Browsing M.Sc. Applied Health Sciences by Author "Mahmud, Meera"
Evaluation of the PLCOM2012 Risk Prediction Model and National Lung Screening Trial Criteria for Selecting Individuals for Lung Cancer ScreeningMahmud, Meera; Applied Health Sciences ProgramBackground: Lung Cancer (LC) is the leading cause of cancer death in North America. Cancer screening trials, such as the PLCO (Prostate, Lung, Colorectal and Ovarian) and NLST (National Lung screening trial) evaluate LC mortality. There has been a growing interest in risk prediction modelling for selecting high-risk individuals for LC screening. Increased risk for developing LC is however associated with greater risk of dying from non-LC (competing) causes; the presence of coexisting illnesses may negate the net benefit of LC screening. Purpose: The focus of this study was to compare the two selection criteria methods, the NLST criteria and the PLCOM2012 model (at 1.5% and 2% 6-year risk), for selecting smokers age of 55 to 75 for screening who may develop LC, and to evaluate which risk factors are strongly associated with competing causes of death (CCoD). Methods: Contingency table and logistic regression analysis, using STATA software, were used to analyze the results of applying both criteria on the PLCO ever-smokers population (N= 74 207). A CCoD logistic regression model was developed using 5-year follow-up data to assess the association between each PLCOM2012 model predictor and non-LC death. Predictor variables were ranked by their ability to predict 6-year LC incidence and 5-year non-LC death. Predictive performance, discrimination (area under the receiving-operating-characteristic curve [ROC-AUC]) and calibration were assessed. Results: Significantly higher LC proportions were found using the PLCOM2012 model than the NLST criteria in various individual characteristics. Increasing the model threshold resulted in less false positives, higher positive predictive value and probability of 6-year LC incidence. The PLCOM2012 model was also shown to be significantly associated with 5-year non-LC death (p<0.001). The CCoD model demonstrated fair discrimination (AUC=0.7114) and good calibration. There was an overall agreement in the rank order of variables predicting LC incidence and non-LC death, with the top four variables being age, smoking intensity and duration, and body mass index (BMI), however the effects were opposite in nature for the latter. Conclusion: Findings show that the PLCOM2012 risk prediction model accurately identifies higher proportions of LC cases compared to the NLST criteria, however it also selects individuals for LC screening who are at risk of dying from competing causes who will not benefit from screening. Decisions on LC screening among individuals at high-risk of developing LC should be based on an appropriate assessment of each person’s health status and life expectancy, in order to maximize the benefit-harm ratio from LC screening. Keywords: Lung cancer, epidemiology, risk prediction modelling, lung cancer screening, competing causes of death