Forecasting individual breast cancer risk using plasma metabolomics and biocontours

Abstract

Breast cancer is a major cause of death for women. To improve treatment, current oncology research focuses on discovering and validating new biomarkers for early detection of cancer; so far with limited success. Metabolic profiling of plasma samples and auxiliary lifestyle information was combined by chemometric data fusion. It was possible to create a biocontour, which we define as a complex pattern of relevant biological and phenotypic information. While single markers or known risk factors have close to no predictive value, the developed biocontour provides a forecast which, several years before diagnosis, is on par with how well most current biomarkers can diagnose current cancer. Hence, while e.g. mammography can diagnose current cancer with a sensitivity and specificity of around 75 %, the currently developed biocontour can predict that there is an increased risk that breast cancer will develop in a subject 2-5 years after the sample is taken with sensitivity and specificity well above 80 %. The model was built on data obtained in 1993-1996 and tested on persons sampled a year later in 1997. Metabolic forecasting of cancer by biocontours opens new possibilities for early prediction of individual cancer risk and thus for efficient screening. This may provide new avenues for research into disease mechanisms.

Keywords: Cancer and health cohort; Chemometrics; Danish diet; Early detection; Metabolomics; Multivariate analysis; NMR; Plasma.

Fig. 1

Resulting ROC curves from a univariate model (left), a model of the 47 lifestyle variables (middle) and the model based on all relevant data (right)

Fig. 2

Regression coefficients of the classification model. Two variables are highlighted as they are discussed below