A meta-analysis of reported correlations between prognostic factors in breast cancer: does axillary lymph node metastasis represent biology or chronology?

Abstract

A statistical overview of published results on correlations between various prognostic factors in breast cancer was undertaken. A distinction was made between clinical (or anatomical) prognostic factors-namely, axillary lymph node status and tumour size-and eight different biological prognostic factors. The latter included: tumour grade, oestrogen and progesterone receptor status, thymidine labelling index, DNA ploidy, S-phase fraction, epidermal growth factor receptor expression and c-erbB-2 gene amplification (or overexpression). 139 articles were eligible for review which reported a total of 432 individual correlations. A simple form of meta-analysis was employed: the counting method, in which the number of studies achieving a statistically significant correlation or not were counted. For each possible correlation examined, the proportion of studies showing a statistically significant correlation was calculated and an exact binomial 99% confidence interval determined for that proportion. If the 99% confidence interval included 5% (the proportion of correlations that would be expected to be statistically significant if the null hypothesis was true), it was taken as failing to exclude the null hypothesis of a zero correlation, while if it excluded 5% it was taken as rejecting the null hypothesis of a zero correlation. A broad agreement was found among published reports on the existence of a statistically significant correlation between the various biological prognostic factors in breast cancer. Of the 20 correlation examined, 18 had a 99% confidence interval excluding 5%, thus rejecting the null hypothesis of a zero correlation. On the other hand, a completely different result was obtained when reports on possible correlations between lymph node status and tumour size on the one hand and the eight biological prognostic factors on the other were analysed. Of the 16 correlations examined, 13 had a 99% confidence interval including 5%, failing to reject the null hypothesis of a zero correlation. These observations suggest the hypothesis that the prognostic influence of node status and tumour size cannot be explained by an analysis of the biology of breast cancer; and is compatible with the contention that axillary node status is merely a reflection of the relative chronological age of breast cancer.