Abstract
Background
The purpose of this study was to conduct a longitudinal analysis of out-of-pocket expenditure (OOPE) trajectories for the assessment of cancer’s lasting financial impact.
Methods
We identified newly diagnosed cancer patients and constructed matched control group of noncancer participants from the 2002-2018 Health and Retirement Study. Outcomes included monthly OOPE for prescription drugs (RX-OOPE_MONTHLY) and OOPE for medical services other than drugs in the past 2 years (non–RX-OOPE_2YR), consumer debt, and new individual retirement account (IRA) withdrawals. Generalized linear models were used to compare OOPEs between cancer and matched control groups. Logistic regressions were used to compare household-level consumer debt or early IRA withdrawal. Subgroup analysis stratified patients by age, health status, and household income, with the low-income group stratified by Medicaid coverage. All statistical tests were 2-sided.
Results
The study cohort included 2022 cancer patients and 10 110 participants in the matched noncancer control group. Mean non–RX-OOPE_2YR of cancer patients was similar to that of participants in the matched control group before diagnosis but statistically significantly higher at diagnosis ($1157, P < .001), 2 ($511, P < .001) years, 4 ($360, P = .006) years, and 6 ($430, P = .01) years after diagnosis. A similar pattern was observed in RX-OOPE_MONTHLY. A statistically significantly higher proportion of cancer patients incurred consumer debt at diagnosis (34.5% vs 29.9%; P < .001) and 2 years after (32.5% vs 28.2%; P = .002). There was no statistically significant difference in new IRA withdrawals. Patients experienced lasting financial consequences following cancer diagnosis that were most pronounced among patients aged 65 years and older, in good-to-excellent health at baseline, and with low income, but without Medicaid coverage.
Conclusions
Policies to reduce costs and expand insurance coverage options while reducing cost-sharing are needed.
Advances in early detection and cancer treatment have substantially improved cancer survival. The 5-year relative survival rate increased from 49% for patients diagnosed between 1975 and 1977 to 67% for those diagnosed between 2010 and 2016 (1). However, high costs of new oncologic therapies, greater adoption, and sometimes prolonged use of these novel therapies threaten the affordability of cancer care and exacerbate the financial hardship patients and families experience (2,3).
Out-of-pocket expenditures (OOPE) for health care are generally higher among cancer patients and long-term survivors than their counterparts without a cancer history (4-7). A Medical Expenditure Panel Survey study reported the average annual OOPE per person was $1000 and $622, for adults with and without a cancer history, respectively (5). Most studies of OOPE among cancer survivors report annual expenditures for a snapshot in time, with little information about cost trajectories. Furthermore, samples from annual household surveys are heterogeneous, with a relatively small proportion of newly diagnosed cancer patients and a larger proportion of survivors diagnosed years prior to the measurement period. Few studies reported cancer patients’ financial hardship trajectories. Shankaran et al. (8) found more than 70% of patients with metastatic colorectal cancer reported major financial hardship 12 months into treatment. Chino et al. (9) reported patients’ willingness to make financial sacrifice for cancer care only decreased slightly, whereas actual sacrifice increased from baseline to 3-month follow-up. Friedes et al. (10) found financial toxicity persisted from baseline to 6-month follow-up. Little is known about whether financial hardship eases years after diagnosis.
Longitudinal assessment of OOPE trajectories allows for comprehensive evaluation of long-term financial consequences from cancer treatments for patients and their families. A lasting financial impact can be especially devastating for older adults who are retired or near retirement age and have limited opportunity to increase earnings. To fill this important research gap, we examined the pattern of OOPE, consumer debts, and new individual retirement account (IRA) withdrawals before and after a cancer diagnosis by comparing cancer patients to a noncancer matched control group from a national cohort of community-dwelling older adults. We stratified the sample by age (younger than 65 years vs 65 years and older), health status, and income. We also compared financial consequences experienced by low-income patients with vs without Medicaid to inform policy discussions on expanding health insurance coverage.
Methods
Data and Study Cohort
We used 2002-2018 data from the Health and Retirement Study (HRS) survey. The HRS is a national, longitudinal biennial survey that collects detailed health, employment, and financial information every 2 years for respondents aged older than 50 years and their spouses (11,12).
We defined the diagnosis interview as the interview that a respondent first reported having a cancer diagnosis. We restricted the study sample of cancer patients to respondents who participated in at least 1 interview prior to diagnosis and referred to the interview preceding the diagnosis interview as the baseline interview (see Figure 1). We used HRS data from 2002 onward because OOPE measurements were inconsistent before 2002. To ensure sufficient duration of follow-up to observe any lasting financial effects, we limited the study sample to patients who were alive at least 6 years (ie, 3 additional interviews) after diagnosis. Respondents diagnosed with cancer between 2014 and 2018 were included but were considered censored as we were unable to assess their survival after 2018. HRS response rates ranged from 86.4% to 89.6% for the study period (13).
