Using Multilevel Models to Explain Variation in Clinical Practice: Surgeon Volume and the Surgical Treatment of Breast Cancer.

Abstract

To investigate the relationship between surgeon caseload and surgery type, and variation in the surgical treatment of early stage breast cancer patients in Alberta, Canada. All women diagnosed with stage I to III breast cancer in Alberta from 2002 to 2010 were identified from the Alberta Cancer Registry. Type of surgery, surgeon (anonymized), and hospital were obtained from provincial physician claims data. Multilevel logistic regression with surgeons and hospitals as crossed random effects was used to estimate adjusted odds ratios (OR) of receiving mastectomy by surgeon volume. Empirical Bayes estimation was used to estimate adjusted OR for individual surgeons and hospitals. Mastectomy was found to be inversely related to surgeon volume among stage I and II patients. Patients whose surgery was conducted by a low-volume surgeon had twice the odds of receiving mastectomy as those that had surgery performed by a very high-volume surgeon (stage I OR 2.36, 95% confidence interval 1.40, 3.97; stage II OR 1.96, 95% confidence interval 1.13, 3.42). OR of mastectomy varied widely by individual surgeon beyond the variation explained by the factors investigated. Surgeon characteristics, including surgeon volume, are associated with surgery type received by breast cancer patients in Alberta. Significant variation in the likelihood of breast-conserving surgery (BCS) by surgeon is concerning given the potential benefits of BCS for those who are eligible.