Financial Toxicities Persist for Cancer Survivors Irrespective of Current Cancer Status: An Analysis of Medical Expenditure Panel Survey

Cancer is the second leading cause of death in the United States and is projected to cost more than 608,570 lives in 2021 (1). Despite the high mortality associated with cancer, substantial progress has been made against cancer in recent decades (1). With improved treatments and newly discovered drugs, the cancer death rate declined by 31% between 1991 and 2018 (1). Although substantial progress has been made to improve survivorship and reduce mortality associated with cancer, the additional burden of cancer-related financial distress has emerged as a matter of serious concern (2). The financial toxicity of cancer (3) and out-of-pocket (OOP) burden on cancer survivors have garnered considerable attention from researchers, as well as policy makers, in recent times (3–6).

Increasing numbers of cancer survivors are now living longer, sometimes without requiring active treatment while in remission (7). However, long-term survivors may report significant symptom burden (8, 9), posttreatment adverse events such as fatigue and pain (9, 10), and treatment-related late toxicities (11). Moreover, psychologic distress, anxiety, depression, and insomnia are pronounced among long-term cancer survivors (9, 12) and may require additional treatments, resulting in increased costs. Previous studies have examined the costs affecting previously versus recently diagnosed cancer survivors (13–15); however, none of these studies specifically examined OOP costs across cancer types. A 2018 study using 2008–2012 Health and Retirement Study data reported significantly higher total costs for recently diagnosed cancer survivors compared with long-term survivors; however, distinctions were not made across cancer types (13). Similarly, two studies using 2001–2007 and 2008–2010 Medical Expenditure Panel Survey (MEPS) data reported significantly higher OOP expenses among both recently and previously diagnosed cancer survivors compared with noncancer controls, without examining cancer types (14, 15).

The goal of this study is to assess OOP expenditure by cancer status and cancer types across sex. We used the current condition designation of MEPS to stratify cancer survivors between those with a current cancer condition and those with no current cancer condition (16), and examined OOP expenditure by cancer types for these subgroups. We examined OOP expenditure across cancer types because treatment approaches and survival for different cancers vary considerably (7), which may in turn cause variations in OOP expenditure. In addition, we stratified our analysis by sex because, as demonstrated for other health conditions such as diabetes, disease-specific OOP expenditure may vary across sex (17). Adopting a more granular approach compared with the previous studies, we examined average OOP expenditure by current versus no current cancer condition and by specific cancer types across sex, which will help facilitate health policy discussions and formulate better targeted intervention strategies.

Data for our study were obtained from MEPS, a nationally representative survey of the noninstitutionalized U.S. population, which collects information on health care use and expenditure (18). The survey oversamples minority groups and provides person weights in the released public use datasets (19). The MEPS design and data collection process have been described elsewhere (20, 21). For our study, we pooled multiple years of data (2009–2015 and 2018) and adjusted the survey weights accordingly (22). Among several publicly available MEPS data files, we used the Full Year Consolidated file—which provided information on demographics, socioeconomic status, insurance coverage, health status, ever having cancer, and health care expenditure (19)—and the Medical Conditions file, which provided information on select current clinical conditions, including cancer (16). MEPS data were deidentified and publicly available, therefore, our study was exempt from Institutional Review Board approval.

In the survey, respondents 18 years or older were asked “Have you ever been told by a doctor or other health professional that you had cancer or a malignancy of any kind?”; those responding yes to this question were then asked “what kind of cancer was it?” (19, 23). On the basis of the responses to these two questions, we identified individuals as cancer survivors and those with no history of cancer.

Information about current conditions was obtained from the Clinical Classifications Software (CCS) codes or CCS Refined (CCSR) codes. Among the cancer survivors, those who had a cancer-specific CCS or CCSR code were identified as (i) survivors with a current cancer condition; the rest of the cancer survivors were identified as (ii) survivors with no current cancer condition. In MEPS, the current condition was defined as “any clinical condition which had an associated health care event or which was being actively experienced by the respondent during the survey year” (16). Thus, respondents with a current cancer condition were those who either had a health care event or reported that they actively experienced cancer in the survey year. Respondents who had CCS or CCSR code for more than one type of cancer or responded that they had history of more than one prior cancer were classified under “two or more cancers” category.

MEPS used International Classification of Diseases, Ninth Revision (ICD-9)-based CCS codes to report current conditions until 2015 and transitioned to ICD-10–based CCSR codes in 2018 (16). Because neither CCS nor CCSR codes were publicly available for the years 2016 and 2017, we were unable to identify cancer cases with a current condition for 2016 and 2017, and excluded these 2 years. Consequently, our final analytic sample consisted of respondents pooled for the years 2009–2015 and 2018. Cases with mismatched cancer types in the survey response and the current condition designation were excluded.

