OBJECTIVE: Accurately predicting Barrett's esophagus (BE) in patients with gastroesophageal reflux disease (GERD) is difficult. Using logistic regression analysis of symptom questionnaire scores we created a model to predict the presence of BE. METHODS: We conducted a logistic regression analysis of symptom data collected prospectively on 517 GERD patients and created a prediction model based on patient gender, age, ethnicity, and symptom severity. RESULTS: There were 337 (65%) males and 180 (35%) females, of whom 99 (19%) had Barrett's esophagus (BE). Multiple logistic regression analysis was performed to determine the predictive ability of gender, age, and ethnicity along with symptoms of heartburn, nocturnal pain, odynophagia, presence of belching, dysphagia, relief of symptoms with food, and nausea. The only significant predictors (at the 0.05 level) were male gender, heartburn, nocturnal pain, and odynophagia (all with positive effects on the presence of BE) and dysphagia (which had a negative effect). A nomogram was produced to show the effect of a given predictor on the probability of having BE in the context of the effects of the other predictors, and to estimate the probability of having BE for a given individual. The mean score (+/- SD) for the BE patients in our sample was 397.4 +/- 46.2 with a range of 292-530. For the patients without BE, the mean score (+/- SD) was 351.3 +/- 60.3 with a range of 190 - 528 (p < 0.001). If screening for BE is performed at a score of 375 or more, our model would have a specificity of 63% with a sensitivity of 77% (95% CI 61-86% given the 63% specificity). CONCLUSIONS: By asking seven questions about symptom severity, clinicians may be able to assign a probability to the presence of BE, and thus, determine the need for endoscopy in GERD patients.
