Public health insurance and cancer‐specific mortality risk among patients with breast cancer: A prospective cohort study in China

Little is known about how health insurance policies, particularly in developing countries, influence breast cancer prognosis. Here, we examined the association between individual health insurance and breast cancer‐specific mortality in China. We included 7436 women diagnosed with invasive breast cancer between 2009 and 2016, at West China Hospital, Sichuan University. The health insurance plan of patient was classified as either urban or rural schemes and was also categorized as reimbursement rate (ie, the covered/total charge) below or above the median. Breast cancer‐specific mortality was the primary outcome. Using Cox proportional hazards models, we calculated hazard ratios (HRs) for cancer‐specific mortality, contrasting rates among patients with a rural insurance scheme or low reimbursement rate to that of those with an urban insurance scheme or high reimbursement rate, respectively. During a median follow‐up of 3.1 years, we identified 326 deaths due to breast cancer. Compared to patients covered by urban insurance schemes, patients covered by rural insurance schemes had a 29% increased cancer‐specific mortality (95% CI 0%‐65%) after adjusting for demographics, tumor characteristics and treatment modes. Reimbursement rate below the median was associated with a 42% increased rate of cancer‐specific mortality (95% CI 11%‐82%). Every 10% increase in the reimbursement rate is associated with a 7% (95% CI 2%‐12%) reduction in cancer‐specific mortality risk, particularly in patients covered by rural insurance schemes (26%, 95% CI 9%‐39%). Our findings suggest that underinsured patients face a higher risk of breast cancer‐specific mortality in developing countries.


| INTRODUCTION
Battling cancer is a crushing burden for all patients, but particularly so for those who are vulnerable to financial stress. It is common that cancer patients experience severe financial stress throughout their survivorship, 1 especially in developing countries where the health system is not ready to ease the burden for everyone. Cancer patients have higher out-of-pocket costs and may be absent from work for quite a while, which further lowers the ability to pay for medical care. 1,2 As an avalanche of "financial toxicity"-the damaging economic side effects of illness, cancer patients are at tremendous risk for debt, bankruptcy and impaired psychological wellbeing. 3,4 It is well-documented that social inequality in health contributes to the disparities in cancer survivorship in both developed and developing countries, including China. [5][6][7][8] The presence of a public health insurance system seems essential for a country to achieve universal healthcare coverage and health equity. 9 Improved health insurance coverage can reduce sociodemographic disparities in cancer care, including breast cancer, through early diagnosis and optimal treatment. 10,11 Fewer studies have paid attention to the impact of health insurance on cancer prognosis. So far, four US studies, [11][12][13][14] support the hypothesis that underinsured patients have a worse breast cancer prognosis, but two other studies from Australia and Brazil reported no clear differences related to the level of health insurance. 15,16 However, it is largely unclear whether the health insurance policies particularly in developing countries, where the patients may face higher financial toxicity, influence breast cancer prognosis.
Moreover, all reports have focused on insurance status or types, while no studies have addressed the out-of-pocket cost as an important barrier to cancer care. [11][12][13][14][15][16] It is, therefore, of critical importance to understand how different insurance plans, featured by varying reimbursement rates (ie, the covered/total charge) within specific insurance type, further contribute to the disparities in diagnosis, treatment and prognosis of breast cancer. The health insurance system was first introduced to China mainland China in 1980s, and drastically expanded to over 95.7% of the total population over the past decades. 17,18 The majority of insurance system is staterun, while commercial insurance may be purchased as a complement. There Leveraging a prospective large-scale cohort of patients with invasive breast cancer in China diagnosed from 2009 to 2016, we aimed to examine the associations of health insurance types and reimbursement rates with the risks of breast cancer-specific mortality.  19 Thus, the benefit packages and financial support are fragmented and inequitable across the schemes. For example, compared to UEBMI and URBMI, NRCMS is limited to a lower reimbursement cap and covers a narrower spectrum of diseases. The mean reimbursement rate for NRCMS is mainly 50% to 65%, which is much lower than UEBMI with a rate of 85% to 95%. 20,21 The disparities between urban and rural health insurances are thus considerable.

| Study population
Therefore, China has been establishing a consolidated health insurance scheme by 2020. For example, the fund pooling and management of NRCMS (from county level) and UEBMI and URBMI (from municipal level) should be moved to provincial and then country levels. The reimbursement rate is defined as the amount of medical expenses covered by insurance divided by the total expense. As the insurance is partly funded by local governments, the reimbursement rates may vary widely across counties, even under the same insurance scheme. Moreover, the rate is individual-based, affected and calculated by age, years of employment, hospital level and treatment modes.
The information on insurance types and reimbursement rates (for the primary treatment) is routinely documented in BCIMS. Specifically, the information on the type of insurance is provided by patients at the registration to BCIMS, while the rate of reimbursement is collected for the primary treatment during follow-up. Given the different administrations and insurance plans (Supporting Information Material), we classified insurance types into urban (ie, URBMI, UEBMI, and/or commercial insurances) and rural (ie, NRCMS) schemes, respectively. In the analysis of insurance type, 139 patients without any insurance and eight patients of unknown insurance status were excluded. Our data showed that the reimbursement rate was different among patients insured by urban or rural schemes ( Figure S1). We also classified patients by reimbursement rate below (0-69%) or above (70%-100%) the median. Patients without insurance were coded as 0 reimbursement rate. Then, 569 patients (192 insured by rural schemes) were excluded from this analysis due to unknown reimbursement rate.

