RESEARCH
Premenopausal breast cancer: potential clinical utility
of a multi-omics based machine learning approach
for patient stratification
Holger Fröhlich
1
&
Sabyasachi Patjoshi
1
&
Kristina Yeghiazaryan
2,3,4
&
Christina Kehrer
3,4,5
&
Walther Kuhn
3,4,5
&
Olga Golubnitschaja
2,3,4
Received: 26 January 2018 / Accepted: 28 February 2018 / Published online: 11 April 2018
#
European Association for Predictive, Preventive and Personalised Medicine (EPMA) 2018
Abstract
Background The breast cancer (BC) epidemic is a multifactorial disease attributed to the early twenty-first century: about two
million of new cases and half a million deaths are registered annually worldwide. New trends are emerging now: on the one hand,
with respect to the geographical BC prevalence and, on the other hand, with respect to the age distribution. Recent statistics
demonstrate that young populations are getting more and more affected by BC in both Eastern and Western countries. Therefore,
the old rule “the older the age, the higher the BC risk” is getting relativised now. Accumulated evidence shows that young
premenopausal women deal with particularly unpredictable subtypes of BC such as triple-negative BC, have lower survival rates
and respond less to conventional chemotherapy compared to the majority of postmenopausal BC.
Working hypothesis Here we hypothesised that a multi-level diagnostic approach may lead to the identification of a molecular
signature highly specific for the premenopausal BC. A multi-omic approach using machine learning was considered as a potent
tool for stratifying patients with benign breast alterations into well-defined risk groups, namely individuals at high versus low risk
for breast cancer development.
Results and conclusions The study resulted in identifying multi-omic signature specific for the premenopausal BC that can be
used for stratifying patients with benign breast alterations. Our predictive model is capable of discriminating individually
between high and low BC-risk with high confidence (>90%) and considered of potential clinical utility. Novel risk assessment
approaches and advanced screening programmes—as the long-term target of this project—are of particular importance for
predictive, preventive and personalised medicine as the medicine of the future, due to the expected health benefits for young
subpopulations and the healthcare system as a whole.
Keywords Predictive preventive personalised medicine
.
Breast cancer
.
Menopause
.
Patient stratification
.
Bioinformatics
.
Machine learning
.
Multi-level diagnostics
.
Biomarker panel
.
Laboratory medicine
Introduction
Breast cancer (BC) epidemic is attributed to the early twenty-
first century as a multi-factorial disease: about two million of
new cases and a half million of deaths are registered annually
worldwide [1]. The highest incidence of BC has been recorded
in Northern and Western Europe, North America, Australia and
New Zealand [2]. Since a couple of years, the USA persistently
demonstrate high incidence rates in BC with by 246,660 new
BC cases and 40,450 BC-related deaths in 2016 [3], which
corresponds to a lifetime risk for one in eight women. The
persisting challenge is prevalent postmenopausal BC frequent-
ly linked to ageing, obesity and metabolic syndrome [4].
* Olga Golubnitschaja
Olga.Golubnitschaja@ukbonn.de
1
Bonn-Aachen International Centre for IT,
Friedrich-Wilhelms-Universität Bonn, Bonn, Germany
2
Radiological Clinic, Rheinische Friedrich-Wilhelms-Universität
Bonn, Bonn, Germany
3
Breast Cancer Research Centre, Rheinische
Friedrich-Wilhelms-Universität Bonn, Bonn, Germany
4
Centre for Integrated Oncology, Cologne-Bonn, Rheinische
Friedrich-Wilhelms-Universität Bonn, Bonn, Germany
5
Centre for Obstetrics and Gynaecology, Rheinische
Friedrich-Wilhelms-Universität Bonn, Bonn, Germany
EPMA Journal (2018) 9:175–186
https://doi.org/10.1007/s13167-018-0131-0
@Phil_Robichaud
@deepthiw
@JoseServera