PURPOSE: We have shown that prediction models by the VA and Mayo estimate poorly the probability of malignancy of solitary pulmonary nodules (SPN) in the Philippines. We developed and validated a clinical prediction model to identify malignant SPN, based on clinical data and radiographic characteristics in the Philippine setting.
METHODS: A prediction model was developed from a retrospective review of patients with SPN from October 2006 to March 2008. Univariate and multiple logistic regression analysis were used to identify independent clinical variables. This model was validated on a prospective cohort of SPN patients (April to August 2008). Receiver operating characteristic (ROC) curves and 95% confidence interval (CI) were constructed. Calibration was done by dividing the sample into five equal groups based on predicted probability and plotting the median probability of each quintile against the observed frequency of malignancy for that group.
RESULTS: Seventy-six SPN patients were included in the development of the prediction model, where size, margin and smoking history were found significant in the multivariate analysis. Prevalence of malignancy was 33%. The area under ROC curve was 0.92; 95% CI of 0.77- 0.95. The equation obtained was: Pre-test probability of a malignant SPN = ex/(1+ex), where x = -4.3368 + (0.3884603*SIZE) + (2.334691*MARGIN) + (2.310823*SMOKING HISTORY). Fifty-eight patients with SPN were included in the validation sample. Prevalence of malignancy was 36%. The ROC curve was 0.91; 95% C.I. of 0.85 to 0.97. Median predicted probabilities in all quintiles were lower than the observed frequency of malignant nodules, probably reflective of the validation sample’ s higher prevalence of malignancy.
CONCLUSION: The local clinical model appeared to be sufficiently accurate to inform clinical decisions about the choice and interpretation of subsequent diagnostic tests. The accuracy of the local clinical prediction model was similar to that reported in its development.
CLINICAL IMPLICATIONS: In the Philippines, and possibly for other countries with a high TB-burden, our clinical prediction model has a better estimate to the probability of SPN than both the VA and Mayo.
DISCLOSURE: Shane Ceniza, No Financial Disclosure Information; No Product/Research Disclosure Information