P3183 - Development and Validation of a Machine Learning<br>Risk-Prediction Model for In-Hospital Mortality from C. difficile: National Inpatient Sample Study
Introduction: Clostridioides difficile (Cdiff) infection carries increased mortality and morbidity in United States. Due to the scarcity of a prognostic scoring system of Cdiff, we aimed to develop a novel machine learning based risk score model to predict in-hospital mortality in Cdiff patients
Methods: The National Inpatient Sample (NIS) 2016-2020 database was used to identify all adult patients (18 years of age) with Cdiff, using ICD-10 code A04.7. Relevant clinical characteristics and outcomes were extracted. The primary endpoint was in-hospital all-cause mortality. The dataset was randomly split into 3 subsets: training, validation, and testing, with a ratio of 0,7, 0.1, and 0.2, respectively. The risk score was generated using the Autoscore package in R software, a machine learning based tool for automatic clinical score generation. The performance was evaluated using area under receiver-operative characteristics curve (AUC) with 95% confidence intervals (95%CI)
Results: Amongst 288,791 Cdiff patients identified [age 64.8318.77 years, female 164,611 (57%)], there were 17.602 (6.09%) primary outcome events. Out of the top 20 covariates seen in the Parsimony plot, we used eight for our risk score (0-100): age, all cause shock, comorbidity burden, ICU admission, acute respiratory failure, race, female sex, and acute kidney injury (Figure 1A). The AUC for the derivation and validation cohort was 0.8698 (0.861-0.8786) and 0.8644 (0.8582-0.8706), respectively (Figure 1B and 1C).
Discussion: We developed and internally validated a machine learning based risk score model to predict outcomes in CDiff patients, achieving a high AUC of 0.8644 (0.8582-0.8706). Risk score models guide in risk stratification and management and large future studies are needed to externally evaluate our results
Figure: AUC curves
Note: The table for this abstract can be viewed in the ePoster Gallery section of the ACG 2024 ePoster Site or in The American Journal of Gastroenterology's abstract supplement issue, both of which will be available starting October 27, 2024.
Disclosures:
Umesh Bhagat indicated no relevant financial relationships.
Ankit Agrawal indicated no relevant financial relationships.
Umesh Bhagat, MD, Ankit Agrawal, MD. P3183 - Development and Validation of a Machine Learning<br>Risk-Prediction Model for In-Hospital Mortality from <i>C. difficile</i>: National Inpatient Sample Study, ACG 2024 Annual Scientific Meeting Abstracts. Philadelphia, PA: American College of Gastroenterology.