Figure 1.
Study timeline (in years) relative to time of diagnosis interview.
We constructed a matched noncancer control group from respondents and spouses to estimate the incremental OOPE, consumer debt, and IRA withdrawal associated with cancer. Respondents whose spouse was diagnosed with cancer were ineligible for the control group because household finances could be impacted by the spouse’s cancer (Supplementary Figure 1, available online). We matched cancer patients to participants in the matched noncancer control group using a 1:5 nearest-neighbor matching with replacement, with demographic, socioeconomic, and health status variables as matching factors. Demographic variables were self-reported and included age, sex (male, female), marital status (married or unmarried, separated, widowed or divorced), race (Black, other [American Indian, Alaskan Native, Asian, Native Hawaiian, and Pacific Islander], and White), and Hispanic ethnicity. We quantified socioeconomic status using 3 binary variables: whether the respondent had a 4-year college degree and was working and if the household annual income was above $75 000. We also matched on health status to participants in the control group for underlying conditions that may affect medical expenditures. We exact matched to the cancer respondents’ baseline interview year to control for time trends and used Mahalanobis distance, a nonparametric approach that does not rely on model specification, for each matching factor (14).
The study protocol was considered exempt by the University of Colorado Multiple Institutional Review Board.
Outcomes
The primary outcome of interest was individual-level OOPE and was calculated using the RAND HRS longitudinal files. The HRS consistently reported OOPE in the following categories since 2002: 1) hospital, 2) doctor visits, 3) outpatient, 4) nursing home, 5) home health care, 6) dentist, 7) special facility costs, and 8) average monthly prescription drugs (RX) expenditures. Except for RX, OOPE for all other categories reported spending in the past 2 years. Therefore, we quantified the primary outcome as OOPE in the past 2 years (ie, since the last survey) for all services other than RX (non–RX-OOPE_2YR) and OOPE for monthly RX (RX-OOPE_MONTHLY).
Other outcomes included household-level consumer debt and new IRA withdrawal. For analyses of household-level outcomes, we excluded 87 households in which both the respondent and spouse were diagnosed with cancer. Household consumer debt was measured as “debts such as credit card balances, medical debts, life insurance policy loans, loans from relatives, and so forth.” We first compared the proportions of cancer patients and participants in the matched noncancer control group with any debt and then compared the debt amount. For new IRA withdrawals, respondents were asked, “Have you (or your spouse) drawn any money or received any payments from this account since previous survey wave?” We compared the proportion of respondents who responded “yes” among those who had 1 or more IRA accounts at baseline. We limited this analysis to households where both the respondent and spouse were younger than age 65 years at diagnosis considering those older than 70 years 6 months would be required to withdraw from their IRA accounts and those who were 65 years and older at diagnosis would reach age 70 years within 6 years of diagnosis. OOPE and the amount of consumer debt were standardized to 2018 dollars and Winsorized at the 99th percentile to lessen the influence of extreme outliers.
Analysis
We employed a panel event study design to examine changes in financial outcomes following a cancer diagnosis (15). Lag and lead indictors represent the pre- and postdiagnosis years and the equivalent years for the matched control group. We examined outcomes using 3 rounds of interviews (approximately 6 years) prediagnosis, the diagnosis interview, and up to 3 interviews postdiagnosis (about 6 years after a cancer diagnosis).
Generalized linear regression models with a gamma distribution and log link were used to estimate changes in average spending over time. Logistic regression models were used to estimate the probability of having any consumer debt or early IRA withdrawal. We applied 2-part models to estimate the change in the amount of RX-OOPE_MONTHLY and consumer debt over time because each variable had a nontrivial proportion of zeros. We adjusted for within-respondent correlation using cluster, a STATA command for clustered standard errors. Interactions between indicators for trend years of interview rounds and cancer diagnosis estimated the difference in OOPE (or consumer debt) between the cancer and control groups. Models were robust to the inclusion of a respondent-level random effect from Generalized Linear Mixed Model.
We conducted stratified analyses based on 3 characteristics of the respondents: aged 65 years and older at diagnosis, health status (good to excellent vs fair to poor) at baseline, and annual household income level (below vs above median income of study sample) at baseline. Of those with annual household incomes below the median, we stratified analyses by Medicaid enrollment to examine whether Medicaid insurance provided financial protection to lower-income cancer patients compared with participants in the matched noncancer control group.
