Type of Document Dissertation Author Deppen, Stephen Andrew Author's Email Address firstname.lastname@example.org URN etd-06062013-113320 Title Development of a lung cancer prediction model for surgeons and factors affecting its national application Degree PhD Department Epidemiology Advisory Committee
Advisor Name Title Eric L. Grogan Committee Chair Jeffrey Blume Committee Member Melinda Aldrich Committee Member Melissa McPheeters Committee Member Pierre P. Massion Committee Member Keywords
- lung cancer
- predictive model
Date of Defense 2013-05-22 Availability unrestricted AbstractLung 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.
Filename Size Approximate Download Time (Hours:Minutes:Seconds)
28.8 Modem 56K Modem ISDN (64 Kb) ISDN (128 Kb) Higher-speed Access Deppen.dissertation06062013.pdf 4.14 Mb 00:19:10 00:09:51 00:08:37 00:04:18 00:00:22