The current approach to train Cardiopulmonary Resuscitation (CPR) is to employ a mannequin device replicating the physical properties of a real human head and torso. This aims to ensure a correct transfer of the cardiac massage location, amplitude and frequency in a real situation. However, this type of training does not replicate the stress that may be elicited in the presence of a real victim ; this may result in reduced CPR performances or even errors. Virtual Reality (VR) may alleviate this lack by adding visual immersion with a Head-Mounted Display (HMD) so that the trainee is cut from the potential distractions of the real surrounding and can fully engage in a more faithful training scenario. However, one must ensure in the first place that using this technology maintains the quality of the CPR. Hence, we have conducted an experimental study to evaluate the potential of visual immersion in such a training context (limited to the cardiac massage). One important requirement was to ensure a correct hand tracking while executing the standard CPR two-hands pose. In the present paper we describe first how we assessed a simple approach using two HTC-Vive trackers. Results show that the proposed minimal setup based on a single hand tracking is validated for frequency and, with correction, for amplitude. Then, to assess the quality of the training, we performed an evaluation study considering the following two factors: Haptic feedback with the mannequin device (with/out) and Real-time Performance feedback (with/out) in the HMD. We observed that the visually immersive experience proposed in this paper delivers a sufficient level of spatial presence, involvement and agency. Integrating the real CPR mannequin in VR has a significantly positive impact on the massage performance quality whereas displaying the real-time performance in the virtual environment tends to be only useful for the frequency when no mannequin is used.