P1440 - Development of a Model to Predict Post-Acute Pancreatitis Diabetes

Award: Presidential Poster Award

Harry Trieu, MD¹, Aneesa Chowdhury, MD¹, Jennifer L. Dodge, MPH², Stephen Pandol, MD³, Marc Goodarzi, MD, PhD³, Ira Shulman, MD¹, Christie Jeon, PhD³, James Buxbaum, MD, MS^1
1Keck School of Medicine of USC, Los Angeles, CA; ²University of Southern California, Los Angeles, CA; ³Cedars-Sinai Medical Center, Los Angeles, CA

Introduction: Acute pancreatitis (AP) is the most common gastrointestinal cause for hospital admission in the United States. A significant proportion of patients will develop post-pancreatitis diabetes mellitus (PPDM-A) several years of their index admission. We aimed to develop a predictive model to help clinicians identify patients at risk for developing PPDM-A.

Methods: A prospective cohort of patients aged 18 or older presenting to LAC+USC Medical Center with a diagnosis of acute pancreatitis according to the Revised Atlanta Criteria between March 2015 and March 2022 was used to build the model. Patients with a diagnosis of diabetes that preceded their index admission for AP were excluded. Patients with a hemoglobin A1c>6.5% or documentation of a diabetes diagnosis after index admission for AP were deemed to have developed PPDM-A.



The dataset was then divided into training and testing sets with 75% and 25% of patients assigned to each, respectively. An elastic net regression model was fit on the training set to select variables correlated with the development of PPDM-A. We then validated the model on previously unseen data (the testing set) and calculated an area under the receiver operator characteristic curve (AUROC).

Results: One thousand and twenty-three patients with median age 42 years, 80.4% Hispanic, and 45.9% female met inclusion criteria for this study (Table). The most common AP etiology was gallstones, seen in 488 (47.7%) patients, followed by alcohol, seen in 275 (26.9%) patients. The training and testing sets consisted of 767 and 256 patients. Thirty-six patients (3.5%) developed PPDM-A.



The best performing elastic net model included alkaline phosphatase, history of smoking, history of pre-diabetes, history of cirrhosis, moderate or severe pancreatitis, ICU level of care, and history of recurrent pancreatitis as predictors (Figure 1). It achieved an AUROC of 0.78 (95% CI 0.69-0.86) in the training and 0.81 (95% CI 0.64-0.98) in the testing set.

Discussion: In this study, we developed an elastic net regression model which predicts development of PPDM-A and achieves an AUROC of 0.81 (95% CI 0.64-0.98) in unseen data. Pending external validation, this model may help clinicians identify patients at risk for developing this serious complication and guide early treatment.



Disclosures:


Harry Trieu, MD¹, Aneesa Chowdhury, MD¹, Jennifer L. Dodge, MPH², Stephen Pandol, MD³, Marc Goodarzi, MD, PhD³, Ira Shulman, MD¹, Christie Jeon, PhD³, James Buxbaum, MD, MS¹. P1440 - Development of a Model to Predict Post-Acute Pancreatitis Diabetes, ACG 2023 Annual Scientific Meeting Abstracts. Vancouver, BC, Canada: American College of Gastroenterology.