Radiologists Outperform Vancouver Risk Model in Detecting Malignant Lung Cancer

Atezolizumab is the most recent in a line of immune checkpoint inhibitors approved for patients with
Atezolizumab is the most recent in a line of immune checkpoint inhibitors approved for patients with
Experienced and trainee radiologists had superior ability to predict the risk for lung cancer compared with the Vancouver model.

Radiologists both experienced and in training have demonstrated better ability to evaluate cancer risk than the Vancouver Lung Cancer Risk Prediction Model, according to a study recently published in CHEST.

A total of 100 nodules (n=20 with proven cancers, n=80 benign) from the National Lung Cancer Screening Trial (NLST) were used in this study. A team consisting of 3 thoracic radiologists with professional experience and 3 radiologists in training estimated the risk for cancer in each nodule. The team worked independently at first, then were given knowledge of the Vancouver model’s outcomes. Receiver operating characteristic analysis was used to calculate results, with differences in area under the curve (AUC) indicating statistical significance via a Dorfman-Berbaum-Metz multi-reader multi-case analysis.

Predicting risk for malignant cancer was significantly better among the human team vs the Vancouver model (AUC, 0.85±0.05 vs 0.77±0.06; P =.0010). Furthermore, use of the Vancouver model was not associated with an improvement in accuracy (0.84±0.06). Radiologists with experience outperformed trainees, although not significantly. Accurate identification of malignant vs benign nodule morphology was better among the human team, as the model largely relies on the size of the nodule to estimate risk.

The study authors concluded that “even with high average accuracy, a computer model may produce misleading results in an important subset of cases, potentially leading to adverse outcomes. Nonetheless, the concept of an objective risk model is logical, and incorporation of additional features into the model in the future using advanced models based on machine learning and texture analysis will likely provide more nuanced weighting of morphological nodule features to enable a more robust and clinically useful decision support tool.”

Related Articles

Disclosure: Multiple authors disclosed affiliations with pharmaceutical companies. See the reference for complete disclosure information.


MacMahon H, Li F, Jiang Y, Armato SG 3rd. Accuracy of the Vancouver Lung Cancer Risk Prediction Model compared with radiologists [published online April 11, 2019]. CHEST. doi:10.1016/j.chest.2019.04.002