LB42
Clinico-pathologic and molecular predictors of axillary lymph node status in early stage breast cancer: a mathematical logistic regression predictive model
Mohammed Aleskandarany1, Emad Rakha2, Andrew Green1, Claire Paish2, Ian Ellis1
1University of Nottingham, UK; 2Nottingham University Hospitals NHS Trust, UK
Background
Axillary nodal (LN) stage is the most important prognostic factor in early
stage breast cancer (BC) which has been in rise as a result of wide spread BC
screening. However, surgical procedures for LN staging carry the risk of early
and long-term post-operative morbidity. Therefore, reliable predictors of nodal
status are needed in order to reduce the extent of axillary surgery and its
consequences.
Method
Predictive factors of axillary LN status at the time of primary diagnosis
have been assessed using the well-established clinico-pathologic parameters and
a panel molecular markers in a well-characterised series (n=1130) of primary
operable invasive BC, temporally divided into training (n=730) and validation
(n=400) sets. Potential predictor factors were initially assessed using
univariate analysis, and then a multivariate logistic regression model was
fitted using backward stepwise variable selection in the training set. The
resulting model was subsequently validated utilising the validation set.
Results
Within the training set, the proportion of cancers with positive nodes was
significantly higher with younger age, larger tumour size, higher grade, NST
tumours, definite vascular invasion (VI), ER-, HER2+,
PIK3CA+, and high MIB1 Labelling Index (MIB1LI). Multivariate
logistic regression model indicated that predictors of nodal positivity
included definite VI, tumour size ≥ 2cm, higher MIB1LI, HER2+,
ER-, and higher tumour grade. This model resulted in 86.6% accuracy
in predicting node positive cases, with area under the curve (AUC) =73.1% and
excellent goodness-of-fit (p=0.981). Model cross validation revealed an AUC of
72.3% in the validation set.
Conclusion
In this study, VI and tumour grade were the strongest independent
predictive factors of nodal status in breast cancer patient at the time of
primary diagnosis. Our predictive model, which incorporates VI, tumour grade,
tumour size, histologic tumour type, HER2status and MIB1LI confers an objective
predictive accuracy relative to single predictive factors.
