Chronic diseases require long-term medical care. Recent advancements in mHealth (mobile health) enables passive, continual and unobtrusive gathering of data with devices such as smartphones and wearables. Detection of chronic disease progression is complicated due to the difficulties of collecting high quality labeled data in real-life monitoring. Interacting with a continuous data stream is laborious and causes patients to drop out. Patient-provided labels are also subjective and prone to bias (e.g. due to memory errors). Furthermore, identifying discrepancies in individual patients is challenging as disease and human behavior changes over time, measurements are person dependent, and patients might start monitoring when they are already in an anomalous situation rendering personal discrepancies less useful. Lastly, it is difficult to learn the causal structures in data to comprehend discrepancies due to the existence of many confounders. To realize prevention with long-term health monitoring, a generic framework will be created for chronic disease monitoring. The following three research objectives are tackled 1) Personal anomaly detection for disease progression using unlabeled mHealth data; 2) Detecting discrepancy using inter and intra-personal features; and 3) Causal discovery in real-life settings through distributed heuristic approaches.