Advancing screening tool for hospice needs and end-of-life decision-making process in the emergency department

Background: Predicting mortality in the emergency department (ED) is imperative to guide palliative care and end-of-life decisions. However, the clinical usefulness of utilizing the existing screening tools still leaves something to be desired. Methods: We advanced the screening tool with the A-qCPR (Age, qSOFA (quick sepsis-related organ failure assessment), cancer, Performance Status Scale, and DNR (Do-Not-Resuscitate) risk score model for predicting one-year mortality in the emergency department of Taipei City Hospital of Taiwan with the potential of hospice need and evaluated its performance compared with the existing screening model. We adopted a large retrospective cohort in conjunction with in-time (the trained and the holdout validation cohort) for the development of the A-qCPR model and out-of-time validation sample for external validation and model robustness to variation with the calendar year. Results: A total of 10,474 patients were enrolled in the training cohort and 33,182 patients for external validation. Significant risk scores included age (0.05 per year), qSOFA ≥ 2 (4), Cancer (5), Eastern Cooperative Oncology Group (ECOG) Performance Status score ≥ 2 (2), and DNR status (2). One-year mortality rates were 13.6% for low (score ≦ 3 points), 29.9% for medium (3 < Score ≦ 9 points), and 47.1% for high categories (Score > 9 points). The AUROC curve for the in-time validation sample was 0.76 (0.74–0.78). However, the corresponding figure was slightly shrunk to 0.69 (0.69–0.70) based on out-of-time validation. The accuracy with our newly developed A-qCPR model was better than those existing tools including 0.57 (0.56–0.57) by using SQ (surprise question), 0.54 (0.54–0.54) by using qSOFA, and 0.59 (0.59–0.59) by using ECOG performance status score. Applying the A-qCPR model to emergency departments since 2017 has led to a year-on-year increase in the proportion of patients or their families signing DNR documents, which had not been affected by the COVID-19 pandemic. Conclusions: The A-qCPR model is not only effective in predicting one-year mortality but also in identifying hospice needs. Advancing the screening tool that has been widely used for hospice in various scenarios is particularly helpful for facilitating the end-of-life decision-making process in the ED.