the survey is too small to analyze reliably; consequently, the data must be pooled over multiple years, with adjustments for time trends in inflation and utilization. Given sample size constraints and different methods in MEPS and NHIS for determining cancer type, we present estimates for all cancers combined and ignore expenditures differences by type of cancer. We could not identify recurrences and second cancers, despite their effect on expenditures.

Another methodological consideration is the comparative strength of different data sources for analyzing expenditures associated with incident and prevalent cases. Although high expenditures associated with the treatment of incident cases disproportionately affect the overall average for all survivors, confidence intervals on expenditure estimates for incident cases are widened by the small sample in a general population survey like MEPS. Furthermore, our MEPS estimates of calendar year expenditures for newly diagnosed survivors encompass the earlier part of the year before diagnosis and should not be confused with estimates of the first 12 months of expenditures following a cancer diagnosis. Given these issues of sample size and timing, data drawn from health insurance claims or provider records are probably better suited to studying expenditures of incident cases than MEPS. On the other hand, as long as the questionnaire identifies everyone who was ever diagnosed with cancer, a nationally representative and population-based survey like MEPS is probably better suited to capturing all expenditures (for all services and payers) for all prevalent cases, including those little affected by cancer.