Skip to main content
Figure 3 | BMC Nephrology

Figure 3

From: Would artificial neural networks implemented in clinical wards help nephrologists in predicting epoetin responsiveness?

Figure 3

Prediction of the epoetin dose required to reach the haemoglobin target. ROC curves plotting sensitivity against 1 minus specificity for a epoetin dose cut-off of 100 IU/Kg/week in the prediction of the dose required for an individual patient to reach the haemoglobin target of 11.5 g/dL obtained from the best performing linear regression (dotted line; using as input variables weight and ferritin) and the best performing ANN (continuous line; using as input variables weight, age, presence or absence of an impaired left ventricular ejection fraction, serum creatinine and ferritin). The areas under the curves, the 95% confidence intervals and the significance P for the linear regression and the ANN were respectively: 0.491 (0.416–0.565), P:n.s. and 0.728 (0.663–0.794), P < 0.001 (P < 0.001 for the difference between the two curves).

Back to article page