Predictors of gastroesophageal reflux disease revisited: a novel gastroesophageal reflux disease clinical prediction score.

Abstract

Symptom assessment and proton pump inhibitor (PPI) trial are limited in predicting pathologic gastroesophageal reflux disease (GERD) on pH testing and endoscopy, which has led to PPI overuse. We aimed to develop a novel scoring system using demographic factors, clinical factors, and symptoms to predict pathologic GERD. Adult patients with GERD were prospectively enrolled into a Heartburn Center Registry from April 2019 to September 2021. Demographics, patient responses to validated questionnaires, results of endoscopy and pH testing were collected. We selected variables associated with pathologic GERD (P < 0.10) in bivariable analyses for inclusion in multivariable models to predict pathologic GERD. A total of 365 patients were included. About 199/365 (33%) of these patients had pathologic GERD based on abnormal pH testing (49.7%) and/or abnormal esophagogastroduodenoscopy (EGD) (61.3%). Our clinical prediction model included 10 variables and had a sensitivity of 90%, specificity of 56%, and area under the receiver operator characteristic curve (AUROC) 0.77, and was then validated using a separate cohort. The model was then refitted using the expanded criteria for definition of GERD from the new Lyon Consensus 2.0. Four new variables led to improved model performance including history of chronic lung disease, history of asthma, history of gastroparesis and heartburn after meals. This expanded model including 14 variables had a sensitivity of 84.2%, specificity of 71.7%, and AUROC of 0.80, which was then validated in a separate cohort. Our novel 10-variable GERD prediction model performs well for estimating risk for pathologic GERD without invasive testing and is potentially a practical clinical screening tool for pathologic GERD.