To account for multiple comparisons in testing the hypothesis of lasting financial burden, we reported sharpened false-discovery rate q values alongside P values for estimates from 3 interviews after the diagnosis interview (16). Analyses were performed using Stata v16.1 (StataCorp LLC, College Station, TX, USA). All statistical tests were 2-sided, and P values of no more than .05 were considered statistically significant.
Results
Study Cohort
The study cohort included 2022 cancer patients and 10 110 participants in the matched noncancer control group. Table 1 reports the baseline characteristics in each group. The mean age of cancer patients was 67.2 (SD = 9.5) years. Most patients were White (78.9%; 15.9% were Black and 5.2% were Other) and married (65.6%). More than 25% were college educated, 37.1% were working, 29.5% had an annual household income of more than $75 000, and 25.8% rated their health status as fair to poor. The postmatch standardized differences were below the standard threshold of 10% for all covariates, suggesting balanced distribution of covariates at baseline between the 2 groups.
Table 1.
Descriptive statistics of respondents diagnosed with cancer and the matched control group at the baseline interview, 2002-2018a
| Characteristics | Cancer group (n = 2022) | Matched control group (n = 10 110) | P b | Standardized differencec |
|---|---|---|---|---|
| Sex % (SD)d | ||||
| Female | 51.3 (50.0) | 51.3 (50.0) | >.99 | <0.001 |
| Male | 48.7 (50.0) | 48.7 (50.0) | >.99 | <0.001 |
| Married, % (SD)d | 65.6 (47.5) | 66.0 (47.4) | .76 | 0.007 |
| Age (mean) | 67.2 (9.5) | 67.0 (9.3) | .34 | 0.023 |
| Race, % (SD)d | ||||
| Black | 15.9 (36.6) | 15.9 (36.5) | .98 | <0.001 |
| Othere | 5.2 (22.2) | 5.2 (22.2) | >.99 | <0.001 |
| White | 78.9 (40.8) | 78.9 (40.8) | .98 | <0.001 |
| Ethnicity, % (SD)d | ||||
| Hispanic | 8.6 (28.0) | 8.6 (28.0) | .99 | <0.001 |
| Non-Hispanic | 91.4 (28.0) | 91.4 (28.0) | .97 | <0.001 |
| College, % (SD)d | 26.0 (43.9) | 26.0 (43.9) | .97 | <0.001 |
| Household income ≥$75k, % (SD)d | 29.5 (45.6) | 29.4 (45.6) | .97 | <0.001 |
| Fair/poor health, % (SD)d | 25.8 (43.7) | 25.5 (43.6) | .82 | 0.005 |
| Working, % (SD)d | 37.1 (48.3) | 37.3 (48.4) | .89 | 0.003 |
Characteristics are presented at baseline (ie, the wave before a reported cancer diagnosis) and equivalent wave for participants in the matched noncancer control group.
Statistical significance was determined by 2-sided t tests between the cancer sample and noncancer control group.
The standardized difference is the absolute value of the mean difference between the cancer and control groups divided by the overall standard deviation.
Statistical significance was determined by 2-sided χ2 tests between the cancer sample and noncancer control group.
Other race group included American Indian, Alaskan Native, Asian, Native Hawaiian, and Pacific Islander.
Trends in OOPE
The comparison of non–RX-OOPE_2YR shows similar 2-year spending between the cancer and the noncancer control group in the preperiod but statistically significantly higher non–RX-OOPE_2YR for the cancer group at the diagnosis and 3 subsequent interviews (Figure 2, A). The mean non–RX-OOPE_2YR for cancer patients was $1157 (P < .001), $511 (P < .001), $360 (P = .006), and $430 (P = .01) higher than participants in the matched noncancer control group at diagnosis interview and 2, 4, and 6 years following diagnosis, respectively (Supplementary Table 1, available online). A similar pattern was observed in the monthly RX-OOPE_MONTHLY (Figure 2, D), with the mean monthly RX-OOPE_MONTHLY $14 (P < .001), $7.2 (P = .01), $8.8 (P = .01), and $10.9 (P = .002) higher at diagnosis and 2, 4, 6, years following diagnosis, respectively, for cancer patients (Supplementary Table 2, available online), although the differences were not statistically significant after multiple comparison adjustment.
Figure 2.
