ARTIFICIAL INTELLIGENCE AS A PREDICTIVE TOOL TO IDENTIFY PATIENTS WITH PULMONARY EMBOLISM AT HIGH RISK FOR 30-DAY READMISSION

Abstract

TOPIC: Education, Research, and Quality Improvement TYPE: Original Investigations PURPOSE: Pulmonary Embolism (PE) is the third leading cause of cardiovascular death in the United States [1]. Venous thromboembolism (VTE) has an annual incidence of 1-2 people per 1000. The 1-year mortality can be as high as 21.6% [2]. The greatest risk factor for VTE, is a prior episode of VTE. Bikdeli et al (2019), found that though hospitalization rates for PE increased from the years 1999 to 2015, the 30-day readmission rates decreased.30-day hospital readmission rates have been used as a metric to evaluate the quality of care provided by hospitals and to determine reimbursement[5]. Risk prediction models have been developed to help identify patients at high risk of readmission with goals to reduce hospital readmissions and healthcare costs. Results with these various models have varied. Machine learning models using artificial intelligence (AI) have been developed to predict hospital readmissions. A commercially available AI/ machine learning tool was deployed through the electronic health record to predict risk of readmission. The aim of this study was to assess the accuracy of artificial intelligence-based tool in predicting which patients are at high-risk for readmission at 30 days by comparing it with a validated prediction tool, the LACE index for patients admitted with PE. METHODS: Data was collected retrospectively on patients 18 years or older with a primary diagnosis of PE and discharged from a tertiary care center from 11/26/2018 to 02/01/2020. We evaluated the effectiveness and accuracy of a proprietary AI product, to identify patients at high risk for hospital readmission and compared the effectiveness with the LACE Index. The statistical method for comparison was logistic regression with LACE score as a single predictor. Sensitivities were compared after choosing a threshold that most closely matched the proprietary AI product specificity. The primary evaluated outcome was readmissions for PE within 30-days from the index hospitalization. RESULTS: Logistic regression with LACE Index score as a single predictor was used. Sensitivities were compared after choosing a threshold which resulted in the specificity most closely matching AI product specificity. At specificity of 96.3% the AI product model had sensitivity of 19.4%. The LACE model had either 2.8 point higher (22.2%) or 5.5 point lower (13.9%) sensitivity depending on the chosen specificities of 94.7% and 97.9% respectively. We conclude that the LACE model performs roughly on par with proprietary AI tool in detecting all-cause readmission for PE. CONCLUSIONS: The LACE model performs roughly on par with proprietary AI tool in detecting 30-day readmission for PE. Proprietary AI tools may be helpful for predicting readmission rates for other diseases, in the case of PE, it did not provide a greater advantage. CLINICAL IMPLICATIONS: The LACE model performs roughly on par with proprietary AI tool in detecting all-cause readmission for PE. DISCLOSURES: No relevant relationships by Asad Cheema, source=Web Response No relevant relationships by radhika deshpande, source=Web Response No relevant relationships by Michael Gleason, source=Web Response No relevant relationships by Daniel Holtz, source=Web Response No relevant relationships by Cameron Koester, source=Web Response No relevant relationships by Abhishek Kalidas Kulkarni, source=Web Response No relevant relationships by Zurain Niaz, source=Web Response No relevant relationships by vivek prakash, source=Web Response No relevant relationships by Prashanth Singanallur, source=Web Response