External validation of a prediction model for pathologic N2 among patients with a negative mediastinum by positron emission tomography.

BACKGROUND

A prediction model for pathologic N2 (pN2) among lung cancer patients with a negative mediastinum by positron emission tomography (PET) was recently internally validated. Our study sought to determine the external validity of that model.

METHODS

A cohort study [2005-2013] was performed of lung cancer patients with a negative mediastinum by PET. Previously published model coefficients were used to estimate the probability of pN2 based on tumor location and size, nodal enlargement by computed tomography (CT), maximum standardized uptake value (SUVmax) of the primary tumor, N1 disease by PET, and pretreatment histology.

RESULTS

Among 239 patients, 18 had pN2 [7.5%, 95% confidence interval (CI): 4.5-12%]. Model discrimination was excellent (c-statistic 0.80, 95% CI: 0.75-0.85) and the model fit the data well (P=0.191). The accuracy of the model was as follows: sensitivity 100%, 95% CI: 81-100%; specificity 49%, 95% CI: 42-56%; positive predictive value (PPV) 14%, 95% CI: 8-21%, and negative predictive value (NPV) 100%, 95% CI: 97-100%. CI inspection revealed a significantly higher c-statistic in this external validation cohort compared to the internal validation cohort. The model's apparently poor specificity for patient selection is in fact significantly better than usual care (i.e., aggressive but allowable guideline concordant staging) and minimum guideline mandated selection criteria for invasive staging.

CONCLUSIONS

A prediction model for pN2 is externally valid. The high NPV of this model may allow pulmonologists and thoracic surgeons to more comfortably minimize the number of invasive procedures performed among patients with a negative mediastinum by PET.