Trends in out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group, all ages and by age group, 2002-2018. A) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group, all ages, 2002-2018. B) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for age group 65 years and older at diagnosis, 2002-2018. C) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for age group younger than 65 years at diagnosis, 2002-2018. D) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched noncancer control group, all ages, 2002-2018. E) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched noncancer control group for age group 65 years and older at diagnosis, 2002-2018. F) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched noncancer control group for age group younger than 65 years at diagnosis, 2002-2018. The dashed vertical line indicates the time of the survey wave when the respondent reported a cancer diagnosis. The solid vertical lines represent the 95% confidence intervals. Outcomes are standardized to 2018 dollars and Winsorized at the 99th percentile. Figure 2, A-C, estimates out-of-pocket expenditures less drug spending for the past 2 years using generalized linear models with a gamma distribution and log link. Figure 2, D-F, estimates monthly out-of-pocket drug spending using 2-part regression models with the first stage estimated using logistic regression and the second using a generalized linear model with a gamma family distribution and a log link. All models used standard errors clustered at the respondent level. P values represent whether the cancer group was statistically significantly different from the noncancer group by trend year. *Denotes sharpened q value above 0.05. The sample size of the full sample is 65 098 and is 44 420 and 20 678 for ages 65 years and older and younger than 65 years groups, respectively. OOPE = out-of-pocket expenditures; RX = prescription drugs.
Analyses stratified by age group indicate that although non–RX and RX-OOPE were statistically significantly higher for patients aged 65 years and older (Figure 2, B and E) and younger than 65 years (Figure 2, C and F) at diagnosis interview, the 6-year financial effect of cancer was only observed among patients aged 65 years and older. Nevertheless, compared with participants in the matched noncancer control group, cancer patients younger than 65 years incurred substantially higher non–RX-OOPE_2YR ($1848; P < .001) at diagnosis interview than those aged 65 years and older ($821; P < .001) (Supplementary Table 1, available online). When stratified by baseline health status, cancer patients in both health groups had statistically significantly higher non–RX-OOPE_2YR than participants in the matched noncancer control group at diagnosis interview, and the higher costs continued for at least 2 years for patients in good-to-excellent health (Figure 3, A). Statistically significantly higher cost was only observed 4 years following diagnosis for those in poor-to-fair health (Figure 3, B). For RX-OOPE_MONTHLY, cancer patients in good-to-excellent health had statistically significantly higher costs at diagnosis ($14; P < .001) and all 3 subsequent interviews (Figure 3, C), whereas statistically significantly higher cost was found at diagnosis and 6 years postdiagnosis for those in fair-to-poor health (Figure 3, D). After adjusting for multiple comparison, costs at all subsequent interviews were not statistically significantly higher for both health groups.
Figure 3.
Trends in 2-year nonprescription out-of-pocket expenditures and monthly out-of-pocket drug spending of cancer patients compared with participants in the matched noncancer control group by health status, 2002-2018. A) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for those in good to excellent health at baseline, 2002-2018. B) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for those in fair to poor health at baseline, 2002-2018. C) Trends in monthly out-of-pocket drug spending of cancer patients compared with participants in the matched noncancer control group for those in good to excellent health at baseline, 2002-2018. D) Trends in monthly out-of-pocket drug spending of cancer patients compared with participants in the matched noncancer control group for those in fair to poor health at baseline, 2002-2018. The dashed vertical line indicates the time of the survey wave when the respondent reported a cancer diagnosis. The solid vertical lines represent the 95% confidence intervals. Outcomes are standardized to 2018 dollars and Winsorized at the 99th percentile. Figure 3, A and B, estimate out-of-pocket expenditures less drug spending using generalized linear models with a gamma distribution and log link. Figure 3, C and D, estimate monthly out-of-pocket drug spending using 2-part regression models with the first stage estimated using logistic regression and the second using a generalized linear model with a gamma family distribution and a log link. All models use standard errors clustered at the respondent level. P values represent whether the cancer group was statistically significantly different from the noncancer group by trend year. *Denotes sharpened q value above 0.05. The sample size of respondents in good-to-excellent health is 49 005 and is 16 093 for those in fair-to-poor health. OOPE = out-of-pocket expenditures; RX = prescription drugs.
Results from income-stratified analyses show that compared with participants in the matched noncancer control group, cancer patients in the high-income group had statistically significantly higher non–RX-OOPE_2YR at diagnosis interview and 1 subsequent interview (Figure 4, A), whereas statistically significantly higher non–RX-OOPE_2YR was reported at diagnosis and all 3 subsequent interviews for cancer patients in the low-income group (Figure 4, B). Further stratification by Medicaid enrollment among the low-income group reveals that cancer’s financial consequences on non–RX-OOPE_2YR was mitigated by Medicaid coverage. Around diagnosis, non–RX-OOPE_2YR was $242 (P = .39) and $969 (P < .001) higher for patients in the low-income group with (Figure 4, C) and without Medicaid (Figure 4, D), respectively. For RX-OOPE_MONTHLY, statistically significantly higher costs were reported among cancer patients in the high-income group at diagnosis and all subsequent interviews (Figure 4, E) but only statistically significantly higher at diagnosis after multiple comparison adjustments. For patients in the low-income group, statistically significantly higher RX-OOPE_MONTHLY was only observed at diagnosis interview (Figure 4, F), largely driven by those without Medicaid coverage (Figure 4, H).
