Abstract
Data on 2170 consecutive patients with breast cancer submitted to curative surgery with or without combined radiotherapy in the period 1968–1972 at the National Cancer Institute of Milan (Italy) and at the University of Texas M.D. Anderson Cancer Center of Houston (Texas, USA) were analyzed to evaluate the prognosis of breast cancer patients after loco-regional treatment only and to verify if different prognostic factors have the same relevance. Forty-four percent of patients were alive without evidence of disease at the end of the follow-up in both centers: 14% of patients treated in Milan died without evidence of breast cancer with an intercurrent disease, whereas the death rate for intercurrent disease was 27 % in Houston. Thirty-seven percent of the patients in Milan and 26% of the patients in Houston died from breast cancer. A considerable percentage of patients (23.4 % in Milano, 38.2% in Houston) had one or more of the required items not specified in the clinical chart. Since the lack of information was considered a possible source of bias, the series were divided into two groups: the first collecting patients with all information available, the second gathering patients with at least one of the required items missing. The latter group was defined « unknown ». Multivariate analysis of survival, carried out by means of Cox's regression model, showed that mortality of these patients for all causes was significantly affected by the following criteria: status of regional nodes (P = 2 × 10−18), unknown (P = 10−9), maximum diameter of primary tumor (P = 7 × 10−10), age of the patients (P = 10−4), site of primary (P = 0.01), and Center (P = 0.04). A significant interaction was found between center and a) age of the patients, b) menopausal status and c) unknown. The relative P values were 6 × 10−7 for age and center, 8 × 10−3 for menopausal and center, 3 × 10−2 for unknown and center. Multivariate analysis of breast cancer mortality was significantly affected by: status of regional nodes (P = 10−18), diameter of primary (P = 5 × 10−14), unknown (P = 2 × 10−13), center (P = 2 × 10−6), site of primary (P = 0.002), and age of the patients (P = 0.03). The same significant interaction as for mortality from all causes was found. It is concluded that comparability of results obtained in different institutions may be dependent on the standardization and availability of patients data. The lack of information may introduce considerable biases in the evaluation of results, as was shown by the relevance of the variable unknown on mortality for all causes and for breast cancer. As regards the number of positive lymph nodes as a criterion to define subgroups of patients with different risks of death, we were unable to identify a definite breaking point. The most widely used categorization of this variable (1–3 positive nodes and 4 or more positive) was not supported by our data.
Get full access to this article
View all access options for this article.