The continuous increasing needs in telemedicine and healthcare, accentuate the need of well-adapted medical alert systems. Such alert systems may be used by a variety of patients and medical actors, and should allow monitoring a wide range of medical variables. This paper proposes Tempas, a personalized temporal alert system. It facilitates customized alert configuration by using linguistic trends. The trend detection algorithm is based on data normalization, time series segmentation, and segment classification. It improves state of the art by treating irregular and regular time series in an appropriate way, thanks to the introduction of an observation variable valid time. Alert detection is enriched with quality and applicability measures. They allow a personalized tuning of the system to help reducing false negatives and false positives alerts