Figure 4.
Trends in out-of-pocket expenditures compared with participants in the matched noncancer control group by annual household income, 2002-2018. A) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for those with annual household income above median income, 2002-2018. B) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for those with annual household income below median income, 2002-2018. C) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for those with annual household income below median income and with Medicaid, 2002-2018. D) Trends in 2-year nonprescription out-of-pocket expenditures of cancer patients compared with participants in the matched noncancer control group for those with annual household income below median income and without Medicaid, 2002-2018. E) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched non-cancer control group for those with annual household income above median income, 2002-2018. F) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched non-cancer control group for those with annual household income below median income, 2002-2018. G) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched non-cancer control group for those with annual household income below median income and with Medicaid, 2002-2018. H) Trends in monthly out-of-pocket expenditures on prescription drugs of cancer patients compared with participants in the matched non-cancer control group for those with annual household income below median income and without Medicaid, 2002-2018. The dashed vertical line indicates the time of the survey wave when the respondent reported a cancer diagnosis. The solid vertical lines represent the 95% confidence intervals. Outcomes are standardized to 2018 dollars and Winsorized at the 99th percentile. Figure 4, A-D, estimates out-of-pocket expenditures less drug spending for the past 2 years. Models were estimated using generalized linear models with a gamma distribution and log link with errors clustered at the respondent level. Figure 4, E-H, estimates monthly out-of-pocket drug spending. Models were estimated using 2-part regression models with the first stage estimated using logistic regression and the second using a generalized linear model with a gamma family distribution and a log link. Standard errors were clustered at the respondent level. P values represent whether the cancer group was statistically significantly different from the noncancer group by trend year. *Denotes sharpened q value above 0.05. The sample size for respondents above and below median income is 32 649 and 32 449, respectively. The sample size for those below median income with Medicaid is 4606 and without Medicaid is 27 794. OOPE = out-of-pocket expenditures; RX = prescription drugs.
Trends in Consumer Debt and Early IRA Withdrawal
Figure 5, A, compares the proportion of households reporting having any consumer debt for cancer and noncancer groups. A statistically significantly higher proportion of cancer patients incurred consumer debt at diagnosis interview (34.5% vs 29.9%; P < .001) and 1 subsequent interview (32.6% vs 28.2%; P = .002). The average amount of consumer debt was statistically significantly higher for cancer patients by $748 (P = .01; Supplementary Table 3, available online) at diagnostic interview but not statistically different in the subsequent interviews (Figure 5, B). Analyses stratified by income showed a statistically significantly higher proportion of cancer patients incur consumer debt at the diagnostic and 2 subsequent interviews in the high-income sample (Figure 5, C) and at the diagnosis and 1 subsequent interview in the low-income sample (Figure 5, D), mostly among those without Medicaid coverage (Figure 5, F). Lastly, among 43% of respondents with IRA accounts, new IRA withdrawals did not differ between cancer and noncancer groups (Figure 6; Supplementary Table 4, available online).
Figure 5.
Trends in household consumer debt before, during, and after a cancer diagnosis compared with a matched control group, 2002-2018. A) Share of respondents with any household consumer debt before, during, and after a cancer diagnosis compared with a matched control group, 2002-2018. B) Amount of debt for respondents with any household consumer debt before, during, and after a cancer diagnosis compared with a matched control group, 2002-2018. C) Share of respondents with any household consumer debt before, during, and after a cancer diagnosis compared with a matched control group for those with annual household income above median income, 2002-2018. D) Share of respondents with any household consumer debt before, during, and after a cancer diagnosis compared with a matched control group for those with annual household income below median income, 2002-2018. E) Share of respondents with any household consumer debt before, during, and after a cancer diagnosis compared with a matched control group for those with annual household income below median income and with Medicaid, 2002-2018. F) Share of respondents with any household consumer debt before, during, and after a cancer diagnosis compared with a matched control group for those with annual household income below median income and without Medicaid, 2002-2018. The dashed vertical line indicates the time of the survey wave when the respondent reported a cancer diagnosis. The solid vertical lines represent the 95% confidence intervals. Consumer debt measures the household debt from credit card balances, medical debts, life insurance policy loans, loans from relatives, and so on. Consumer debt is standardized to 2018 dollars and Winsorized at the 99th percentile. Figure 5, A, estimates the share of respondents with any debt using a logistic regression model, and Figure 5, B, estimates the amount of debt using a 2-part model with a logistic regression model in the first stage and the second stage estimated using a generalized linear model with a gamma family distribution and log link. Both models used standard errors clustered at the respondent level. The sample size is 64 137. Figure 5, C-F, estimates the share of respondents with any debt using a logistic regression model. The sample size for respondents above and below median income is 32 038 and 32 099, respectively. The sample size for those below median income with Medicaid is 4554 and without Medicaid is 27 496. P values represent whether the cancer group was statistically significantly different from the noncancer group by trend year. *Denotes sharpened q value above 0.05. OOPE = out-of-pocket expenditures; RX = prescription drugs.
