1 Universitaetsklinikum Carl Gustav Carus, Internistische Onkologie, Dresden; 2 Medizinische Hochschule, Biometrie, Hannover, Germany
*E-mail: koehne@mkl.med.tu-dresden.de
Watine raises the concern that the laboratory parameters identified in our prognostic model may underestimate the true prognostic value of LDH, albumin, CEA, bilirubin and the aminotransferases, due to the unfortunate under representation of these variables in our model.
We addressed this point in our discussion: "Any multivariate analysis faces the problem that only those variables that have been put into the model come out as important predictors. It is possible that other potential parameters or constellation of variables may better describe and separate the prognosis of patients with metastatic colorectal cancer. Nevertheless, our learning sample clearly identifies three separated risk groups validated in the validation set. Whether other variables including those with a high rate of missing values might add any information to our model remains speculative and at present less important, as we were able to clearly separate patient risk groups with the available data. The difference in median survival in the three groups is also of definite clinical relevance." We would like to remind the readers that our analysis is based on the largest number of patients and variables. Those investigations cited by Watine represent smaller cohorts seldom exceeding 400 patients.
Concern was also raised that differences in the laboratory methodology may have influenced the results. The laboratory values examined are routine in all hospitals. Most, if not all laboratories are subject of strict quality control measures. Nevertheless, different normal ranges are in practice in each laboratory. It is unlikely that different laboratory methods may have played a major role. Indeed, our model was validated in an independent cohort, and not only were the three risk groups reproduced, but also the survival curves of the validation sample were within the 95% confidence interval of the learning set.
The future of prognostic parameters will be to determine potential molecular markers that not only have a statistically significant correlation but also may be causally related to the aetiology of the disease or indicate possible treatment targets.
C.-H. Köhne1* & H. Hecker2
1Universitaetsklinikum Carl Gustav Carus, Internistische Onkologie, Dresden; 2Medizinische Hochschule, Biometrie, Hannover, Germany (*E-mail: claus-henning.koehne@uniklinikum-dresden.de)