Wearable devices performing advanced bio-signal analysis algorithms will foster a revolution in healthcare provision of chronic cardiac diseases. In this context, energy efficiency is of paramount importance, as long-term operations must be performed within an extremely tight power budget. Aggressive voltage scaling is an effective strategy towards this goal, but presents the adverse effect of causing bit-flip errors in the memory sub-system. The application of error correction codes to the whole memory results in large area and energy overheads. To cope with this challenge, in this paper we propose to selectively protect only data presenting high significance, i.e.: highly impacting the end-to-end quality of service. The resulting heterogeneous error protection scheme allows substantial energy savings with respect to complete protection, while resulting in negligible results degradation of cardiac bio-signal processing algorithm