Figure 6.
Share of respondents younger than age 65 years at diagnosis with new IRA withdrawal before, during, and after a cancer diagnosis compared with a matched control group. The dashed vertical line indicates the time of the survey wave when the respondent reported a cancer diagnosis. The solid vertical lines represent the 95% confidence intervals. Respondents reported whether they or their spouse withdrew from their IRA within the past 2 years. The analysis is conditional on respondents who had an IRA account at baseline and both members of the household were aged younger than 65 years at the diagnosis interview. The analysis is further censored if both members of the household were aged younger than 70 years postdiagnosis. A logistic regression model was used with errors clustered at the respondent level. P values represent whether the cancer group was statistically significantly different from the noncancer group by trend year. The sample size is 4918. IRA = individual retirement account.
Discussion
Rising costs of cancer care threatens the financial well-being and health of patients with cancer and their families. In this longitudinal, national study of older adults, we found patients with cancer not only had higher OOPE during the diagnosis interview (ie, within 0-2 years of diagnosis) but also continued to experience higher financial burden for approximately 6 years following diagnosis compared with their counterparts without a cancer history. The lasting financial impact of cancer was largely driven by OOPE for non-RX spending and more pronounced among patients who were aged 65 years and older, in good-to-excellent health prior to diagnosis, and had annual household incomes lower than the sample median (approximately $44 135), and it was especially notable for the patients with annual household incomes below the median and without Medicaid coverage.
The lasting financial consequences of cancer observed here are deeply concerning as many patients age into retirement, if not retired already, with diminishing earning potential. Although Medicare offers some financial protection for patients aged 65 years and older, traditional Medicare fee-for-service plans do not have an annual cap on OOPE as do many private insurance plans or Medicare Advantage ($7550 in 2021) (17). As a result, older cancer patients who underwent high-cost treatments are more susceptible to higher OOPE. Medicaid, on the other hand, offered financial protection for low-income cancer patients. The trends in non–RX-OOPE_2YR exhibit a sharp difference in the long-term financial burden of cancer between those with and without Medicaid (Figure 4, C and D). Low-income patients lacking financial protection from Medicaid were particularly vulnerable; even 5 to 6 years after cancer diagnosis, their mean non–RX-OOPE_2YR was statistically significantly higher than participants in the matched noncancer control group (more than $653) and was also higher than the difference found between high-income cancer patients and the noncancer control group (approximately $291). These findings add to our growing understanding of the benefits associated with Medicaid expansion, which shields low-income patients from further financial devastation (18–23). States choosing not to expand Medicaid leave their residents more vulnerable to the long-term financial consequences of cancer.
Analysis stratified by patients’ health status prior to cancer diagnosis indicated that lasting financial impact of cancer was more pronounced among patients in good-to-excellent health prediagnosis. This likely captures the “health shock” triggered by a cancer diagnosis among otherwise healthy individuals, which reflects the health status of participants in the matched noncancer control group. It is also possible that cancer patients in better health may be treated more aggressively than those in poor health, thus accumulating higher medical costs over time. For patients in fair-to-poor health prediagnosis, a cancer diagnosis was still associated with higher OOPE when compared with their matched control group in fair-to-poor health who presumably also incurred higher medical costs.
Although most cancer patients in our study were aged older than 50 years, they appeared to partly finance their care through consumer debt rather than withdrawing savings from their IRAs. Given the few years remaining in the labor market to contribute earnings to IRAs, it is reassuring to find that cancer patients were not more likely to have new IRA withdrawals than their counterparts without a cancer history. Approximately one-third of cancer patients report consumer debt. Although the HRS does not ask if borrowing was for cancer treatment, these observations highlight the importance of patients having protection against health expenditures and lowering OOPE to levels that are affordable by most households.
