Objective: Reports on the accuracy of computed tomographic colonography (CTC) mainly involve series from expert institutions. The aims of this study were to assess CTC accuracy in a nationwide population and to relate it to radiologist performance in their initial training.
Design: Nationwide multicentre trial.
Setting: Twenty-eight radiologists, working in 26 mostly academic clinical units, were involved in the study after having attended a formal specialised 2-day training session on CTC. They worked through a training set of 52 cases with automatic feedback after an attempt at each case.
Patients: The study enrolled 845 patients with average and high risk of colorectal cancer, 737 of whom had both complete CTC and videocolonoscopy data, which constituted the dataset.
Interventions: Patients underwent same-day CTC followed by videocolonoscopy with segmental unblinding of CTC results.
Main outcome measures: Sensitivity, specificity and positive and negative predictive values for detection of polyps ≥ 6 mm in per-patient and per-lesion analyses of CTC without computer-aided detection.
Results: Sensitivity, specificity and positive and negative predictive values for patients with polyps ≥ 6 mm were 69% (95% CI 61% to 77%), 91% (95% CI 89% to 94%), 67% (95% CI 59% to 74%) and 92% (95% CI 90% to 94%), respectively. Univariate analysis showed that the detection rate for polyps ≥ 6 mm was linked to neither radiologist case volume nor number of polyps, but was related to sensitivity achieved in the training set. Pooled sensitivity was 72% (95% CI 63% to 80%) versus 51% (95% CI 40% to 60%) for radiologists achieving above and below median sensitivity in the training set (61%), respectively. Multivariate analysis showed that sensitivity for polyps ≥ 6 mm in the training set was the only remaining significant predictive factor for subsequent performance.
Conclusions: Radiologist sensitivity CTC for detection of polyps ≥ 6 mm in training was the sole independent predictor for subsequent sensitivity in detection of such polyps.