Descriptive statistics
Of the 483 eligible patients, 389 (81%) and 94 (19%) had breast cancer and NPC, respectively. Detailed sample characteristics and their relevant statistical tests are presented in Table 1. Statistical tests indicate significant differences between survey participants (n = 76) and non-participants (n = 407, either uncontacted or contacted but did not participate in the survey) for the cancer site, gender, and race (P = 0.000). These differences are expected and directly attributed to our intention to sample similar proportions of breast cancer and NPC patients. Patients who had been involved in a clinical trial were more likely to participate in our survey (P < 0.01). We also compared survey participants whose medical bill data were extracted for further analysis (n = 41) to those who did not provide consent for us to do so (n = 35 either refused or no response). There were no statistically significant differences between these two groups except that there were more patients from the Malay ethnicity in the former group (P < 0.01, results not shown).
Characteristics of our 76 survey participants are depicted in Table 2. Most respondents are married (76%), attained secondary and below education (63%), and live in a public Housing Development Board (HDB) flat (90%). For medical expenses, around 10% of the respondents qualify for MediFund, which provides further financial assistance based on means-testing. Another 10% received PG and CSC subsidies. About one-fifth of the respondents reported using their family’s Medisave to pay for treatment.
Direct non-medical expenses include transport cost, use of Complementary and Alternative Medicine (CAM), and hiring a formal caregiver. Most respondents took public transport (66%), followed by taxi/private hire (25%) and private car (25%), some respondents selected more than one mode of transport in the questionnaire. The use of complementary and alternative medicine (CAM) was not prevalent. For the 31 (41%) respondents who used CAM, 11 (31%) of them stopped due to the cancer treatment. The reasons include CAM being not suitable, not effective, finances and others. Only six (8%) respondents hired a formal caregiver after their cancer diagnosis.
Respondents making adjustments to their living arrangement as a result of their cancer diagnosis are in the minority (18, or 24%). They cited financial reasons or to improve access to care. Some respondents have rented out rooms or moved to a smaller flat to raise funds, and some have moved in with family or friends to receive care. Seven (9%) respondents indicated that they or their household member(s) took on loans, including borrowing money from relatives and friends, to finance the cost of cancer treatment. Four borrowed less than $10,000, one borrowed $20,000 to $30,000, and one borrowed $50,000 and above.
The mean COST and FACT-G scores among the survey participants were 18.0 (out of 44) and 68.3 (out of 108), respectively. Higher scores indicate better feelings of financial wellbeing and better QOL respectively. The mean scores, as well as distribution of COST scores, were not significantly different between the breast cancer and NPC cancer patients (Table 3). For our sample, the two scores are positively and moderately correlated (r = 0.45).
Cross-tabulations
To gain a deeper insight into employment status and income for both the patient and household, we cross-tabulated the survey results for these two variables. Indirect costs of cancer include lost economic productivity for the patients and their household members. Only a quarter of the respondents who earned less than $1,000 per month before cancer diagnosis reported being affected by changes in employment status or income compared to respondents (Table 4A). One reason is that these respondents were typically homemakers, retirees or have been out of the workforce for a significant period. In contrast, about 70% of the respondents who had higher earnings experienced an adverse impact on employment and income.
Around 60% of the respondents reported no change in the employment status of their household members (Table 4B). Accordingly, most of these respondents did not experience a fall in household income. For the affected households, most had employed household members taking on extra work or unemployed members seeking employment to supplement the income, while a minority had employed household members seeking alternative work arrangements (such as taking unpaid leave from work, working fewer hours, resigning from job, and/or early retirement) to devote more time towards caregiving. The reduction in household income ranged from $250 to $4,000 per month.
Medical Bill
All 41 patients, for whom medical bill data were retrieved, incurred out-of-pocket expenses. Table 5 reports the average and median OOP expenses, alongside standard deviation (SD) and interquartile range (IQR), for inpatient and outpatient settings as well as both combined. There was no significant difference between the total average OOP expense between breast cancer and NPC patients (average $26,818 vs $24,206 respectively). The outpatient component ranged between 26 – 34% of the total costs. However, the average total OOP expense was skewed by five breast cancer patients with extremely high OOP (Supplementary Figure 1). Comparing the median total OOP costs, there was a significant difference between breast cancer and NPC ($15,910 vs $21,593, P=0.008).
Regression Results
Financial Toxicity
For the first set of regression analysis, COST score is the dependent variable and the sample consists of 76 survey participants without medical bill data. The results are reported in Supplementary Table B. Specification 1 contains all the variables listed in Equation 1, except that due to the low utilisation of CAM and the possibility of recall bias for the transport mode used for hospital visit, these two variables were not included. Based on the adjusted R-squared, its explanatory power is relatively lower compared to Specification 2, which was obtained after systematically dropping the variables with coefficients that were small in magnitude or highly insignificant.
According to the parsimonious specification, patients who live in HDB flats were still found to have significantly lower scores (-8.3, P = 0.02), with education level being marginally significant (-4.4, P = 0.06). Patients who hired formal caregiver due to the cancer diagnosis, as well as those whose household member(s) needed to earn extra income to finance medical expenses were found to be associated with significantly lower scores (-7 and -6 respectively, P = 0.05). On the contrary, patients requiring inpatient admission had significantly higher scores (4, P = 0.05), symbolising increased feelings of financial well-being.
The second set of regression analysis also has COST score as the dependent variable, but includes the total OOP expense as an additional variable, for which data is limited to a subset of 36 survey participants excluding five outliers (Supplementary Table C). We did not find any meaningful correlation between OOP expense and FT for both the ordinary least squares (OLS) and two-stage least squares (2SLS) specifications.
Quality of Life
Results of the pre-planned regression analysis with FACT-G score as the dependent variable are reported in Supplementary Table D. We found that the feeling of financial well-being is positively and significantly correlated with a patient’s quality of life, controlling for other factors. Particularly, a one-point increase in COST score is associated with a 0.7 increase in FACT-G score (P = 0.002). A loss of quality of life was found to be significantly associated with certain adverse changes that his/her household member(s) experienced, such as when household member(s) had to work more to supplement household income (-15.2, P = 0.02) or when a household member’s Medisave account is being used to pay for medical expenses (-0.3, P = 0.04). Patients of Malay and Indian ethnicity are more likely to report higher quality of life.