Physicians use colonoscopy to detect recurrent colorectal cancer for patients who have had colorectal cancer surgery. Clinical studies show both underuse and overuse of this test among survivors of colorectal cancer in relation to guideline recommendations. Yet few studies have examined referrals and test findings. To identify determinants of colonoscopy referral for colorectal cancer survivors and ex-post test outcome, Chapter 1 examines Texas cancer registry data linked with Medicare claims from 2000 to 2009 for colorectal cancer survivors with a history of resection surgery. Risk-adjusted regression analyses are used to measure the association of patient, referring physician and clinical factors with referrals and test results. Intestinal symptoms, the timing of referrals, and referring physician specialty are associated with referral decisions and test results. Gastroenterologists are more likely to refer patients for colonoscopy than oncologists, surgeons or primary care physicians, but their rates of positive test results are the lowest. The discrepancy between referral decisions and test results suggests suboptimal test use.

Chapter 2 studies referral patterns for colonoscopies and applies cost-benefit analysis to examine whether this test has been overused or underused among patients with a cancer history. A key aspect of the analysis is that the ex post value of colonoscopy is partially observable in insurance claims records based on whether the test identifies recurrent cancer. Estimating the physician-specific parameter representing physicians' practice styles, I find that referral patterns exhibit physician-level heterogeneity. The percentage of physicians who overuse colonoscopy varies with physician specialty: A significant number of gastroenterologists overuse colonoscopy, whereas a much lower portion of oncologists and primary care doctors overuse it. These findings illustrate the need for well-targeted health care policies to curb growing health care costs.

Chapter 3 measures the relationship between hospital volume and patient mortality for six cancer operations (colectomy, esophagectomy, pancreatic resection, pneumonectomy, pulmonary lobectomy, and rectal resection). Analyzing hospital discharge data from Florida, New Jersey, and New York for the 12 years 2000 to 2011, we find that the statistical significance of hospital volume depends critically on the regression model used: for the data, logistic and random-effects models suggest that higher volume is associated with lower mortality, but fixed-effects models do not. Model-specification tests support either the fixed-effects or random-effects model, depending on the surgical procedure; the basic logistic model is always rejected. These findings illustrate the importance of testing alternative model specifications, especially when drawing policy conclusions about promoting high-volume facilities.