| Breast cancer-specific and overall mortality
All patients were actively followed through telephone contact and medical visits until death or May 17, 2017, whichever came first. The underlying cause of death was ascertained from the medical records, whenever possible, or informed by the immediate family members.
We studied breast cancer-specific mortality as the primary outcome and overall mortality as the secondary outcome.

| Statistical analysis
First, we described the demographic and clinical characteristics among patients with different insurance types and reimbursement rates.
Demographic and clinical characteristics were obtained from BCIMS and classified as showed in Table 1. We examined the associations of health insurance type and reimbursement rate with different treatment modes, using logistic regression with adjustment for demographic and clinical characteristics. To account for correlations between treatment types, we additionally adjusted for other types of treatment.
We examined the associations of health insurance type and reimbursement rate with different treatment modes, using logistic regression with adjustment for demographic and clinical characteristics. To account for correlations between treatment types, we additionally adjusted for other types of treatment.
Next, we calculated and plotted the cumulative rates and 95% confidence intervals (CIs) of breast cancer-specific and overall mortality by insurance type and reimbursement rate up to 5 years after cancer diagnosis using a competing risk model. 21 Hazard ratios (HRs) and 95% CIs of breast cancer-specific and overall mortality were then estimated from Cox regression by contrasting patients insured by the rural scheme to patients insured by the urban scheme, as well as patients with low reimbursement rate to those with high. The proportional hazards assumption, tested based on Schoenfeld residuals, was not violated. To illustrate the joint effect of insurance type and reimbursement rate, we further examined the association of every 10% increase in reimbursement rate with mortality risks by insurance type.
In Model A, we adjusted for demographic factors, including age (as a continuous variable), calendar year at diagnosis, ethnic group, educational level (as a proxy for socioeconomic status, SES) and marital status. In Model B, we additionally adjusted for clinical characteristics (as potential mediators), including comorbidity, histological type, tumor stage, hormone receptor status (including both estrogen and progesterone receptors), HER2 status and Ki-67 level. In Model C, we additionally controlled for treatment modes, namely surgery, chemotherapy, radiotherapy, hormonal therapy and trastuzumab therapy.
Age was treated as continuous variables, whereas other covariates were categorized as showed in Table 1.
Because body mass index (BMI) would be neither the cause nor consequence of different insurances, we did not adjust for it in the primary analysis. We, however, noted that patients with different insurance were characterized by different BMI. We, therefore, performed an additional analysis by adjusting for BMI at diagnosis.
According to the recommendation for Asian populations, 22 we classified BMI into <23 kg/m 2 (nonoverweight) and ≥23 kg/m 2 (overweight). SES and accessibility to medical service are highly correlated with individual insurance plans. To further disentangle the potential influence of SES and accessibility to health-care, we performed a sensitivity analysis by clustering patients residing in the same community/county through the zip code of residence.
All analyses were performed in STATA statistical software (version 14; STATA, College Station, Texas). Value of P < .05 indicated statistical significance.

| Health insurance and breast cancer-specific mortality
During follow-up (median 3.1 years, interquartile range 1.4-5.1 years), 372 deaths were observed and 326 of them were due to breast cancer.
The cumulative rates of breast cancer-specific mortality were higher among patients insured within rural insurance schemes and with reimbursement rates ≤69%, compared to patients with urban insurance schemes and higher reimbursement rates, respectively ( Figure 1). Similar patterns were noticed for overall mortality.
When adjusting for demographic characteristics, patients insured by rural insurance schemes had a 46% increased risk of cancer-specific mortality (95% CI 14%-87%) compared to patients within urban insurance schemes ( Note: Patients with missing information on insurance type (n = 147, 1.98%) or reimbursement rate (n = 569, 7.65%) were not included for the corresponding analysis. Body mass index (BMI) was classified into <23 kg/m 2 (non-overweight) and ≥23 kg/m 2 (overweight). Tumor stage was categorized as localized (no nodal or metastatic disease), regional (nodal disease), or distant (any metastatic disease). Pearson's χ 2 statistic was used to assess significance of the difference between proportions in assessment of univariable associations. Abbreviation: HER2, human epidermal growth factor receptor 2.
T A B L E 2 Associations of insurance type and reimbursement rate with treatment type, by demographic and clinical characteristics   HRs were additionally adjusted for surgery (yes or no), chemotherapy (yes or no), radiotherapy (yes or no), hormonal therapy (yes or no) and trastuzumab therapy (yes or no). treatment modes, the association was attenuated somewhat yet remained significant (HR 1.29, 95% CI 1.00-1.65). Similarly, patients with low reimbursement rate had a 42% increased risk of cancer-specific mortality (95% CI 11%-82%) compared to patients within high reimbursement rate. Similar patterns were found for overall mortality (Table 3).
Largely similar results were yielded for both insurance type and reimbursement rate after additional control for BMI (Table S1).
Comparable but less significant associations were observed by conditioning on residential areas to further address SES and accessibility to care (Table S2).

