Cardiopulmonary diseases affect millions of people and cause high costs in health care systems worldwide. Patients should perform regular endurance exercises to stabilize their health state and prevent further impairment. However, patients are often uncertain about the level of intensity they should exercise in their current condition. The cost of continuous monitoring for these training sessions in clinics is high and additionally requires the patient to travel to a clinic for each single session. Performing the rehabilitation training at home can raise compliance and reduce costs. To ensure safe telerehabilitation training and to enable patients to control their performance and health state, detection of abnormal events during training is a critical prerequisite. Therefore, we created a model that predicts the heart rate of cardiopulmonary patients and that can be used to detect and avoid abnormal health states. To enable external feedback and an immediate reaction in case of a critical situation, the patient should have the possibility to configure the system to communicate warnings and emergency events to clinical and non-clinical actors. To fulfill this task, we coupled a personal health record (PHR) with a new component that extends the classic home emergency systems. The PHR is also used for a training schedule definition that makes use of the predictive HR model. We used statistical methods to evaluate the prediction model and found that our prediction error of 3.2 heart beats per minute is precise enough to enable a detection of critical states. The concept for the communication of alerts was evaluated through focus group interviews with domain experts who judged that it fulfills the needs of potential users.