Award: ACG Governors Award for Excellence in Clinical Research
Introduction: Esophageal varices (EV) are a common and potentially lethal complication of cirrhosis. National and international guidelines recommend universal EV screening of all patients with cirrhosis. However, a significant number of patients undergoing screening esophagogastroduodenoscopy (EGD) do not have varices. Therefore, the aim of this study was to identify clinical predictors of esophageal varices in order to create a clinically-practical predictive score to potentially screen out low-risk patients from unneeded EGDs.
Methods: Patients with cirrhosis undergoing first screening EGD for EV from April 2016 to December 2017 at both Olive View-UCLA (OV) and the VA Greater Los Angeles Healthcare System (VA) were included. Demographic, clinical, biochemical, radiologic, and endoscopic data were abstracted using a standardized collection form. Patients from OV and VA were pooled, and Chi-squared test and student’s t-test were used to compare differences in distribution and means, respectively, between those with and without EV. Pooled patients were then randomly assigned into two groups in a 2:1 ratio: a discovery cohort and a validation cohort. Within the discovery cohort, a random forest classifier was used to predict the presence of EV on EGD. Variables with the greatest accuracy score were used to create a predictive score. The EV predictive score with the largest area under the receiver operating curve (AUC) in the discovery cohort was selected and then applied to the validation cohort.
Results: A total of 166 patients were included in the study (discovery cohort n=110, validation cohort n=56). Patients with EV on EGD had a lower mean hemoglobin, platelet count, BUN, and albumin and a higher mean INR compared to those without EV (Table 1). The following EV score formula was developed:
INR x 105 (Platelet X Hemoglobin X Albumin X BUN)
AUCs in the discovery and validation cohort for predicting EV on EGD were similar at 0.768 and 0.771, respectively. A score cutoff value of 1.2 had a sensitivity of 90.9%, specificity of 55.9%, and NPV of 91.6% of predicting the absence of EV on EGD. Using this score in the validation cohort would have reduced the number of unneeded EGDs by 70.5%.
Discussion: These results suggest that noninvasive clinical markers may be used to accurately predict the presence or absence of EV. Using this EV score would greatly decrease the number of unneeded EGDs, thereby reducing potential procedure and anesthesia related risks and costs.