Promotion of end-of-life discussions in terminally ill patients in Taiwan
The Taiwan National Health Insurance Administration, since 2012, has initiated an EOL discussions program that encourages healthcare providers to provide these discussions to terminally ill patients [9]. During the discussions, healthcare providers discuss patients’ goals and preferences concerning medical care and life-sustaining treatments towards the end of their lives [9]. Each EOL discussion with a patient awards the healthcare provider with the equivalent of US$75 [9]. There should be a record of all discussion content in the patient’s medical chart and each session must be at least an hour long [9].
Study population
This nationwide cohort study used the Taiwan National Health Insurance Research Database (NHIRD), which contains healthcare data from more than 99% of the population in Taiwan [10]. It is a large-scale database derived from the national health insurance system, which consists of registration files and original claims data. The database de-identified and scrambled the patients’ identification codes before releasing the data for research purposes [11].
This cohort study selected subjects aged 18 years or older who had received a cancer diagnosis and died between January 1, 2012 and December 31, 2018. We identified these patients from the Registry for Catastrophic Illness. In Taiwan, the Registry for Catastrophic Illness requires a peer review of pathohistological reports before a cancer diagnosis is reported [12]. This registry was also linked to Taiwan’s death certificate database to confirm the demise of cancer patients [13]. The Institutional Review Board of TCH (no. TCHIRB-10709107-W) approved this study and we performed all methods in accordance with TCH IRB guidelines and regulations.
Outcome variable
The outcome was life-sustaining treatments during the last three months of life in cancer patients. Life-sustaining treatments included cardiopulmonary resuscitation, intubation, and defibrillation [14].
Main explanatory variable
The main explanatory variable was EOL discussions with physicians, which was determined by patients’ medical records.
Controlling variables
The controlling variables included sociodemographic characteristics and comorbidities. Sociodemographic factors included age, sex, urbanization, and income level. Urbanization described whether subjects resided in urban, suburban, or rural areas [15]. We calculated the income level from the average monthly income of the insured person and grouped it into three categories: low (≤ 19,200 New Taiwan Dollars [NTD]), intermediate (19,201 NTD to < 40,000 NTD), and high (≥ 40,000 NTD). We defined the comorbidities according to the International Classification of Diseases, Ninth and Tenth Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) code. Subjects’ comorbidities include diabetes, chronic kidney disease, congestive heart failure, coronary heart disease, liver cirrhosis, chronic obstructive pulmonary disease, dementia, and cerebrovascular disease (Supplementary table 1). A person was considered to have a comorbidity only if the condition occurred in an inpatient setting or in three or more outpatient visits [16].
Statistical analysis
First, we analyzed the subjects’ demographic data. We then analyzed categorical data using the Pearson χ2 test where appropriate. We presented continuous data as mean ± standard deviation (SD), and conducted a two-sample t-test to compare outcomes between patients who underwent EOL discussions and those who did not.
We assessed the crude associations of EOL discussions and other covariates with the outcome (utilization of life-sustaining treatments during the last three months of life) by computing the odds ratios (ORs) and corresponding 95% confidence intervals (CIs). We then performed a multivariate logistic regression to estimate the association between EOL discussions and the utilization of life-sustaining treatments after adjusting for potential confounders. A variable with p<0.05 was defined as a significant factor associated with the utilization of life-sustaining treatments in the multivariate analysis. Adjusted odds ratios (AOR) with 95% confidence intervals (CI) indicated the strength and direction of these associations.
We conducted subgroup and sensitivity analyses to examine the robustness of the main findings, after stratifying participants by age and sex. We performed all data management and analyses using the SAS® v9.4 statistical software package (SAS Institute, Cary, NC, USA).