An Analysis of Predictor Variables for Adjuvant Treatment of Breast Cancer
Published in: Cancer Chemotherapy and Pharmacology, v. 2, no. 3, 1979, p. 147-158
Modern computer-based statistical analysis of a well-documented data base can facilitate the selection of breast cancer patients for adjuvant chemotherapy. Traditional selection criteria are dominated by the number of positive axillary lymph nodes found by pathologic examination. However, patients of the same nodal status (0+, 13+, 47+, >7+) are still heterogeneous with regard to risk of metastatic recurrence and benefits of adjuvant treatment. A pilot study of 103 patients of similar nodal status (13+) followed for up to 10 years was undertaken to determine whether data already in the patient's record at the time of pathologic examination and derived from the patient's history as well as from clinical, surgical and pathologic examination could supplement nodal status in predicting disease recurrence. Preliminary processing and screening for promising variables were performed with CLINFO; maximum-likelihood procedures were then used to relate the probability of disease recurrence to those variables that appeared to be significant. Parametric models of hazard rate for the individual patient were employed corresponding to both exponential and Wiebull distributions of disease-free interval. The hazard rate was related log-linearly to a set of prognostic variables, and model parameters were determined by fitting to the data. Factors that favor longer disease-free intervals (in quantitative order of importance) are: (1) Nipple involved clinically at presentation; (2) Lesion had soft or rubbery consistency on palpation; (3) Disease discovered by physician; (4) Homolateral lymph nodes not involved clinically; (5) Margin from tumor to fascia >1 cm; (6) No maternal history of breast cancer; (7) Increasing age of patient; (8) Presence of specialized histology. Based on the findings of this pilot study, a quantitative summary of personal (SK) and institutional experience is developed in which the probability of recurrence for the individual patient and the associated confidence intervals are used to classify patients with regard to risk of recurrence.