Deppen, Stephen Andrew.
Development of a lung cancer prediction model for surgeons and factors affecting its national application.
Degree: PhD, Epidemiology, 2013, Vanderbilt University
Lung cancer is deadly, killing more people than breast, colon and prostate cancer combined. Surgeons evaluating patients for lung cancer face a dilemma: to operate and subject the individual to operation associated mortality and morbidity or not operate and possibly miss early diagnosis and treatment. No models designed for surgeons evaluating lung lesions. We successfully estimated the TREAT model. A model designed for surgeons with an internally validated AUC of 0.87 and Brier score of 13.
If the TREAT model is applied to a national population, its accuracy may decrease due to local conditions. To determine the possible extent of such variation, benign disease prevalence after lung surgery was estimated using 2009 Medicare hospital discharge data. Significant variation in benign disease prevalence between states was observed with a low of 1.3% in Vermont and a high of 25% in Hawaii. The causes for this observed variation are unknown. Residence in a county with high fungal lung disease prevalence was not associated with increased likelihood of benign disease.
FDG-PET scan variance was observed in the national ACOGOS Z4031 trial. FDG-PET sensitivity (82%) and specificity (31%) were significantly lower than in previous published studies. Granuloma occurred in 68% of the false positive FDG-PET scans and sensitivity varied significantly between sites. Scan accuracy increased with increasing lung lesion size. Whether the observed variation is caused by practice variation, referral patterns, fungal lung disease, or other factors is unknown.
A meta-analysis examined FDG-PET accuracy to diagnose lung lesions sought to determine if other researchers had observed variance in FDG-PET accuracy. Seven studies reported false
positive scans arising from granulomas caused by infectious lung disease. Specificity of those studies was 59%, significantly lower than the specificity (77%) observed in the remaining 53 studies. Studies whose mean lesion size was less than or equal to 20 mm had significantly lower sensitivity than studies conducted in larger lesions.
The TREAT model shows clinical promise and should be externally validated. The causes of observed variation in benign disease prevalence and FDG-PET accuracy should be investigated with particular attention made to measuring infectious disease exposures that cause granulomas.
Advisors/Committee Members: Pierre P. Massion (committee member), Melissa McPheeters (committee member), Jeffrey Blume (committee member), Melinda Aldrich (committee member), Eric L. Grogan (Committee Chair).
Subjects/Keywords: lung cancer; diagnosis; surgery; predictive model; meta-analysis; epidemiology
to Zotero / EndNote / Reference
APA (6th Edition):
Deppen, S. A. (2013). Development of a lung cancer prediction model for surgeons and factors affecting its national application. (Doctoral Dissertation). Vanderbilt University. Retrieved from http://hdl.handle.net/1803/12484
Chicago Manual of Style (16th Edition):
Deppen, Stephen Andrew. “Development of a lung cancer prediction model for surgeons and factors affecting its national application.” 2013. Doctoral Dissertation, Vanderbilt University. Accessed May 09, 2021.
MLA Handbook (7th Edition):
Deppen, Stephen Andrew. “Development of a lung cancer prediction model for surgeons and factors affecting its national application.” 2013. Web. 09 May 2021.
Deppen SA. Development of a lung cancer prediction model for surgeons and factors affecting its national application. [Internet] [Doctoral dissertation]. Vanderbilt University; 2013. [cited 2021 May 09].
Available from: http://hdl.handle.net/1803/12484.
Council of Science Editors:
Deppen SA. Development of a lung cancer prediction model for surgeons and factors affecting its national application. [Doctoral Dissertation]. Vanderbilt University; 2013. Available from: http://hdl.handle.net/1803/12484