Clinical Prediction Models

Published Date: 19th July 2016

Publication Authors: Jones C

​

Background 

Management of solitary pulmonary nodules (SPNs) depends critically on the pre-test probability of malignancy. Several quantitative prediction models have been developed using clinical and radiological criteria. Three models include CT criteria (Mayo, Veterans Association, Brock University) with a fourth model (Herder) incorporating FDG avidity on CT-PET scan in addition. These models have not been validated in a UK population, and the current study aimed to compare their performance in a population of patients recruited from a UK teaching hospital.

Methods 

Patients with SPNs (4–30 mm) were retrospectively identified from the lung cancer MDT and a nodule follow-up clinic (n = 246). All patients had a final diagnosis confirmed by histology or radiological stability on a 2-year follow up. For each patient, the probability of the pulmonary nodule being malignant was calculated using the four models described. The models were used both in a restricted cohort of patients based on their respective exclusion criteria, and in the total cohort of patients. The accuracy of each model was assessed by calculating the area under the receiver operating characteristic (ROC) curve.

Results 

The median age of the patient population was 69 years (range 32–94) and 50% were male. The prevalence of malignancy was 40.6% (33.3% primary lung cancer, 7.3% metastatic disease). Figure 1 shows the distribution of the probabilities of malignancy according to the four different models.

The areas under the ROC curves for the cohorts restricted by respective exclusion criteria were (AUC, 95% CI): Mayo 0.892 (0.847–0.937); VA 0.736 (0.672–0.801); Brock 0.901 (0.855–0.947) and Herder 0.924 (0.875–0.974). For the total cohort, the AUC values were Mayo 0.873, VA 0.736, Brock 0.867 and Herder 0.916.

There was no statistical difference between the Mayo and Brock models, but both were significantly better than VA (AUC difference of 0.14 and 0.13 respectively, p ≤ 0.0001 for both). The Herder model performed significantly better than both Mayo and Brock models (AUC difference of 0.10 and 0.14 respectively, p ≤ 0.01).

Conclusion 

Both the Mayo and the Brock models perform well in a UK population, but accuracy is improved by incorporating CT-PET findings using the Herder prediction model.

Al-Ameri, A; Malhotra, P et al. (2014).  Clinical prediction models for malignancy in solitary pulmonary nodules-a validation study in a UK population . Thorax. 69 (Suppl 2), A40.

« Back