The derivation of treatment processes based on medical patient data and the comparison of those with existing guideline for the treatment of patients is a central topic and follows the aim of the project of a conjoint view from the micro and macro levels of medical information. In first publications [Dunkl2011, Dunkl2012] a methodology for compliance checking has been introduced which is still a central topic within the project. The necessary data sources for analysis have been identified and analyzed for usability [Dunkl2012]. The presented methodology also provide for the generation of synthetic log data [Dunkl2011]. For that purpose parts of the guideline have been modeled as a petri net to produce synthetic log data using CPN-Tools. This research direction will be followed in the course of the project.
The topic of data integration and quality improvement was addressed parallel to the other topics. To be able to derive models from data using techniques of process mining it is necessary to integrate data from different data sources and to transform it for further processing. For that purpose a data integration model for process mining was iteratively developed [Dunkl2012]. Due to low data quality referring to a process management context of medical data the derived models were not usable for compliance checking. As an essential outcome we were able to show the potential of process mining as an auditing tool for medical experts to find impossible or seldom situations in treatment processes [Dunkl2012]. At the moment we are continuing the work on a data integration layer for real world heterogeneous data with partly insufficient data quality. Diverse concepts for preprocessing, filtering and enrichment of data in meta information are developed to support data integration and to improve data quality [Dunkl2013]. A first Prototype (PTDoc) based on the data integration model has been used to recollect 10 patient cases [Dunkl2012].
The increasing amount of healthcare data in electronic form constitutes a knowledge base of high value for medical research and a optimization of individual patient treatment processes.
In a first application area the research cluster set the goal to evaluate the adherence of evidence based guidelines using the treatment of cutaneous melanoma. Till now we got access to diverse data sources through our cluster cooperation, e.g. data from Statistics Austria, data from the Main Association of Austrian Social Security Institutions, data from actual studies at the university hospital for dermatology, etc. Intensive analysis on the mentioned data sources shows that direct process mining is not applicable due to missing data structure. Especially the temporal reference of the actual existing Austrian data structures are not detailed enough. Additionally a separation of patient data from office-based physicians and the hospital sector is noticeable. Nevertheless we are optimistic to have soon data in appropriate quality for process mining to compare real treatment processes with recent melanoma guidelines.
Data integration and analysis: In the area data quality improvement we want to embed established as well as new concepts into the data integration model. The concepts will be evaluated with the use case of melanoma. to collect the necessary information for this use case a close cooperation between the department of dermatology and computer science is required.
Pathodanymics: A connection between clinical protocols and the medical logic of diagnose and therapy processes can be established through the formulation of a pathodynamic model. In this model a (in general multidimensional) phase space of a biographic system (in the actual domain typically a patient) will be mapped to an abstract discrete sequence of clinical relevant states. Starting from the concept of homeostasis "disease" and "healing" are interpreted as aberrancy (noticeable from the symptomatic) from the return (by therapeutic intervention) to the homeostatic norm. Diagnosis are made in the light of given causal knowledge of therapeutic effects (viewed normative) in terms of a maximization of the healing process (return to homeostasis). The research of the cluster tries to explore the structures and conditions of modeling as a basis for targeted investigation of clinical process data and their structural analysis (process structure mining). Therefor selected use cases as skin disease like psoriasis, melanoma but also chronic diseases like diabetes mellitus are used.
From a medical viewpoint the remaining project runtime should be used for further medical use cases with high relevance in terms of health policy for process mining. Section criteria should be more than 2 percent of prevalence and a chronic characteristic. Such use cases from the dermatology domain are e.g. the psoriasis vulgaris and the atopic dermatitis, from the domain of internal medicine diabetes mellitus and chronic inflammatory musculoskeletal disorders as well as chronic decease of the vascular system. The medical university / general hospital Vienna provide access to these patients and data can be collected sufficiently respectively used from existing data sources.