Development of a predictive model using endoscopic features for gastric cytomegalovirus infection in renal transplant patients.

Background. Cytomegalovirus (CMV) is a common viral pathogen in transplant patients which often targets the stomach. However, the endoscopic characteristics of gastric CMV infection are not well established. We aimed to develop a predictive model using endoscopic findings for gastric CMV infection in renal transplant patients. Methods. A retrospective study of 287 kidney transplant recipients who underwent endoscopy with biopsy for suspected CMV infection from January 2006 to November 2015 at a tertiary referral hospital was performed. CMV infection was defined based on inclusion bodies in hematoxylin and eosin and immunohistochemical staining. Endoscopic and clinical parameters related to gastric CMV infection were selected by univariate analyses. Multivariate logistic regression was used to create a predictive model from beta-coefficients. Results. CMV was present in 107 (37.7%) of the 287 patients. Multivariate analysis found age (odds ratio [OR], 0.964; 95% confidence interval [CI], 0.938-0.99; P = 0.008), erosions with surface exudate (OR, 5.34; 95% CI, 2.687-10.612; P < 0.001), raised shape of erosions (OR, 3.957; 95% CI, 1.937-8.083; P < 0.001), and antral location of ulcers (OR, 15.018; 95% CI, 5.728-39.371; P < 0.001) as independent predictive factors for gastric CMV infection. Using the predictive model created from this analysis, the sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were 71.03%, 85.56%, 74.51%, 83.24%, and 80.14%, respectively. The area under the receiver operating characteristic curve of this model for detecting CMV infection was 0.850 (95% CI, 0.803-0.889; P < 0.001). Conclusions. The predictive model with typical endoscopic findings may be useful for detecting gastric CMV infection in renal transplant patients.