Total OOP expenditure per person per year was the primary outcome variable in our analysis. The total OOP expenditure was the sum of all-cause OOP expenditure incurred per person per year for any health care event, including office-based visits, outpatient visits, emergency room (ER) visits, inpatient stays, prescription medication purchases, home health care events, and other medical equipment and services use (19). All dollar values were inflation adjusted to 2018 U.S. dollars using the consumer price index for Medical Care (24). Expenditure data in MEPS were primarily self-reported with a subset of the responses verified with the health care providers (19). Even with the possibility of underestimation of cost in MEPS (25), as established in previous studies, use of MEPS data enabled us to examine OOP burden of cancer at the national level (14, 26).

The covariates in each of the estimation model were age, cancer types, race/ethnicity, marital status, educational attainment, income level, insurance status, survey year, number of comorbid conditions, and self-reported health status. All covariates except age were categorical variables (Table 1). In the race/ethnicity variable, non-Hispanic White, non-Hispanic Black, and Hispanic were separate categories, while all other race/ethnicities were grouped together into the “Others” category. Marital status was dichotomized to two groups: single (which included individuals who never married or were widowed, divorced, or separated) and married. Educational attainment was categorized into: less than high school diploma, high school diploma, college education or higher, and missing. Income level was categorized into: <200% of federal poverty level, 200% to <400% of federal poverty level, and ≥400% of federal poverty level (27). Insurance status had five categories: private (employer sponsored), private (nonemployer sponsored), Medicaid/dual eligible, Medicare, and Uninsured. The Medicaid/dual eligible category included the individuals who were covered by both Medicaid and Medicare.

Because our data consist of three types of participants (i) those without cancer (ii) survivors with current cancer condition, and (iii) survivors with no current cancer condition, a substantial number of survey respondents had zero OOP expenditure. To account for the heterogeneous samples and zero inflation, we adopted a two-part regression model. We used logistic regression as the first part to model the probability of incurring any expenditure and used generalized linear model regression with log link and gamma distribution as the second part to model the nonzero expenditure (28). This technique was used in the second part because the gamma distribution models the non-negative and right-skewed expenditure data appropriately, whereas the log link helps avoid retransformation (28). Also, we stratified the analyses by those with current cancers and those with no current cancer. Within each analysis, types of cancer were used as covariate.

We estimated the adjusted average OOP expenditure for several cancer types (29). A permutation test was used to estimate the P values for the OOP expenditure difference for each category compared with the “no history of cancer” reference category (30). The permutation test is a nonparametric method which allowed us to construct the empirical null distribution of the incremental mean values for each cancer category with respect to the “no history of cancer” reference. P values represent the statistical significance obtained using 1,000 permutated replicates to test the hypothesis that the estimated average OOP expenditure for each cancer category is different than the “no history of cancer” category (two-sided P value; ref. 30). We conducted the permutation test in several steps. First, we permuted the outcome variable (i.e., OOP expenditure) 1,000 times. Then we estimated the average OOP expenditure for all 1,000 permutated outcome variables by applying the two-part model, which formed the empirical null distribution. Finally, two-sided P values were obtained by comparing the estimated OOP expenditure from the actual data with the empirical null distribution generated through permutation. We conducted analyses by stratifying our sample by “current cancer condition” and “no current cancer condition” status. All analyses for male and female survivors were conducted separately.

To compare the differences in cancer-attributable OOP expenditure between female and male survivors, we first subtracted the estimated OOP expenditure for the “no history of cancer” category from each cancer type for women and men separately. Subtracting these cancer-attributable OOP expenditure values for men from the respective values for women yielded the incremental cancer-attributable OOP expenditures for each cancer type. The P values were obtained by comparing these differences in estimated OOP expenditures between women and men to the respective differences in 1,000 replicate data.

All analyses were conducted in SAS 9.4 (SAS Institute; RRID:SCR_008567) and Stata15 (StataCorp; RRID:SCR_012763) software, and two-sided P < 0.05 was considered statistically significant. We incorporated survey weights in all our descriptive and covariate adjusted analyses and employed survey specific commands (i.e., svyset and svy: prefix) in Stata.

MEPS data analyzed in this study are publicly available from the Agency for Healthcare Research and Quality website at: https://meps.ahrq.gov/data_stats/download_data_files.jsp.

