Date of Completion
5-7-2016
Embargo Period
4-16-2018
Advisors
Miranda Lynch, PhD; Brady McKee, MD
Field of Study
Public Health
Degree
Master of Public Health
Open Access
Open Access
Abstract
Background: United States Preventive Services Task Force (USPSTF) and Centers for Medicare & Medicaid Services (CMS) recommendations for annual screening for lung cancer with low dose CT (LDCT) scans rely on age and smoking history to identify those at high risk for lung cancer. The Tammemagi et al. six year lung cancer risk prediction model, PLCOm2012, developed and validated in large lung cancer screening clinical trials, demonstrated good predictive performance in screening selection. However, the model has not been validated in clinical practice. Validating the model in clinical practice would increase confidence in its ability to provide information for shared decision making discussions in the near term and would potentially allow for selection of other high risk groups, not currently recommended to be screened, in the future.
Methods: Retrospective evaluation of the predictive performance of the Tammemagi et al. six year lung cancer risk prediction model in the Lahey Hospital & Medical Center, Lahey physician referred patients enrolled in the lung cancer screening program between January 1, 2012 and November 30, 2015 (n=2302). Predictor variable data were gathered from the program clinical data base and program participant clinic medical records. All patients met the National Comprehensive Cancer Network (NCCN) Lung Cancer Screening Guidelines Group 1 or Group 2 high-risk criteria.
Results: The model six year mean risk for lung cancer was higher for participants with lung cancer, 4.56%, as compared to those without lung cancer, 3.55% (p=0.0265). Area under the curve (AUC) of the receiver operator characteristics (ROC) was 0.63 (95% CI 0.57 – 0.69). The mean absolute difference between observed and predicted risk was 0.013 or less for the first 9 deciles.At the 1.51% predicted risk recommended screening threshold; sensitivity = 85.7%, specificity = 29.7%, and PPV = 3.7%. In sub-group analysis, for NCCN Group 2 (younger, lighter smoking history, no limit on time quit and one additional risk factor) the mean predicted risk for participants with lung cancer was 2.39% as compared to 1.83% for those without lung cancer but the difference was not statistically significant; p=0.2507. However, the incidence of lung cancer was the same for NCCN Group 2 as for the complete sample. NCCN Group 2 model AUC was 0.634 (95% CI 0.522 – 0.746), the sensitivity and specificity of the model at the recommended screening threshold were 64.7% and 56.0%, respectively and PPV was 4.2%.
Conclusions: Lung cancer risk prediction model, PLCOm2012noEd, predictive performance in a clinical lung cancer screening program was adequate to help patients and their physicians assess individual risk of lung cancer relative to the recommended model risk screening threshold (1.51%) and to supplement USPSTF and CMS screening program entry criteria for shared decision making discussions. Model risk predictive capability for the NCCN Group 2 subgroup did not match actual screening program lung cancer results.
Recommended Citation
Borondy Kitts, Andrea K., "A Retrospective Study Assessing the Predictive Performance of a Lung Cancer Screening Risk Prediction Model in a Clinical Lung Cancer Screening Program" (2016). Master's Theses. 909.
https://digitalcommons.lib.uconn.edu/gs_theses/909
Major Advisor
Julie Robison, PhD