The advances in IoT and wearable sensors enable long term monitoring, which promotes earlier and more reliable diagnosis in health care. This position paper proposes a probabilistic method to address the challenges in handling longitudinal sensor signals that are subject to stochastic uncertainty in health monitoring. We first explain how a longitudinal signal can be transformed into a Markov model represented as a matrix of conditional probabilities. Further, discussions are made on how the derived models of signals can be utilized for anomaly detection and classification for medical diagnosis.