This is a retrospective cohort study using a prospectively-collected database of patients admitted to the PCU at Kouseiren Takaoka Hospital, Toyama prefecture in Japan. Data were obtained from the electronic medical records. The hospital was an acute hospital with 533 beds, and the PCU has 16-beds and provides an active home support services in addition to end-of-life care [16, 17]. We chose to examine this study at our institution and with all eligible cases during the time period covered, rather than selecting a sample size in terms of power. This study was conducted with the approval of the Clinical Research Ethics Review Committee of the Kouseiren Takaoka Hospital (Approval No.: #20,190,829,003). We obtained informed consent from all participants.

All consecutive patients admitted to the PCU between October 2016 and March 2020 were eligible for this study. No case exclusion criteria were established for this study.

On the basis of literature reviews [12,13,14,15, 18,19,20,21], variables potentially associated with survival and place of death were extracted from the medical records: patient age, sex, primary tumor sites, length of hospital stay, presence or absence of metastases, Palliative Prognostic Index (PPI) [22], symptoms, vital signs (i.e., systolic blood pressure, pulse rate, and SpO2), opioid dose (oral morphine equivalent), marital status, the number of co-habiting family members (including patient), presence or absence of a daytime caregiver, whether the primary caregiver was a spouse, family members’ preferred place of care, and family members’ preferred location of death. Further, calorie intake on the first day and presence/absence of delirium within three days of admission were recorded. Primary tumor sites were classified into hepatobiliary pancreatic cancer, respiratory cancer, gastrointestinal cancer, head and neck cancer, urologic cancer, skin cancer, gynecologic cancer, and others. The symptoms were classified as pain, fatigue, dyspnea, disturbance of consciousness, nausea and vomiting, anorexia, abdominal distention, and others.

Patient survival was defined as the periods from the day of admission to the PCU to death. Each patient was followed up to seven months. Place of death was also recorded.

Patients who were discharged to home from the PCUs and were treated at home for at least one day, were grouped into the discharge-to-home group, and those who were treated in the PCU from admission until death were grouped into the PCU care group.

For comparisons of survival, propensity score matching was estimated using a logistic regression model adjusted for age, sex, PPI, and cancer type. Propensity score matching was implemented using a nearest neighbor matching approach without replacement, with a caliper of 0.04 for optimal precision. Standardized differences were employed as a metric to assess the balance achieved through the matching process. Kaplan‒Meier curves and log-rank tests were utilized to compare survival between the PCU care group and the discharge-to-home group. To evaluate the Kaplan‒Meier curves after propensity score matching, curves were drawn for patients before propensity score matching as a sensitivity analysis. To analyze the factors affecting survival, a Cox proportional hazards model was used, and hazard ratios (HRs) and 95% CIs were calculated.

The primary outcome measure for this study is whether there is a significant difference in survival after propensity score matching. The following secondary outcome measures are also to be evaluated. To identify the factors associated with death at home after discharge, patient backgrounds were compared between the patients who were discharged to home and eventually died at home and those who were discharged to home but eventually died at the hospital (PCU). Comparisons were performed using Student’s t-test or χ2 test wherever appropriate. Multiple logistic regression analysis was performed as a form of multivariate analysis to investigate factors affecting the place of death remaining in the final model, and HRs and 95% CIs were calculated.

The significance level was set at 5%, and all analyses were conducted using IBM SPSS Statistics ver. 27 (IBM Corporation).