This study has limitations. First, both cancer diagnoses and OOPE were self-reported. The HRS collects limited information on cancer stage or treatments received and self-reported OOPE, especially non–RX OOPE that spans a 2-year period, is subject to recall bias. Second, although information on cancer site is available, the number of newly diagnosed cancer patients was too small to analyze findings by cancer site. Third, our analysis excluded patients who died within 5 years of diagnosis to allow a sufficient observation period, which may underestimate OOPE because of high medical costs associated with end-of-life care. Fourth, the time of cancer diagnosis spans from 2004 to 2018; thus, OOPE reported may understate more contemporary and often also more costly treatment. Lastly, financial burden estimated from the HRS, which is limited to adults aged 50 years and older, would likely be an underestimation as research has showed financial toxicity disproportionally affects younger cancer patients (24).
Our findings suggest that cancer has immediate and lasting financial consequences and point toward 3 policy recommendations to ease the short- and long-term financial burden of cancer. First, policies that support Medicaid expansion are critical to protect low-income patients from further financial loss. Patients with few assets prior to diagnosis are vulnerable to subsequent loss that could have generational implications following their death. Increasing subsidies in the Health Insurance Marketplace for patients who exit labor market following a cancer diagnosis may also provide financial relief. Second, bringing fee-for-service Medicare in line with private insurance coverage and Medicare Advantage by including an annual cap on OOPE would protect older patients from long-term financial loss. Given the ever-increasing expenses associated with cancer treatment, OOPE can quickly add up, making it difficult to cover the costs of daily living for older adults with fixed incomes and little to no opportunity for replenishment. Finally, policies that curb the cost of cancer care so that OOPE is less burdensome will benefit all patients, regardless of income at diagnosis. Few families are sufficiently well off to absorb the long-term cost of care that remains a financial burden well into the future. As of January 2019, there are 16.9 million cancer survivors in the United States, representing approximately 5% the population (25). Therefore, policies to reduce cancer’s OOPE are impactful and affect a large and growing segment of our national economy.
Funding
Shih acknowledges funding from the National Cancer Institute (R01CA225646, R01CA225647 and CCSG P30 CA016672). Bradley and Owsley were partially supported by National Cancer Center Core Grant P30CA46934.
Notes
Role of the funders: The National Cancer Institute funded this research but had no role in the design or conduct of the study; analysis or interpretation of the data; review, or approval of the manuscript; or the decision to submit the manuscript of the publication.
Disclosures: YTS served as a consultant for a review panel for Pfizer Inc and an advisory board for AstraZeneca in 2019. KRY serves on the Flatiron Health Equity Advisory Board. All other authors have no conflicts of interest to disclose on the subject of this manuscript. YTS and KRY, who are JNCI Associate Editors and co-authors on this paper, were not involved in the editorial review or decision to publish the manuscript.
Author contributions: Conceptualization: YTS, KMO, LHN, KRY, CJB. Resources, funding acquisition, and supervision: YTS, CJB. Data curation, software, formal analysis, and project administration: KMO. Methodology and visualization: YTS, KMO, KRY, CJB. Validation, investigation, writing-original draft, and writing-review & editing: YTS, KMO.
Data Availability
The raw/processed data required to reproduce the above findings cannot be shared under the data use agreement between University of Colorado and HRS Survey Research Center.
Supplementary Material
Contributor Information
Ya-Chen Tina Shih, Section of Cancer Economics and Policy, Department of Health Services Research, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
Kelsey M Owsley, Department of Health Systems, Management, and Policy, University of Colorado Comprehensive Cancer Center and Colorado School of Public Health, Aurora, CO, USA.
Lauren Hersch Nicholas, Department of Health Systems, Management, and Policy, University of Colorado Comprehensive Cancer Center and Colorado School of Public Health, Aurora, CO, USA.
K Robin Yabroff, Surveillance and Health Equity Science Department, American Cancer Society, Atlanta, GA, USA.
Cathy J Bradley, Department of Health Systems, Management, and Policy, University of Colorado Comprehensive Cancer Center and Colorado School of Public Health, Aurora, CO, USA.
