Setting:
We conducted this study in Singapore, a high-income Asian economy. Cancer is the leading cause of death in Singapore25. Approximately 30% of health expenditure in Singapore is out-of-pocket, greater than the average for high-income countries26. A mandatory low-cost public health insurance with high deductibles (Medishield Life) partially meets patients’ healthcare cost. Individuals can top-up this insurance coverage by purchasing a private health insurance27.
Study Design and Participants:
We used data from COMPASS (Cost of Medical Care of Patients with Advanced Serious Illness in Singapore), a cohort of 600 patients diagnosed with a metastatic cancer. Following written consent, the study recruited patients (between July 2016 and March 2018) who met inclusion criteria for having a diagnosis of a solid metastatic cancer and, being 21 years or above, from two major public hospitals.
Participants were surveyed at baseline and every three months until they died and their billing data (starting January 2015) was obtained from hospital records. The study was approved by the Institutional Review Board at SingHealth. Details of the study protocol are published28.
Study variables:
Outcomes:
Low and high intensity admissions –We calculated the cost per day for each admission (total gross cost/number of days in hospital) and the median cost per day, separately for admissions in private and subsidised wards. Admissions incurring lower than median cost per day in each type of ward were classified as low-intensity admissions.
Total inpatient cost –We calculated total cost incurred in all inpatient admissions (gross cost before tax and subsidy), and separately for low- and high-intensity admissions.The cost was adjusted for inflation to 2019 Singapore dollars (SGD) using the consumer price index from the Department of Statistics, Singapore29.
Covariates:
Age – of patient during the baseline survey.
Symptom burden – At each survey, we assessed symptom burden using a composite measure of 10 items (shortness of breath, constipation, losing weight, vomiting, swelling, pain, dryness in mouth/throat, lack of energy, nausea, other symptoms); each item rated 0 (not at all) to 4 (very much). Total score calculated as sum of all items ranged from 0 to 40; a higher score indicated higher symptom burden.
Preference for life extension - We asked patients (at each survey) “If you had to make a choice now, would you prefer treatment that extends life as much as possible, or would you want treatment that cost you less?” Patients responded on a scale of 1 to 9, with 1 representing ‘extend life as much as possible at high cost’ and 9 representing ‘no life extension at less cost’. We reverse coded it so that higher values indicate higher preference for life extension.
Prognostic belief - At each survey, patients were asked if they thought that the current treatment that they are taking would cure them (1 = inaccurate; 0 = correct/unsure).
Cancer site- was categorized as breast, gynecologic/genitourinary, gastrointestinal, respiratory, and other.
Analysis:
Using a sample of 439 deceased patients, we examined inpatient cost in the last year of life. To do this, we calculated the number of days between the date of each admission and death, and then the cost of all inpatient admissions within specified time intervals before death. If there were no admissions during the time interval, costs were considered to be zero. Hence, the number of patients in each interval is the same. We decomposed the total inpatient cost (for all admissions, low- and high-intensity admissions) into cost incurred for - ward, investigations, surgery, maintenance, ICU and related units, procedures, consumables, drugs and others (Appendix 1). We examined the proportion of cost incurred on each inpatient service at 6–12, 0–6, 0–2 and 1 month before death. We plotted the mean total inpatient cost, total number of admissions and total number of days in hospital and the mean cost per day (total inpatient cost/total number of days in hospital) for each monthly interval over the last year of life, for all admissions, and low- and high-intensity admissions separately.
To assess patient characteristics associated with higher inpatient spending (Aim 2), our dependent variables were total inpatient cost, inpatient cost for low- and high-intensity admissions, incurred by patients between two survey rounds n and n + 1 (conducted approximately 90 days apart). If for any patient, n + 1 survey was missing and an admission was more than 90 days apart from round n, the admission was dropped from the analysis. Independent variables included patients’ symptom burden, age, preference for life extension and prognostic belief in round n. The regression controlled for cancer site. As many patients incurred zero inpatient cost between two survey rounds, we ran two-part models. The first part of the two-part model used a logit function to identify patient characteristics associated with having non-zero inpatient cost. Conditional on cost being non-zero, the second part used a generalized linear model with log link and gamma distribution to identify patient characteristics associated with higher cost. For patients who survived less than 90 days after survey n, the total cost in the next 90 days might be artificially lower, which might bias our results. We therefore conducted a sensitivity analysis in which we dropped patient-time intervals where patient died within 30 days of survey n and estimated two-part models using average monthly cost between survey n and n + 1 or survey n and death date.