Our study sample included 189,285 adult individuals (weighted N = 233,221,635) with an average age of 46.61 years. The weighted percentage of non-Hispanic White, non-Hispanic Black, Hispanic, and other race/ethnicity was 64.88%, 11.83%, 15.35%, and 7.95%, respectively. The study sample included 15,010 cancer survivors (weighted n = 22,631,973) with average age of 63.99 years. Among the cancer survivors, the weighted percentage of non-Hispanic White, non-Hispanic Black, Hispanic, and other race/ethnicity was 84.78%, 6.19%, 5.78%, and 3.25%, respectively. Of the cancer survivors, 10.57% (weighted percentage) did not have any high school diploma and 28.27% lived below 200% of the federal poverty level.

The average age of the female cancer survivors (62.44 years) was lower than the average age of the male survivors (66.06 years). Among the 8,902 (weighted n = 12,910,571) female survivors, 42% had a current cancer condition; among the 6,108 (weighted n = 9,721,402) male survivors, 52.39% had a current cancer condition.

Table 1 illustrates the sociodemographic characteristics of the study sample stratified by sex and cancer status (i.e., no history of cancer, current cancer condition, no current cancer condition). The percentage of non-Hispanic White respondents was similar between survivors with a current cancer condition (female 82.18%, male 85.71%) and survivors with no current cancer condition (female 84.41%, male 87.37%), whereas it was lower among those with no history of cancer (female 62.41%, male 63.09%). There was no substantial difference in educational attainment between respondents with a current cancer condition versus no current cancer condition among either female or male cancer survivors. Most cancer survivors had income ≥400% of the federal poverty level, with a higher percentage of male survivors (current cancer condition 51.60%, no current cancer condition 50.21%) in this category compared with female survivors (current cancer condition 43.10%, no current cancer condition 38.99%). Although the uninsured rate was similar among female and male survivors with a current cancer condition, among the survivors with no current cancer condition, more women (6.61%) were uninsured than men (3.94%).

Among female cancer survivors with a current cancer condition, those with breast cancer ($1,730, P < 0.001), lung cancer ($1,679, P = 0.009), colon cancer ($1,595, P = 0.010), melanoma ($1,783, P = 0.002), non–Hodgkin lymphoma ($1,656, P = 0.018), nonmelanoma skin cancer (NMSC)/other skin cancer ($2,118, P < 0.001), and two or more cancers ($2,310, P < 0.001) had statistically significantly higher OOP expenditures compared with the females with no history of cancer ($853); however, the difference was not statistically significant for females with a current cervical cancer condition ($882, P = 0.855; Table 2).

Among female cancer survivors with no current cancer condition, those with cervical cancer ($1,207, P = 0.007), breast cancer ($1,364, P < 0.001), melanoma ($1,396, P = 0.015), NMSC/other skin cancer ($1,506, P < 0.001), and two or more cancers ($1,578, P = 0.007) had significantly higher OOP expenditures compared with the females with no history of cancer ($857; Table 2).

Among male cancer survivors with a current cancer condition, those with prostate cancer ($1,457, P < 0.001), lung cancer ($1,131, P = 0.027), colon cancer ($1,471, P = 0.001), melanoma ($1,474, P < 0.001), non–Hodgkin lymphoma ($1,653, P = 0.005), NMSC/other skin cancer ($1,789, P < 0.001), bladder cancer ($2,157, P < 0.001), and two or more cancers ($2,641, P < 0.001) had statistically significantly higher OOP expenditures than men with no history of cancer ($621; Table 3).

Among male cancer survivors with no current cancer condition, those with prostate cancer ($1,152, P = 0.002), colon cancer ($966, P = 0.028), melanoma ($1,351, P < 0.001), NMSC/other skin cancer ($1,478, P < 0.001), bladder cancer ($1,321, P = 0.019), and two or more cancers ($1,433, P = 0.009) had significantly higher OOP expenditures compared with men with no history of cancer ($621; Table 3).

Table 4 shows incremental cancer-attributable OOP expenditures for female cancer survivors compared with male cancer survivors. Among survivors with current cancer condition, cancer-attributable OOP expenditures for females with two or more cancers was significantly lower than for males with two or more cancers (difference in cancer attributable OOP = −$564, P = 0.021). Among cancer survivors with no current cancer condition, cancer-attributable OOP expenditures for females with NMSC/other skin cancer was significantly lower than for males with NMSC/other skin cancer (difference in cancer attributable OOP = −$208, P = 0.044; Table 4).

