Statistical and machine learning analysis of non-clinical factors impacting emergency room delays

Abstract

IntroductionPlenty is known about the clinical factors impacting the time patients spend in emergency departments (EDs), but less is known about associated non-clinical factors. Here we fill this knowledge gap through a statistical and machine learning (ML) analysis of the non-clinical factors impacting ED length of stay (LoS). MethodsData from adult patients visiting the tertiary healthcare organization ED, Riyadh between January 2017 and October 2019 were subjected to logistic regression and multiple ML classification models to predict non-clinical factors associated with delayed ED LoS (>6 h). ResultsThere were 352,230 emergency visits by 135,185 patients. Most were under 6 h, with only 18% exceeding that benchmark. In logistic regression analysis, weekend visits, in the summer, by elderly male patients were associated with delayed ED LoS. The results of this study and the analysis of non-clinical factors in general can be used to aid in administrative decisions in EDs, such as deciding the number of staff per shift or provide the patients with the expected LoS to help them make decisions whether to stay or leave. ConclusionLogistic regression revealed some associations between ER LoS and the included factors. However, ML models showed disappointing performance with the included non-clinical predictive variables. Further research to examine the impact of both clinical and non-clinical parameters on ED delay is now needed.