Latent treatment pattern discovery for clinical processes

A clinical process is typically a mixture of various latent treatment patterns, implicitly indicating the likelihood of what clinical activities are essential/critical to the process. Discovering these hidden patterns is one of the most important components of clinical process analysis. What makes the pattern discovery problem complex is that these patterns are hidden in clinical processes, are composed of variable clinical activities, and often vary significantly between patient individuals. This paper employs Latent Dirichlet Allocation (LDA) to discover treatment patterns as a probabilistic combination of clinical activities. The probability distribution derived from LDA surmises the essential features of treatment patterns, and clinical processes can be accurately described by combining different classes of distributions. The presented approach has been implemented and evaluated via real-world data sets.