The influence of risk factors for breast cancer on mammographic patterns: a multivariate analysis

Abstract

The relationship between the mammographic parenchymal patterns and known risk factors for breast cancer was investigated in the population of Merseyside. Two sample populations of self-referred women were studied. The ‘index’ population was drawn from the employees of a number of major companies on Merseyside and comprised 5413 women aged 35–60 years. The ‘test’ population of 2224 women aged 35–65 years was drawn from the wider Merseyside population. Full histories were recorded on questionnaires and the women were examined by a nurse examiner who recorded her findings on the same proforma. Mammography was carried out and the films were then coded according to Wolfe's classification of mammographic parenchymal patterns. All this data was transferred to the University of Liverpool mainframe computer. Analysis of the data was undertaken using the Statistical Package for the Social Sciences-X (SPSSX). The frequency of the occurrence of the relevant factors within both populations was determined. The relationship between the various factors and the parenchymal patterns in the index population was calculated using crosstabulation analysis. Those factors which showed a statistically significant relationship to the breast patterns were then entered in to a multivariate analysis using a ‘logit’ regression form of loglinear regression. The equations derived by this method were tested by calculating predicted values for the proportion of given breast patterns in the subgroups of the population with specified characteristics and comparing the predicted values with the actual values within the population. Those equations which were found to be satisfactory were then tested in the ‘test’ population. From this study it was deduced that the parenchymal patterns are associated strongly with body build. Parity and a history of breast biopsy also have a significant effect. Age at first pregnancy and a history of breast feeding have a minor effect on some of the patterns. The equations for the DY pattern and for ‘high risk’ versus ‘low risk’ patterns are accurate predictors of the proportion of these patterns in the population. The equations for the P 2 pattern are poor predictors of the proportion of P 2. The existence of a relationship between risk factors for breast cancer and the mammographic parenchymal patterns has been confirmed.