[Bayes analysis in clinical decision-making for solitary pulmonary nodules].

OBJECTIVE: To explore the feasibility and the value of Bayes analysis in clinical decision-making for solitary pulmonary nodules (SPNs). METHODS: We collected 352 consecutive SPN patients (malignancy, n=135; benignity, n=217) retrospectively to form the training set. Utilizing Bayes analysis, the prior odds of malignant SPNs and the likelihood ratios of clinical and CT findings were derived from the training set, which were then used to calculate the probability of malignancy in each SPN. Bayes analysis was also tested prospectively for its diagnostic validation and precision of predictive probability on the test set of 132 SPN patients (malignancy, n=61; benignity, n=71), and compared with the performance of physicians using routine judgment. The actual results of patients diagnosis were analyzed according to the scale of calculated malignant probability in SPNs. RESULTS: The sensitivity, specificity, and accuracy of Bayes analysis for the training samples were 88.9%, 93.1%, and 91.5%, respectively. In the test set, the sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of Bayes analysis were 88.5%, 85.9%, 87.1%, 84.4%, and 89.7%, respectively. The accuracy of Bayes analysis had no statistically significant difference with that of senior physician A (80.3%, chi2=2.37, P=0.122) and B (79.5%, chi2=3.12, P=0.076), and was higher than that of junior physician C (74.2%, chi2=7.05, P=0.012) and D (74.2%, chi2=6.56, P=0.009); The Brier score was 0.099, 0.140, 0.137,0.154, and 0.179 for Bayes analysis,senior physician A, senior physician B, junior physician C, and junior physician D, respectively. Excluding the solitary metastasis (n=11) misclassified, the false negative rate of Bayes analysis was 1.0% (5/484) for SPNs with <20% estimated probability of malignancy. CONCLUSION: Bayes analysis is accurate in qualitative diagnosis, precise in forecasting the malignant probability, and has low false negative rate for SPNs. It is feasible to use Bayes analysis for the management of SPNs.