Machine learning prediction of pathological complete response and overall survival of breast cancer patients in an underserved inner-city population

Table 4 Multivariate analysis of top predictors of pCR (N = 240)

(B)	Predictors for pCR as the outcome	Accuracy	Sensitivity	Specificity	AUC
NN	ER+/HER2−, ER−/HER2+, TS, BPE, Mu, SI, Age, NG, Nstage, T-stage	0.731 ± 0.046	0.799 ± 0.092	0.7360 ± 0.09	0.7540 ± 0.039
RF	ER+/HER2−, ER−/HER2+, TS, BPE, Age, Mu, SI, NG, NS, T-stage	0.720 ± 0.056	0.583 ± 0.147	0.774 ± 0.068	0.7520 ± 0.064
LR	ER+/HER2−, ER−/HER2+, TS, BPE, Mu, SI, Age, NG, T-stage, Nstage	0.720 ± 0.075	0.555 ± 0.118	0.890 ± 0.052	0.7550 ± 0.043
GBR	ER+/HER2−, ER−/HER2+, TS, BPE, Age, Mu, SI, T-stage, NG, Nstage	0.697 ± 0.055	0.666 ± 0.209	0.785 ± 0.054	0.7430 ± 0.084

The performance metrics are mean ± standard deviation for four different models: Neural network (NN), random forest (RF), logistic regression (LR), and gradient boosted regressor (GBR)
TS tumor size, BPE background parenchymal enhancement, Mu multifocal, SI skin involvement, NG Nottingham grade

ISSN: 1465-542X