SAT0557 OUTPATIENT READMISSION IN RHEUMATOLOGY: A MACHINE LEARNING PREDICTIVE MODEL OF PATIENT'S RETURN TO THE CLINIC. (June 2019)

• Record Type: Journal Article
• Authors: Rodriguez, Luis Rodriguez; García, Alfredo Madrid; Font-Urgelles, Judit; Freites, Dalifer; Lajas, Cristina; Pato, Esperanza; Angel Jover, J; Fernandez, Benjamin; Abasolo, Lydia
• Is Part Of: Annals of the rheumatic diseases. Volume 78(2019)Supplement 2
• Page Start: 1371
• Page End: 1371
• Publication Date: 2019-06
• Subjects: Rheumatism -- Periodicals; 616.723005
• DOI: 10.1136/annrheumdis-2019-eular.7125
• Languages: English
• ISSNs: 0003-4967

Abstract

Abstract : Background: Readmissions can be defined as the return of a patient to a healthcare setting after a discharged. Attention has been mainly focused on readmissions following inpatient hospitalizations. In the outpatient setting, readmissions have been far less studied. Premature outpatient discharges can have negative impacts at multiple levels, as they may prolong disability, and increase the chance of disease chronification, the demands to the immediate patient's support system, and healthcare resources utilization. As the first step in preventing outpatient readmission, the assessment of the individual patient's risk could be useful to help identify those subjects at greatest risk, so, in a further step we could focus the delivery of an intervention in those patients to reduce their risk. Objectives: To develop and validate a machine learning predictive model based on Random Forest, to estimate the risk of readmission in an outpatient rheumatology clinic after discharge (outpatient readmission). Methods: Patients stored in a departmental electronic health record from April 1st, 2007 to November 30th, 2016, and followed-up until November 30th, 2017, were included in this study. Only readmissions taking place between 3 and 12 months after discharge were analyzed. Discharge episodes were split into training, validation and test datasets. Clinical and demographic variables, including diagnoses, treatments, quality of life, and comorbidities, were used as predictors. Models were developed using Random Forest in the training dataset, though the combination of several tuning parameters. Models that maximized the area under the receiver operating characteristic curve (ROC-AUC) in the validation set were assessed in the test set. The model with the highest AUC-ROC in the test dataset was considered as the best final model. Results: 17, 473 patients (18, 117 discharges episodes) were analyzed and 1, 960 (10.8%) discharges episodes were classified as outpatient readmissions. 48, 654 models were finally developed. The best final model showed an AUC-ROC of 0.674 a sensitivity of 0.330 and a specificity of 0.867. The most relevant variables in the model were the number of diagnoses given at discharge, follow-up duration, age, number of previous discharges, previous corticosteroids use and disability. Conclusion: We have developed a predictive model for outpatient readmission in a rheumatology setting. Clinical, demographical characteristics as well as medication and disability were the most important predictors. Disclosure of Interests: None declared