References
- 1.American Cancer Society. Cancer Facts & Figures 2021. Atlanta, GA: American Cancer Society; 2021. [Google Scholar]
- 2. Sullivan R, Peppercorn J, Sikora K, et al. Delivering affordable cancer care in high-income countries. Lancet Oncol. 2011;12(10):933–980. [DOI] [PubMed] [Google Scholar]
- 3.Institute of Medicine. Delivering High-Quality Cancer Care: Charting a New Course for a System in Crisis. Washington, DC: Institute of Medicine; 2013. [PubMed] [Google Scholar]
- 4. Narang AK, Nicholas LH. Out-of-pocket spending and financial burden among Medicare beneficiaries with cancer. JAMA Oncol. 2017;3(6):757–765. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Ekwueme DU, Zhao J, Rim SH, et al. Annual out-of-pocket expenditures and financial hardship among cancer survivors aged 18-64 years–United States, 2011-2016. MMWR Morb Mortal Wkly Rep. 2019;68(22):494–499. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Langa KM, Fendrick AM, Chernew ME, et al. Out-of-pocket health-care expenditures among older Americans with cancer. Value Health. 2004;7(2):186–194. [DOI] [PubMed] [Google Scholar]
- 7. Park T, Hwang M. Health care use and expenditures attributable to cancer: a population-based study. Res Social Adm Pharm. 2021;17(7):1300–1305. [DOI] [PubMed] [Google Scholar]
- 8. Shankaran V, Unger JM, Darke AK, et al. S1417CD: a prospective multicenter cooperative group-led study of financial hardship in metastatic colorectal cancer patients. J Natl Cancer Inst. 2022;114(3):372–380. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Chino F, Peppercorn JM, Rushing C, et al. Going for broke: a longitudinal study of patient-reported financial sacrifice in cancer care. J Oncol Pract. 2018;14(9):e533–e546. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Friedes C, Hazell SZ, Fu W, et al. Longitudinal trends of financial toxicity in patients with lung cancer: a prospective cohort study. J Clin Oncol Pract. 2021;17(8):e1094–e1109. [DOI] [PubMed] [Google Scholar]
- 11. Juster FT, Suzman R. An overview of the health and retirement study. J Hum Resource. 1995;30(suppl):S7–S56. [Google Scholar]
- 12. Kelley AS, Langa KM, Smith AK, et al. Leveraging the health and retirement study to advance palliative care research. J Palliat Med. 2014;17(5):506–511. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Health and Retirement Study. Sample Sizes and Response Rates. https://hrs.isr.umich.edu/sites/default/files/biblio/ResponseRates_2017.pdf. Accessed April 2017.
- 14. Rubin DB. Bias reduction using Mahalanobis-metric matching. Biometrics. 1980;36(2):293–298. [Google Scholar]
- 15. Clarke D, Schythe KT, Implementing the Panel Event Study. Discussion Paper Series. Bonn, Germany: IZA - Institute of Labor Economics; 2020. [Google Scholar]
- 16. Anderson ML. Multiple inference and gender differences in the effects of early intervention: a reevaluation of the abecedarian, Perry preschool, and early training projects. J Am Stat Assoc. 2008;103(484):1481–1495. [Google Scholar]
- 17. Freed M, Biniek JF, Damico A, et al. Medicare Advantage in 2021: Premiums, Cost Sharing, Out-of-Pocket Limits and Supplemental Benefits. San Francisco, CA: Kaiser Family Foundation; 2021. [Google Scholar]
- 18. Lam MB, Phelan J, Orav EJ, et al. Medicaid expansion and mortality among patients with breast, lung, and colorectal cancer. JAMA Netw Open. 2020;3(11):e2024366. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Moss HA, Wu J, Kaplan SJ, et al. The Affordable Care Act’s Medicaid expansion and impact along the cancer-care continuum: a systematic review. J Natl Cancer Inst. 2020;112(8):779–791. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lin L, Soni A, Sabik LM, et al. Early- and late-stage cancer diagnosis under 3 years of Medicaid expansion. Am J Prev Med. 2021;60(1):104–109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Barnes JM, Johnson KJ, Boakye EA, et al. Early Medicaid expansion and cancer mortality. J Natl Cancer Inst. 2021;113(12):1714–1722. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Han X, Jemal A, Zheng Z, et al. Changes in noninsurance and care unaffordability among cancer survivors following the affordable care act. J Natl Cancer Inst. 2020;112(7):688–697. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Kluender R, Mahoney N, Wong F, et al. Medical debt in the US, 2009-2020. JAMA. 2021;326(3):250–256. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Smith GL, Lopez-Olivo MA, Advani PG, et al. Financial burdens of cancer treatment: a systematic review of risk factors and outcomes. J Natl Compr Canc Netw. 2019;17(10):1184–1192. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.American Cancer Society. Cancer Treatment & Survivorship Facts & Figures 2019-2021. Atlanta, GA: American Cancer Society; 2019. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The raw/processed data required to reproduce the above findings cannot be shared under the data use agreement between University of Colorado and HRS Survey Research Center.