Use and Interpretation of Hui and Walter's Latent- Class Reference-free Method to Estimate the Accuracy of a Diagnostic Test: Implications from a Study Assessing Preoperative Ct Scan Evaluation of Lung Cancer Staging

From data on preoperative thoracic CT scan, we estimated the sensitivity and specificity of CT scan in detecting locoregional extension of lung cancer by two approaches, one requiring a reference test and one not requiring such a test. Four senior radiologists independently interpreted CT scans of 85 patients diagnosed with resectable lung cancer. Radiologists' sensitivities and specificities were estimated by comparison with pathology findings from surgical specimens, and using Hui and Walter's reference-free latent class approach [1,2]. We investigated reasons why the two sets of estimates significantly and markedly differed. From simulations, we found that small sample bias in Hui and Walter's estimates could be a major factor [3]. Furthermore, errors in pathology findings could account for part of the difference. Finally, our analyses also revealed that the latent classes, on which Hui and Walter's approach rests, may differ intrinsically from the diagnostic classes as defined from pathology findings and may have a different interpretation [4]. Diagnostic parameters estimated with respect to