A 2-week prognostic prediction model for terminal cancer patients in a palliative care unit at a Japanese general hospital.

Abstract

We aimed to develop a prognostic prediction model for 2-week survival among patients with terminal cancer in a palliative care unit (PCU). A prospective cohort study was conducted on terminal cancer patients in the PCU for 11 months at a general hospital in Tokyo, Japan. We collected data regarding demographics, treatment history, performance status, symptoms, and laboratory results. Patients who survived more than 2 weeks were labeled 'long survivors' and those who died within 2 weeks were grouped as 'short survivors'. Stepwise logistic regression model was constructed for the model development and bootstrapping was used for the internal model validation. In 158 subjects whose data were available for the analysis, 109 (69%) subjects were categorized as long survivors and 49 (31%) subjects as short survivors. A prognostic prediction model with a total score of 8 points was constructed as follows: 2 points each for anorexia, dyspnea, and edema; 1 point each for blood urea nitrogen >25 mg/dl and platelets <260,000/mm(3). Area under the receiver operating characteristic (ROC) curve of this model was 83.2% (95% CI: 75.3-91.0%). Bootstrapped validation beta coefficients of the predictors were similar to the original cohort beta coefficients. Our prognostic prediction model for estimating 14-day survival for patients with terminal cancer on the PCU ward included five clinical predictors that are readily available in the clinical setting and showed a relatively high accuracy. External validation is needed to confirm the model's generalizability.