A model for prediction of recurrence of non-small cell lung cancer based on clinical data and CT imaging characteristics

Abstract

Objectives

To establish a model for prediction of recurrence of non-small cell lung cancer (NSCLC) based on clinical data and computed tomography (CT) imaging characteristics.

Methods

A total of 695 patients with surgically resected NSCLC confirmed by pathology at three centers were retrospectively investigated. 626 patients from center 1 were randomly divided into two sets in a ratio of 7:3 (training set, n = 438; testing set, n = 188), 69 patients from center 2 and 3 were assigned in the external validation set. Univariate and binary logistic regression analyses of clinical and CT imaging features determined the independent risk factors used to construct the model. The receiver-operating characteristic curve nomogram and decision curves analysis were used to evaluate the predictive ability of the model.

Results

The mean patient age was 63.3 ± 10.1 years, and 44.7 % (311/695) were male. The univariate and binary logistic regression analyses identified four independent risk factors (age, tumor markers, consolidation/tumor ratio, and pleural effusion), which were used to construct the prediction model. In the training set, the model had an area under the curve of 0.857, an accuracy of 71.7 %, a sensitivity of 88.1 %, and a specificity of 70.0 %; in the testing set, the respective values were 0.867, 75.5 %, 94.4 %, and 73.5 %; in the external validation set, the respective values were 0.852, 79.7 %, 83.3 %, 78.9 %.

Conclusion

A prediction model based on clinical data and CT imaging characteristics showed excellent efficiency in prediction of recurrence of NSCLC. Clinical use of this model could be useful for selection of appropriate treatment options.