Using Test Results to Compare the Efficiency of Physicians in Identifying Cancer Recurrence

Physicians use colonoscopy to detect recurrent colorectal cancer for patients who have had colorectal cancer surgery. Although there exist evidence-based guidelines for follow-up colonoscopy, clinical studies show both underuse and overuse of this test among survivors of colorectal cancer. Patient-level sociodemographic and clinical factors have been studied to investigate the variation in use of follow-up colonoscopy, but less attention has been paid to factors at the physician level. In particular, no studies have analyzed physicians' referral patterns to explain the variation in test use, even though referral is the main way most survivors receive this test.

This paper 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. Following previous literature, I assume that a physician refers a patient for a colonoscopy if her perceived probability of cancer recurrence is greater than a physician-specific probability threshold. The variation in thresholds across physicians shows physicians' hetergeneous reactions to perceived cancer risk. Moreover, I use an alternative assumption that a physician refers a patient for a colonoscopy if the net benefit of the test exceeds a physicians-specific dollar-value threshold. The sign of this dollar-value threshold indicates which physicians are overusers or underusers.

There may exist other physician-level unobserved factors which may confound the estimation of the thresholds of each physician. In this paper, I consider two common physician-level factors: differences in patients' risk across physicians, and differences in physicians' diagnostic skills. A key aspect of the analysis is that the ex post value of colonoscopy is partially observable based on whether the test identifies recurrent cancer. The ex post test result is used to control for these unobserved physician-level confounders to obtain consistent estimates of thresholds of each physician.

Using Texas cancer registry-Medicare linked data for the years from 2000 to 2009, I find that referral patterns exhibit physician-level heterogeneity. Physician specialty is the main determinant in explaining the variation in thresholds. The percentage of physicians who overuse colonoscopy also varies with physician specialty. A significant portion of gastroenterologists overuse colonoscopy, whereas a much lower portion of oncologists and primary care physicians overuse it. Finally, physicians with a high threshold, i.e. physicians who are less aggressive in utilizing the test, tend to make more guideline-adherent referrals.

These findings suggest that health policy interventions targeting all medical providers may be suboptimal because they do not provide well-targeted incentives for physicians who overuse or underuse medical resources. Rigorous individual-level measures, such as a physician's testing threshold, may help stratify medical providers in terms of their relative efficiency. These measures can enhance the efficiency of health care provision in accountable care organizations and optimize the Medicare physicians' reimbursement scheme.