Models Estimate Probability of Cancer in Pulmonary Nodules Detected on First Screening CT
In a study reported in The New England Journal of Medicine, Annette McWilliams, MD, of Vancouver General Hospital, and colleagues analyzed data from subjects undergoing low-dose computed tomography (CT) screening for lung cancer to identify factors that might predict whether lung nodules found on first screening are malignant. They developed models that accurately estimated likelihood of malignancy.
Study Details
Data from two cohorts of participants undergoing low-dose CT screening for lung cancer were analyzed. The development data set included participants in the Pan-Canadian Early Detection of Lung Cancer Study (PanCan). The validation data set included participants involved in chemoprevention trials at the British Columbia Cancer Agency (BCCA). The final outcomes of all nodules of any size detected on baseline low-dose CT scans were tracked. Parsimonious (including only predictors that were significant at P < .05) and fuller (including additional predictors that were thought a priori to be associated with risk of lung cancer and with P < .25) multivariable logistic-regression models were developed to estimate the probability of lung cancer.
Frequency of Cancer
In the PanCan data set, 1,871 persons had 7,008 nodules, 102 of which were malignant. In the BCCA data set, 1,090 persons had 5,021 nodules, 42 of which were malignant. Among persons with nodules, the rates of cancer in the two data sets were 5.5% and 3.7%, respectively.
Factors Included in Predictive Models
In a univariate analysis, significant consistent predictors of lung cancer included the size, type (nonsolid, part-solid, or solid), and location of the nodules and the number of nodules detected. Spiculation was a significant predictor in the PanCan data, but data on spiculation were not available in the BCCA data set. In a parsimonious model without spiculation, diagnosis of cancer in a nodule was associated with female vs male sex, increasing nodule size, and nodule location (upper vs middle or lower lobe). In a parsimonious model with spiculation, cancer was associated with each of these factors plus spiculation. The fuller models (without and with spiculation) included older age, family history of lung cancer, emphysema, lower nodule count, and part-solid nodules vs solid nodules in addition to factors in the parsimonious models.
Performance of Models
Both parsimonious and full models showed excellent discrimination in the PanCan and BCCA data sets, with all having receiver-operating-characteristic area under the curve (AUC) > 0.90. The models performed well when applied to nodules ≥ 10 mm or smaller; for example, for such nodules the AUCs using the parsimonious model without spiculation were 0.894 in the PanCan data set and 0.907 in the BCCA data set.
With regard to how such models would perform in the screening setting, it was estimated, for example, that on the assumption of a 5% risk of cancer in a population, the parsimonious model including spiculation would have 71.4% sensitivity, 95.5% specificity, 18.5% positive predictive value, and 99.6% negative predictive value. Only 5.5% of nodules would be classified as positive.
The investigators concluded, “Predictive tools based on patient and nodule characteristics can be used to accurately estimate the probability that lung nodules detected on baseline screening low-dose CT scans are malignant.”
The study was funded by the Terry Fox Research Institute and others.
The content in this post has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the ideas and opinions of ASCO®.
