Estimation of surgeon effects in the analysis of post-operative colorectal cancer patients data

There has been increasing interest in the assessment of surgeon effects for survival data of post-operative cancer patients. In particular, the measurement of surgeon's surgical performance after eliminating significant risk variables is considered. The generalized linear mixed model approach, which assumes a log-normal-distributed surgeon effects in the hazard function, is adopted to assess the random surgeon effects of post-operative colorectal cancer patients data. The method extends the traditional Cox's proportional hazards regression model, by including a random component in the linear predictor. Estimation is accomplished by constructing an appropriate log-likelihood function in the spirit of the best linear unbiased predictor method and extends to obtain residual maximum likelihood estimates. As a result of the non-proportionality of the hazard of colon and rectal cancer, the data are analyzed separately according to these two kinds of cancer. Significant risk variables are identified. The 'predictions' of random surgeon effects are obtained and their association with the rank of surgeon is examined.