| Joint effect of insurance type and reimbursement rate
We showed that every 10% increase in the reimbursement rate was associated with a 7% reduced risk of cancer-specific mortality (95% CI 2%-12% after full adjustment; Table 4). Particularly, every 10% increase of reimbursement rate in rural insurance schemes was associated with remarkable risk reduction of cancer-specific mortality (HR 0.74, 95% CI 0.61-0.91), compared to that in urban insurance schemes (HR 0.94, 95% CI 0.86-1.03, P for difference = .039). Similar results were found in overall mortality (Table 4).

| DISCUSSION
To the best of our knowledge, this is the first study to demonstrate that underinsured patients with invasive breast cancer are at increased risk of cancer-specific mortality in a country with less developed health insurance system. Importantly, our findings strongly suggest that a higher reimbursement rate, particularly in rural scheme insurance, is associated with a remarkable risk reduction of breast cancer-specific mortality. These associations are partly but not entirely explained by known prognostic indicators, including tumor characteristics and cancer treatment.
Findings from several studies in developed countries, mostly from the US, have shown that underinsured patients with breast cancer are more likely to suffer an increased risk of cancer-specific mortality, compared to those with adequate insurance. [10][11][12][13] Only one study from developing countries showed that breast cancer prognosis is comparable between patients insured by public and private health T A B L E 4 Association of every 10% insurance reimbursement rate increase with risks of cancer-specific or overall mortality insurances. 16 Our data further illustrated the impact of public health insurance status on breast cancer prognosis in developing countries independent of clinical factors. Most importantly, in addition to insurance status or type, we are the first to reveal that the low reimbursement rate is associated with an excess risk of breast cancer-specific mortality. In many developed countries, insurance plans usually come with a fixed coinsurance or reimbursement rate. Our setting therefore provides a unique opportunity to understand the potential mechanisms underlying the relationship between insurance and cancer prognosis, which highlights the urgent need of promoting reimbursement rate in rural insurance schemes, to significantly improve breast cancer prognosis and reduce health disparities at large.  26 It is not implausible that the lack of financial support may impact breast cancer prognosis through psychological stress.
Our findings may partly reflect the difference of SES across rural and urban regions as well as between individuals. Of note, SES is highly correlated with, and to some extent reflected by, health insurance status. As health insurance is likely underlying the causal pathway between SES and cancer prognosis, we did not consider it as a confounder in the studied association. However, we have adequately addressed educational attainment (as a proxy for SES) in all analyses. To further separate the influence of SES, we performed a sensitivity analysis by conditioning on 88 residence areas to better control for SES and accessibility to healthcare. Increased risks of cancer-specific and overall mortality are still suggested, although some are not significantly likely due to power issues. This largely refutes the possibility that our findings are completely explained by the differential socioeconomic status.
One major merit of our study is the large-scale prospective cohort design with virtually complete follow-up, largely limiting the common sources of bias. The rich information on demographic and clinical characteristics helped to disentangle the direct influence of health insurance on cancer-specific mortality, from the influence through tumor characteristics and treatment modes. Our study also has several limitations to consider. First, some deaths due to other causes may be misclassified as breast cancer-specific mortality. However, in our data, 297 out of 326 cancer-specific deaths (91.1%) entailed a clinically detected local recurrence or distant metastasis, which largely alleviates such concerns.
Moreover, the 5-year breast cancer-specific survival rate in our cohort is comparable to other Chinese cohorts 27,28 and cohorts from developed countries, 6,29 given a similar distribution of tumor stage. Furthermore, we have little information regarding extra insurances beyond the basic/public insurance. However, there were only 11 patients with commercial insurances included in our study and it is less likely to impact our results. As this cohort is based on a regional medical center, the findings may not be generalized to the entire population. The major selection forces include urban and well-educated residents, as well as advanced disease yet eligible for surgery and chemotherapy/radiotherapy due to referrals from other hospitals. We, however, observed similar associations across regions of residence, educational levels, and tumor stages (data not shown). We may also miss the patients that are most financially vulnerable, because of the nature of our study setting.
Reassuringly, we noted the strongest association in the youngest patients (aged 18-39 years), where we should have a smaller selection force because young patients with breast cancer were more likely to seek healthcare in a tertiary hospital.
In conclusion, our findings suggest that underinsured patients face a higher risk of breast cancer-specific mortality in China, which may provide fresh insights into the role of reimbursement rate in cancer health disparities in China and likewise developing countries.