Miniprobe confocal laser microscopy for the detection of invisible neoplasia in patients with Barrett's oesophagus

Objective: The biggest challenge in endoscopic surveillance of Barrett's oesophagus is better detection of neoplasia in mucosa of normal macroscopic appearance. We evaluated in vivo miniprobe confocal laser microscopy (CLM) for the detection of invisible Barrett's neoplasia. Design: Prospective two-centre trial in two phases: phase I to establish criteria of Barrett's neoplasia and phase II to test these criteria. Patients and intervention: 296 biopsy sites in 38 consecutive patients with Barrett's oesophagus (mean age 62.1 years, 89.5% men, median length of the Barrett's oesophagus, 3 cm) were examined with standard high-resolution endoscopy and by miniprobe CLM, with precise matching of CLM recordings to biopsy sites. CLM image criteria for normal versus neoplastic Barrett's oesophagus were established from 95 biopsies of 15 patients (phase I); these criteria were then prospectively tested on 201 biopsies from the remaining patients without visible focal changes (phase II). All 201 CLM video recordings from phase II cases were randomised and blindly evaluated by two gastroenterologists. Main outcome measure: The primary endpoints were accuracy values in diagnosing HGIN or early carcinoma (EC) on a per-biopsy basis. Secondary endpoints included inter-observer agreement. Results: All initially defined miniprobe CLM criteria (phase I) were significantly more frequently detected in HGIN/EC sites compared with sites with no or low grade neoplasia (phase II). In a per-biopsy analysis, sensitivity and specificity for two independent investigators were 75.0% and 88.8%, and 75.0% and 91.0%, respectively, translating at best into a positive predictive value of 44.4% and a negative predictive value of 98.8%. Interobserver agreement was good (kappa 0.6). Conclusion: Miniprobe CLM showed a high negative predictive value for the diagnosis of endoscopically invisible neoplasia in Barrett's oesophagus; sensitivity, however, has still to be improved.