EHR-Based Model Accurately Identifies High-Risk Patients for Gastric Cancer

An electronic health record (EHR)-based logistic regression model demonstrated strong performance in identifying individuals at high risk for noncardia gastric cancer (NCGC), offering a potential tool to guide targeted screening in the US, according to a study published in Gastro Hep Advances.

“In this study, we developed and assessed the performance of an EHR-based logistic regression predictive model in accurately identifying individuals at high risk for NCGC,” wrote Michelle Kang Kim, Department of Gastroenterology, Hepatology, and Nutrition, Cleveland Clinic in Cleveland, Ohio, and coauthors.

Using data from the Cleveland Clinic health system, the researchers conducted a retrospective case-control study of patients aged 40 to 80 years who received care between 2010 and 2021.

A total of 614 patients with confirmed intestinal-type NCGC were matched to 6331 controls without gastric cancer, with baseline demographic and clinical characteristics drawn from the EHR. The model incorporated variables such as age, sex, race, ethnicity, smoking history, body mass index, and comorbidities, including anemia and pernicious anemia, without reliance on endoscopic or pathology data.

Multivariable analysis identified several independent predictors of NCGC: age (odds ratio [OR] = 1.16 per 10-year increase), male sex (OR, 1.97), Black (OR, 3.07) or Asian (OR, 4.39) race, tobacco use (OR, 1.61), anemia (OR, 1.35), and particularly pernicious anemia (OR, 6.12). Conversely, hypertension, hypercholesterolemia, and liver disease were associated with lower odds of NCGC.

The model demonstrated good discrimination, with a median 0.632 area under the curve (AU) estimator of 0.731. Varying the risk threshold allowed optimization of sensitivity and specificity; for example, at a threshold of 0.172, sensitivity was 39.6%, specificity was 90.2%, and positive predictive value reached 1.0%—a benchmark relevant for cancer screening programs.

“In summary, we demonstrate the feasibility of an EHR-based logistic regression model in accurately predicting the probability of NCGC,” concluded the study authors.

Reference

Kim MK, Rouphael C, Wehbe S, et al. Using the electronic health record to develop a gastric cancer risk prediction model. Gastro Hep Adv. 2024;3(7):910-916. doi:10.1016/j.gastha.2024.07.001