Methods of speed control for implantable rotary blood pumps (iRBPs) are vital for providing implant recipients with sufficient blood flow to cater for their physiological requirements. The detection of pumping states that reflect the physiological state of the native heart forms a major component of such a control method. Employing data from a number of acute animal experiments, five such pumping states have been previously identified: regurgitant pump flow, ventricular ejection (VE), nonopening of the aortic valve (ANO), and partial collapse (intermittent [PVC-I] and continuous [PVC-C]) of the ventricle wall. An automated approach that noninvasively detects such pumping states, employing a classification and regression tree (CART), has also been developed. An extension to this technique, involving an investigation into the effects of cardiac rhythm disturbances on the state detection process, is discussed. When incorporating animal data containing arrhythmic events into the CART model, the strategy showed a marked improvement in detecting pumping states as compared to the model devoid of arrhythmic data: state VE--57.4/91.7% (sensitivity/specificity) improved to 97.1/100.0%; state PVC-I--66.7/83.1% improved to 100.0/88.3%, and state PVC-C--11.1/66.2% changed to 0.0/100%. With a simplified binary scheme differentiating suction (PVC-I, PVC-C) and nonsuction (VE, ANO) states, suction was initially detected with 100/98.5% sensitivity/specificity, whereas with the subsequent improved model, both these states were detected with 100% sensitivity. The accuracy achieved demonstrates the robustness of the technique presented, and substantiates its inclusion into any iRBP control methodology.