Affiliations: Department of Pathology, University of Turku, Kiinamyllynkatu 10, FIN-20520 Turku, Finland, and Department of Pathology, Jyväskylä Central Hospital, Keskussairaalantie 19, FIN-40620 Jyväskylä, Finland
Abstract: Background Quantitative pathology deals with graded or continuous features. Often these features are prognosticators, and able to tell something about the outcome of disease. Many features are also expected to be predictors of treatment response, especially in association with drug therapy. The basicly intuitive knowledge that the expression of estrogen and progesterone receptors on the surface of cancer cells could reflect good response during tamoxifen or toremifen treatment is much valued in clinical practice, as is the overexpression of erbB2 receptor antigen on the cell surface in potentially predicting a good response during trastuzumab treatment. Material and methods Prognostication, and prediction of therapy outcome is not always perfect, even with multivariate methodology. Because of this, we have looked for a method which could extract most of the prognostic value from basicly quantitative histological or immunohistochemical features. For this purpose we have applied various morphometric tests, and studied various antibodies, including antibodies against cystatin A, bcl-2, erb-B2, and E-cadherin in immunohistochemistry. In immunohistochemistry the staining can be subjectively graded, but usually the immunohistochemical staining index turns out to be the best prognosticator. In our studies we tried to define the prognostically most valuable cutpoints, both in subjective grading and in calculating immunohistochemical staining indices. Cutpoint optimization was done with the help of the khi-square test. After optimization, the prognostic evaluation, in the form of a survival study was carried out. Corresponding studies were carried out on various morphometric features, associated with the development of a morphometric grading system for breast cancer. Results Many quantitative features had a cutpoint or cutpoints which showed highly significant survival difference between patients on the two sides of the cutpoint Some cutpoints were especially significant and associated with prominent peaks in the khi-square curve or ditches in the p-value curve. These cutpoints were chosen as optimal cutpoints for further evaluation. In the following survival analysis the cutpoint presented the best alternative for the survival analysis. Prognostic comparison of the features also appeared more reliable when the optimal cutpoint was known. However, also after these studies the final prognostic conclusions should be based on an independent material, suitable for corresponding analysis. Conclusions A reliable analysis of potential prognosticators or predictors can be based on optimization of the decision cutpoint. After the procedure different prognosticators can be compared reliably. In clinical implementation separate the thinking of group statistics should be changed to thinking of diagnostic statistics.
Keywords: breast cancer, prognosis, diagnosis, grading, khi square, univariate methods, multivariate methods, prognosticators, predictors, survival analysis, histopathology, clinical chemistry