Predicting hospital admissions to reduce crowding in emergency department s of the the integral healthcare system for public use in Catalonia
Objective: This study analyzed data from Emergency Departments (EDs) from more than 60 different centers embedded in the Integral Healthcare System for Public Use in Catalonia (SISCAT) to predict hospital admissions based on information readily available at the moment of arrival to the ED. The predi...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/374381 |
| Acceso en línea: | https://hdl.handle.net/2117/374381 |
| Access Level: | acceso abierto |
| Palabra clave: | Hospitals -- Admission and discharge Hospitals -- Emergency services Hospitals -- Waiting lists Sanitat Salut Hospital Urgències Temps d'espera SISCAT Model predictiu Gradient boosting Admissions Sanidad Salud Urgencias Tiempo de espera Modelo predictivo Healthcare Emergency department Crowding Waiting time Predictive model Hospitals -- Ingressos i altes Hospitals -- Serveis d'urgències Hospitals -- Llistes d'espera Àrees temàtiques de la UPC::Economia i organització d'empreses |
| Sumario: | Objective: This study analyzed data from Emergency Departments (EDs) from more than 60 different centers embedded in the Integral Healthcare System for Public Use in Catalonia (SISCAT) to predict hospital admissions based on information readily available at the moment of arrival to the ED. The predictive models might help reduce overcrowding at EDs and improve the service delivered to patients. Method: A retrospective analysis was conducted using data from the SISCAT collected during the year 2018. Gradient boosting machine was used to train and test the predictive models in R, splitting the data in a 70/30 partition. Variable importance for each of the models was analyzed. Receiver Operating Characteristic (ROC) curves were created, and the Area Under the Curve (AUC) was obtained from each of them as a measure of predictive performance. The first part of the study targeted the obtention of models with high accuracy and AUC, while the second part targeted the obtention of models with a sensitivity > 0.975 and analyzed the possible benefits that could come from the application of such models. Results: From the 3,189,204 ED visits included in the study, 11.02% ended in admission to the hospital. Gradient boosting machine proved to be a good method to predict for a binary outcome of either admission or discharge. The best performance for all the models was obtained at a 0.5 probability of admission threshold. The largest AUC was obtained for the complete dataset and yielded a result of 0.8938 with a 95% CI of 0.8929-0.8948. The best results for the sensitivity tests were obtained with the adults’ dataset, with a model that gave a 0.4344 specificity and 0.5033 accuracy for a 0.975 sensitivity level. Conclusion: This study reaffirms on the belief that gradient boosting machine is a powerful tool to use in binary outcome predictive models. It shows that data collected at the moment of arrival to the ED can be used to predict hospital admissions accurately, and that a model including data from a comprehensive hospital network has a better predictive performance when compared to a similar model developed with data from one unique health center only. It discusses the huge potential that the application of the models obtained could have in fighting crowding in EDs by allowing for an early start of the bed allocation process, making it possible to do all the required procedures for admission simultaneously to the patient being visited by the doctor, instead of doing it in a sequential manner after the visit, which unnecessarily crowds ED rooms and generates a nonoptimal use of the available resources in EDs. The study also suggests the application of this predictive technique to develop models with proven high sensitivity to digitalize the patient-hospital relationship to allow for a first contact between both parties before the visit to the ED, which can potentially regulate the inflow of patients in this department and reduce ED overcrowding significantly |
|---|