Objectives: Few studies have evaluated the ability of the endoscopist to predict the presence of Barrett's esophagus (BE) at index endoscopy. The goals of this study were to determine the operating characteristics of endoscopy in diagnosing BE, and to determine the clinical and endoscopic predictors of BE in suspected BE patients at the index endoscopy.
Methods: From September 1993 to October 1997, endoscopic reports were examined to identify patients with suspected BE. All esophageal pathology reports during the same period were evaluated for the presence of specialized intestinal metaplasia.
Results: During the study period, 4053 endoscopies were performed on 2393 patients. Eight percent of all procedures were performed for suspected or confirmed BE. Fifty-three patients were known to have BE and thus their reports were excluded from this analysis. Five hundred seventy of the remaining patients had esophageal biopsies performed, and were included in this analysis. Among these 570 patients, 146 were suspected to have BE on endoscopy, while 424 were not suspected to have BE at the time of endoscopy. There were no differences among the two groups in terms of gender, race, and dyspepsia as an indication for the endoscopy. However, suspected BE patients were slightly younger and were more likely to have heartburn, but were less likely to have dysphagia as an indication for the endoscopy. The sensitivity and specificity of the endoscopists' assessments were 82% (95% confidence interval [CI], 72-92) and 81% (95% CI, 78-84), respectively. The positive predictive value and the negative predictive value were 34% and 97%, respectively. The positive likelihood ratio was 4.32 (95% CI, 3.49-5.31) and the negative likelihood ratio was 0.22 (95% CI, 0.13-0.38). Univariate analysis showed that endoscopists diagnosed BE in those with long-segment BE (LSBE) more accurately than in those with short-segment BE (SSBE) (55% vs 25% p = 0.001; odds ratio [OR] = 3.63, 95% CI, 1.71-7.70). Barrett's esophagus was correctly diagnosed in 38.5% of white patients but in only 14.7% of black patients (p = 0.01; OR = 3.63, 95% CI, 1.31-10.13). Multivariable logistic regression identified only the length of the columnar-appearing segment (p = 0.002; OR = 3.33, 95% CI, 1.54-7.17) and race (p = 0.08; OR = 2.31, 95% CI, 0.88-6.03) to be associated with the presence of BE on biopsy.
Conclusions: Barrett's esophagus is frequently suspected at endoscopy; SSBE was more frequently suspected than LSBE, but was correctly diagnosed only 25% of the time, versus 55% for LSBE. Endoscopists diagnosed BE with a sensitivity of 82% and a specificity of 81%. However, the positive predictive value was only 34%, whereas the negative predictive value was 97%. The length of the columnar-appearing segment is the strongest predictor of BE at endoscopy. Alternative methods are needed to better identify BE patients endoscopically, especially those with SSBE.