Table 5. Multivariate logistic regression modeling.

Variable	OR (95% CI)	P value
Number of anthracycline cycles	1.64 (1.17–2.30) †	0.004
BSA (m2)	2.08 (1.36–3.20) ‡	0.001
Elevated BP (≥140/90mmHg)	5.36 (1.73–17.61)	0.006
Trastuzumab therapy	3.35 (1.18–9.51)	0.02
Model performance indices
Area under ROC curve	0.78 (0.70–0.86)
Akaike Information Criterion (AIC)	145.0
Positive predictive value, PPV (<5% FPR)	62.5%
Negative predictive value, NPV (<5% FPR)	84.1%

FPR, False positive rate.

†OR per cycle.

‡OR per standard deviation (0.16 m2) increase in body surface area (BSA). Analysis undertaken using the 161 subjects who received epirubicin and had complete dose and blood pressure (BP) data, of whom 33 were in the cAC group. The input variables were: epirubicin ≥450mg/m2 (yes/no), number of anthracycline cycles, height, weight, BSA (as continuous variable and >2.00), BP ≥140/90mmHg (yes/no) and BP<100/60mmHg (yes/no), Trastuzumab (yes/no), and Interval from final anthracycline cycle. For calculation of model score: yes = 1 and no = 0. Model 1 score = 0.4955 x (number of cycles) +4.552 x (body surface area) +1.679 x (blood pressure≥140/90) +1.209*(Trastuzumab). The same variables were retained in the model if baseline LVEF was used as an input, thus confirming minimal confounding effect of regression towards the mean. A 2nd analysis performed to exclude any influence of the time to follow-up (by omitting both the interval and Trastuzumab treatment status) performed similarly: area under the ROC curve of 0.74 (0.65–0.83), positive predictive value (PPV) 64.7%, negative predictive value (NPV) 84.7%.