P0891 - Development and Validation of a Machine Learning Risk-Prediction Model for In-Hospital Mortality from Clostridiodes Difficile in Patients with Inflammatory Bowel Disease: National Inpatient Sample-Based Study
Introduction: Clostridioides difficile (Cdiff) infection is associated with increased mortality risk in patients with underlying inflammatory bowel disease (IBD).Due to the scarcity of a prognostic scoring system, we aimed to assess for risk factors and to develop a novel machine learning-based risk score model to predict in-hospital mortality in Cdiff inpatients with IBD.
Methods: The NIS 2016-2020 database was used to identify all adult patients ( >18 years) with primary diagnosis of Cdiff and secondary diagnosis of IBD, using ICD-10 code A04.7, K50, and K51. Relevant clinical characteristics and outcomes were extracted. The primary endpoint was in-hospital, all-cause mortality. We performed multivariate analysis for predictors of mortality. We then generated a risk score by dividing the dataset into three subsets: training, validation, and testing. Risk scores were generated using the Autoscore package in R software, a machine learning-based tool, which uses random forest-based algorithm that combines the output of multiple decision trees. Performance was evaluated using the area under the ROC (AUC) with 95% CI
Results: Among 13,791 inpatients identified with Cdiff and IBD [age 52.7 21.5 years, 7,999 female (58%)], there were 446 (3.23%)deaths. On multivariate analysis, factors significantly associated with the risk of death at a P-value of < 0.001 were age (OR=1.04 [1.03-1.05]), malnutrition (OR=1.56 [1.23-1.98]),AKI (OR=1.66 [1.28-2.13]), acute liver failure(ALF) (OR=3.73 [2.31-6.01]), shock (OR=5.88 [4.43-7.81]), ICU admission (OR=7.55 [5.72-9.96]), and malignancy (OR=2.45 [1.78-3.83])(Table 1). We then used 14 covariates (derived from a Parsimony plot) for our risk score (0-100): age, ICU admission, all-cause shock,AKI, malnutrition, female sex,CKD, transfusion of RBC, malignancy, electrolyte abnormality, PVD, ALF, CHF, and severe sepsis (Figure 1B). The AUC for the derivation and validation cohort was 0.9185(0.8949-0.942) and 0.9647(00.953-0.9763) respectively (Figures 1 and 2)
Discussion: We identified risk factors associated with mortality among inpatients with IBD and Cdiff. We also developed and internally validated a machine learning-based risk score model to predict outcomes in CDiff patients with IBD, achieving a high AUC of 0.9647 (00.953-0.9763). Risk score models play a crucial role in risk stratification and management. Future studies are necessary to externally validate our findings and to include concomitant medication use which is not available in the NIS database
Figure: Figure showing the AUC
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.
Prabhat Kumar indicated no relevant financial relationships.
Ankit Agrawal indicated no relevant financial relationships.
Jean-Paul Achkar indicated no relevant financial relationships.
Umesh Bhagat, MD1, Prabhat NA. Kumar, MD2, Ankit Agrawal, MD1, Jean-Paul Achkar, MD3. P0891 - Development and Validation of a Machine Learning Risk-Prediction Model for In-Hospital Mortality from <i>Clostridiodes Difficile</i> in Patients with Inflammatory Bowel Disease: National Inpatient Sample-Based Study, ACG 2024 Annual Scientific Meeting Abstracts. Philadelphia, PA: American College of Gastroenterology.