In this nationally representative study, we estimated average total per year OOP expenditures for several common cancer types among survivors with current and no current cancer conditions. Our results show that the OOP expenditures among survivors with a current cancer condition of breast cancer (female only), prostate cancer (male only), lung cancer, colon cancer, melanoma, non–Hodgkin lymphoma, NMSC/other skin cancer, bladder cancer, and two or more cancers were significantly higher than the OOP expenditures among individuals with no history of cancer of respective sex. These differences were observed in female survivors with breast cancer; male survivors with prostate, lung, colon, and bladder cancer; and survivors of both sexes with melanoma, NMSC/other skin cancer, and two or more cancers even when survivors had no current cancer condition. Among women with cervical cancer, average OOP expenditure was not significantly higher for those with a current cancer condition compared to women with no history of cancer; however, it was higher for those with no current cancer condition compared with women with no history of cancer.

As expected, we observed higher OOP expenditure among survivors with a current cancer condition compared with those with no current cancer for most cancer types (except cervical cancer). The higher OOP estimates are likely attributable to the greater health care needs among individuals recently diagnosed with cancer (31, 32). Cancer treatment incurs its highest costs in the initial and terminal phases of care, and the cost is usually lower in the continuing phase. In addition to cancer-related care, additional health service needs, such as home health care and mental health care, are elevated among recently diagnosed survivors. According to Chesney and colleagues, home health care is utilized by 43.7% of elderly cancer survivors in the first month after surgery, and the percentage decreases to 12.6% 5 years after surgery (33). The initial treatment cost, coupled with the elevated supportive health care needs, may have resulted in the higher OOP expenditures reported in our study among those with a current cancer condition.

One notable finding in our study is that the OOP expenditures for survivors with no current cancer condition for several cancer types (breast cancer among women; prostate, lung, colon, and bladder cancers among men; and melanoma, NMSC/other skin cancer, and two or more cancers among both sexes) were significantly higher compared to those with no cancer history. This finding highlights the persistence of higher health care spending among survivors who do not currently experience cancer or actively receive treatment for cancer. Although the maintenance phase of cancer care may incur lower costs than the initial phase (31), long-term cancer survivors may still experience heightened health needs due to several persistent psychologic and physiologic conditions. Compared with the general population, significantly higher depression and anxiety have been reported among younger (<60 years) long-term cancer survivors (34). Although the literature on depression and anxiety related to OOP burden in cancer survivors is lacking, total financial burden is well reported (35, 36). Diagnosis of depression results in around 32% higher total expenditures among cancer survivors (35), which may be associated with higher OOP expenditure. In addition to mental health care, supportive services such as home health care are used by more than 12% of long-term cancer survivors (33). The persistent mental and home health care needs are possible reasons for the higher OOP expenditures among long-term survivors with no current cancer condition.

An interesting finding in our study was that OOP expenditure among women with a current cervical cancer condition was not significantly higher than women without a history of cancer. A possible explanation for this finding may be the way the treatment-related cost is transferred to the survivors by the insurers. Although extensive treatment may be required for cervical cancer (37, 38), the insurers transfer only a fraction of the total treatment-related costs to the patients (39). Blanco and colleagues reported that the median OOP cost for commercially insured women with cervical cancer in the first 12 months after diagnosis was $2,253, which was only 3.9% of the total treatment-related cost (39). This lower cost transfer to recently diagnosed patients with cervical cancer may help explain the reduced OOP burden on this subgroup.

In contrast, we found that OOP expenditure for cervical cancer survivors with no current cancer was significantly higher than the OOP expenditure for women without a cancer history. Substantial physiologic and psychologic needs (40–43) demonstrated by long-term cervical cancer survivors may explain this finding. Long-term cervical cancer survivors experience several physiologic issues, many of which are associated with treatment interventions in the pelvic region (40). Treatment-related adverse events include bladder dysfunction, gastrointestinal complications, sexual dysfunction, and lymphedema (40–43). Bladder symptoms are very common, with 96.2% of cervical and endometrial cancer survivors reporting bladder storage issues and 82.7% reporting incontinence issues 1 year after treatment, and these percentages are significantly higher than in people with no cancer history (40, 41). In addition, lymphedema, chronic radiation proctitis with late onset, and sexual dysfunction may affect long-term cervical cancer survivors (40, 42). The clinical care related to these physiologic conditions is the most likely cause of higher OOP burden observed in this subgroup. This finding underscores the need to provide financial support to cervical cancer survivors even when they are a few years removed from their cancer diagnosis.

Similar to cervical cancer, lymphedema is observed in more than 40% of breast cancer survivors and in lower percentages among several other cancers (44). In addition, chronic radiation proctitis is observed in prostate, urinary bladder, uterine, and anal cancers, where radiotherapy poses a risk of rectum injury (45). In our study, some of these cancer types, namely breast cancer (women) and prostate and bladder cancers (men), among survivors with no current cancer condition demonstrated significantly higher OOP expenditures compared with individuals without a cancer history. This is an indication that the long-term adverse effects related to cancer treatment may prevent cancer survivors’ OOP costs from returning to their precancer level. Specific aspects of these long-term adverse effects causing higher OOP costs should be investigated further in future research.

