Temporal mining of integrated healthcare data: Methods, revealings and implications

The increasing integration and availability of healthcare data triggers new opportunities for an adequate discovery and use of temporal patterns to support medical decisions. However, adequate data mappings are still lacking for the application of temporal mining methods over healthcare databases. Additionally, existing methods commonly do not allow for flexible knowledge representation to guide time partitioning and to support the use of multiple temporal granularities. Finally, existing predictors are not able to rely on these temporal patterns. In this paper, cyclic rules that integrate multiple medical aspects are discovered using an expressive data mapping. Second, temporal constraints are proposed to guide the discovery of patterns under multiple time scales. Third, a classifier relying on cyclic patterns is defined to predict healthcare conditions. The conducted experiments hold evidence for the utility and efficiency of the proposed methods in characterizing and predicting integrated healthcare profiles