Existing General Population Models Inaccurately Predict Lung Cancer Risk in Patients Referred for Surgical Evaluation

Abstract

Background atients undergoing resections for suspicious pulmonary lesions have a 9-55% benign rate. Validated prediction models exist to estimate the probability of malignancy in a general population and current practice guidelines recommend their use. We evaluated these models in a surgical population to determine the accuracy of existing models to predict benign or malignant disease. Methods We conducted a retrospective review of our thoracic surgery quality improvement database (2005-2008) to identify patients who underwent resection of a pulmonary lesion. Patients were stratified into subgroups based on age, smoking status and fluorodeoxyglucose positron emission tomography (PET) results. The probability of malignancy was calculated for each patient using the Mayo and SPN prediction models. Receiver operating characteristic (ROC) and calibration curves were used to measure model performance. Results 89 patients met selection criteria; 73% were malignant. Patients with preoperative PET scans were divided into 4 subgroups based on age, smoking history and nodule PET avidity. Older smokers with PET-avid lesions had a 90% malignancy rate. Patients with PET- non-avid lesions, or PET-avid lesions with age<50 years or never smokers of any age had a 62% malignancy rate. The area under the ROC curve for the Mayo and SPN models was 0.79 and 0.80, respectively; however, the models were poorly calibrated (p<0.001). Conclusions Despite improvements in diagnostic and imaging techniques, current general population models do not accurately predict lung cancer among patients ref erred for surgical evaluation. Prediction models with greater accuracy are needed to identify patients with benign disease to reduce non-therapeutic resections.

Evaluation of Individuals With Pulmonary Nodules: When Is It Lung Cancer?

“…( Tables S5, S6 ). [33][34][35] Another used data from the PLCO (Prostate, Lung, Colorectal, Ovarian Cancer Screening Trial) of lung cancer screening with chest radiography and found that although a lung mass (not surprisingly) was highly predictive of malignancy (OR, 11.2; 95% CI, 6.3-19.9), the fi nding of a lung nodule was not (OR, 1.4; 95% CI, 0.8-2.5). 36 The most extensively validated model was developed by investigators at the Mayo Clinic who used multiple logistic regression analysis to identify six independent predictors of malignancy in 419 patients with noncalcifi ed nodules that measured between 4 and 30 mm in diameter on chest radiography.…”

Accuracy of Clinicians and Models for Estimating the Probability That a Pulmonary Nodule Is Malignant

“…For those who are more quantitatively inclined, clinical prediction models (or risk assessment models) have been developed to provide a more explicit, transparent, and reproducible assessment of this risk. One commonly cited model was developed and internally validated at the Mayo Clinic (4) and subsequently validated in independent samples of U.S. veterans with lung nodules, a surgical population from an academic center in the southeastern United States, and patients from the Netherlands who underwent positron emission tomography (5)(6)(7). A similar model was developed and validated using samples of U.S. veterans with a higher prevalence of malignancy (8).…”

Segmental 18F-FDG-PET/CT in a single pulmonary nodule: a better cost/effectiveness strategy

“…The adoption of this latter strategy, defined on the basis of risk of malignancy, young age, patient preference (as in anxious subjects), may be better justified, keeping in mind the ALARA principle, choosing a segmental scan with respect to an unjustified whole body PET/CT. On the other hand, the advantage of a surveillance strategy, for 4 years, has to be weighed against the disadvantage of delaying diagnosis of malignant nodules [7], the incidence of which is reported up to 62 % in patients with higher likelihood of benign disease [27].…”