In addition to total OOP expenses incurred by female and male survivors separately, we investigated incremental cancer-attributable costs for female survivors compared with male survivors. We observed significantly lower cancer-attributable OOP expenditures for female survivors only among those with two or more cancers (among survivors with current cancer) and NMSC/other skin cancer (among survivors with no current cancer). These results suggest that sex does not play a significant role in OOP expenditure variations across most cancer types.

Cancer is physically and psychologically debilitating, and it reduces survivor's ability and engagement to work. Considering this aspect of cancer, it is vitally important to adopt policy actions targeting the most vulnerable survivors. A good example of a federal initiative to alleviate financial distress related to cancer screening and diagnosis is the National Breast and Cervical Cancer Early Detection Program (NBCCEDP). This program reduces financial barriers related to breast and cervical cancer screening and diagnosis among underserved women and provides Medicaid access after diagnosis. Public health interventions similar to the NBCCEDP for other cancer types could lessen the financial burden on a substantial number of cancer survivors. To implement such interventions for other cancers in a cost-efficient manner, providing targeted assistance to the individuals with most in need is of vital importance. Targeting long-term survivors with financial intervention is equally important as targeting the survivors with current cancer because both groups may experience high OOP burden depending on the cancer type. To that extent, our study reports the specific cancer types with high OOP expenditures among survivors with current and no current cancer across both sexes, which may help identify the most financially vulnerable cancer survivors.

Our study was conducted using nationally representative data, which is a strength of this study. Despite this strength of our study, there are a few limitations. First, although based on nationally representative general adult population, the study may not be representative of cancer survivors because survey participants are usually a self-selected group in the general population and they are generally healthier than the nonparticipants (46). The self-reported nature of MEPS carries a possibility of recall bias (47); however, possibilities of self-selection bias prevalent in web surveys (48) is potentially reduced through the implementation of personal interviews (49) with a population-representative complex survey design in MEPS. Second, the cost amounts might also be underestimated in MEPS due to the self-reported nature of the survey (50, 51); however, a subset of the responses were verified by MEPS with health care providers data (19). Third, we were unable to incorporate cancer-related clinical information (e.g., age at diagnosis, time since diagnosis, stage) in our analyses due to the unavailability of those variables in MEPS. Finally, we had to exclude 2016 and 2017 data because CCS codes were not available for those years. Despite these limitations, MEPS is a valuable data source because it is the only nationally representative survey that collects health care utilization and expenditure data in the United States (50–52).

We estimated the OOP expenditures for female and male survivors for several cancer types across current versus no current cancer conditions. Financial distress affects all aspects of life for cancer survivors, from negatively affecting the purchase of basic necessities like food to contributing as a risk factor for mortality (53). Amid an ongoing discussion on financial toxicity of cancer, several interventions to alleviate the OOP burden on the survivors have been suggested (54, 55). Our study highlights that the financial distress varies across cancer types among the survivors with and without a current cancer condition. This highlights the need for targeted intervention to alleviate the burden on most financially vulnerable cancer survivors. Our findings will inform policy discussions around the financial toxicity of cancer and help formulate targeted interventions.

M.A. Karim reports grants from Cancer Prevention and Research Institute of Texas during the conduct of the study. No disclosures were reported by the other authors.

M.A. Karim: Conceptualization, data curation, formal analysis, methodology, writing-original draft. R. Talluri: Data curation, formal analysis, methodology, writing-review and editing. S.S. Shastri: Writing-review and editing. H.-C. Kum: Writing-review and editing. S. Shete: Conceptualization, formal analysis, supervision, funding acquisition, methodology, project administration, writing-review and editing.

This study was supported by the NCI grant 5P30CA016672 (S. Shete, PhD), the Duncan Family Institute for Cancer Prevention and Risk Assessment (S. Shete, PhD), the Betty B. Marcus Chair in Cancer Prevention (S. Shete, PhD), and a Cancer Prevention Fellowship supported by the Cancer Prevention and Research Institute of Texas (CPRIT) grant award, RP170259 (to M.A. Karim, PhD; PI: Shine Chang, PhD and S. Shete, PhD). The study funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the article; and decision to submit the article for publication.

No other financial disclosures were reported by authors.

Editorial support was provided by Bryan Tutt, Scientific Editor, Research Medical Library, The University of Texas MD Anderson Cancer Center.