Study design and participants
We performed a post-hoc analysis of a multicenter, prospective, observational cohort study known as the East-Asian collaborative cross-cultural Study to Elucidate the Dying process (EASED), which addresses the dying process and end-of-life care in patients with terminal cancer admitted to PCUs in Japan.20 The EASED enrolled and followed up participants from 23 palliative care institutions across Japan between January 1, 2017 and June 30, 2018. Consecutive eligible patients were enrolled if they had been newly referred to the participating PCU during the study period. All interventions and observations were carried out within routine clinical practice. The inclusion criteria for the present study were as follows: (1) age ≥18 years), (2) diagnosis of locally extensive or metastatic cancer (including hematological neoplasms), and (3) admission to the PCU. Patients who were scheduled for discharge within 1 week or those who did not want to be enrolled, were excluded.
Using the EASED data, we aimed to assess the impact of paracentesis on prognosis in patients with MA. The participants in this study were patients with ascites on admission to PCUs. Patients without symptoms caused by ascites on admission were not excluded, as they may have become symptomatic and undergone paracentesis during hospitalization.
The EASED study was performed in accordance with the ethical standards of the Helsinki Declaration and the ethical guidelines for epidemiological research presented by the Ministry of Health, Labour and Welfare in Japan. The study protocol was reviewed and approved by the local institutional review boards of all participating institutions. Written consent was waived in accordance with local regulations.
In this study, we defined paracentesis as the removal of ascitic fluid from the abdominal cavity via a temporarily inserted needle or catheter to relieve abdominal pressure and alleviate ascites-related symptoms.1,10,21 This definition did not include diagnostic paracentesis. As part of routine clinical practice, paracentesis was performed when clinically indicated (i.e., mainly based on the patient’s symptoms). The volume of paracentesis and whether additional artificial hydration or albumin infusion was performed was decided by the primary palliative care physician.
We collected data regarding patient’s age, sex, primary tumor site, and the presence/absence of metastatic lesions at admission. Laboratory data (i.e., albumin, total bilirubin, creatinine, and C-reactive protein) had been recorded because many patients underwent routine blood tests at admission. Furthermore, we recorded the Karnofsky Performance Status (KPS), co-treatments (i.e., hydration volume and the use of diuretics, corticosteroids, opioids, and albumin infusion), and ascites features (i.e., gross appearance, volume of paracentesis, history of paracentesis before admission) on the day of the first paracentesis in the PCU. In addition, we used a numerical rating scale (NRS) to assess abdominal distension prior to paracentesis and on the following day. Adverse events that may have been caused by paracentesis were recorded in accordance with the Common Terminology Criteria for Adverse Events version 4.0. On the day of death, the date of death and number of paracentesis procedures performed during the PCU stay were recorded. All measurements were performed by patients’ primary palliative care physicians during daily clinical practice using a structured data-collecting sheet designed for the study. Patients who were discharged were followed up for 6 months from baseline.
First, the characteristics of patients with MA were described, and their survival was compared with that of patients without MA using the log-rank test.
Then, we constructed a propensity score (PS) model (the conditional probability of undergoing paracentesis) by selecting a set of confounders between treatment assignment (undergoing paracentesis) and outcome (survival from admission to death), based on a backdoor criterion using a directed acyclic diagram (DAG) that draws the causal network linking receiving paracentesis, survival, and other variables.22–25 The DAG included both measured variables at admission (i.e., patient characteristics [age, sex, KPS, primary tumor site, and liver metastasis], complications [malignant bowel obstruction], laboratory data [albumin, C-reactive protein, total bilirubin, and creatinine], presence of symptoms owing to ascites, and opioid consumption) and an unmeasured variable (i.e., ascites volume). Variables selected as confounders (see Figure, Supplemental Digital Content 1, which demonstrates these variables). Albumin and C-reactive protein were used as markers of cachexia, while total bilirubin and creatinine were used as markers of hepatic and renal failure, respectively.
Next, under the missing at random assumption, we performed multiple imputations by chain equations to impute missing values for KPS (0.1%), albumin (12.7%), total bilirubin (13.6%), creatinine (11.4%), and C-reactive protein (13.6%).27 The variables included in the imputation models were the same as those in the PS model. In total, ten complete datasets were generated for subsequent analyses.
We compared baseline characteristics between patients who did not undergo paracentesis (non-paracentesis group) and those who did (paracentesis group). The balance in covariates was assessed using the standardized mean difference (SMD). SMD >0.1 was interpreted as a meaningful difference.
To account for selection and confounding biases, the observed differences in baseline covariates between the two groups were adjusted using the inverse probability of treatment weighting (IPTW) method.23,24 In this method, the PS for each patient was estimated using multivariate logistic regression of the PS model. Subsequently, the PS values from the ten imputed datasets were pooled according to Rubin’s rule.27 Finally, scores in the non-paracentesis group were weighted by the average treatment effect for treated weight: [PS/(1-PS)]. This method produces a weighted pseudo-sample of patients in the reference group with the same distribution of measured covariates as in the exposed group.
Survival was calculated from the day of admission to the day of death. Adjusted Kaplan–Meier curves were computed based on inverse probability weights, and a univariate inverse probability-weighted Cox proportional hazards model was used to estimate the IPTW-adjusted hazard ratio (HR) for patient survival in the paracentesis and non-paracentesis groups. Furthermore, subgroup analyses were performed to investigate the IPTW-adjusted HR of the paracentesis and non-paracentesis groups according to some of the baseline covariates, including age, sex, KPS, primary tumor site, liver dysfunction (defined as total bilirubin higher than 4.0 mg/dL at admission), and renal dysfunction (defined as creatinine higher than 1.5 mg/dL at admission).
Finally, two sensitivity analyses were conducted to assess the robustness of the results. First, 265 patients without symptoms of ascites at admission were excluded to determine whether results varied according to patient selection. Second, patients who were discharged alive were excluded to determine whether results were affected by the censored population.
All statistical analyses were performed using R version 3.5.3 (R Core Team 2019, Vienna, Austria). All p-values were two-sided, and p-values <0.05 were considered significant. Imputation of missing data was conducted using the ‘‘MICE’’ package, and survival analyses were performed using the ‘‘survival’’ package.