Page - 19 - in Applied Interdisciplinary Theory in Health Informatics - Knowledge Base for Practitioners

3. Success factors in process mining of care pathways 3.1. Connected Health Cities The Connected Health Cities (CHC) project in the North of England aims to implement a region-wide LHS through a range of initiatives linking and using health data and sharing insights and best practice (www.connectedhealthcities.org). The approach has included the development of federated data repositories of EHR data as advocated by Friedman, the development of a learning culture for sharing and disseminating knowledge and a focus on care pathways that can be mined, analyzed and improved. Challenges have included: developing architectures and consent models for ethical access to health data; linkage of health data from different sources, standards and variable data quality; engagement with multi-disciplinary teams across multiple organizations; engagement with busy clinicians and already stressed organizations; and the development of better methods for process mining of care pathways. Solutions have included: national level engagement on legal and ethical frameworks; public engagement through a social media campaign (called #datasaveslives) and citizens juries; Trusted Research Environments (TREs) for the secure curation of data; developing experience in multi-disciplinary collaboration; a focus on specific high-impact problem areas; and ClearPath, a novel method for care pathway process analysis that draws on GST and, more generally from a systems thinking approach. 3.2. The ClearPath Method The ClearPath method [11] is an extension of an established process mining method (called PM2, see [14]) that incorporates a stronger systems method of enquiry and produces care pathway simulations that can be used for experimentation and learning. In our work in this area it became evident that a more holistic systems approach was essential to address what have been called “data quality” issues. From the perspective of GST we see health data not as the product of a machine but as the product of a highly complex sociotechnical healthcare system that is evolving, adapting and responding to its environment. We would argue that the failure of “big data” methods in healthcare is due to a failure to apply GST. A conventional approach to healthcare data mining includes complaining about data quality, cleaning data to suit the analysis and assuming that more data means less unknown systemic bias. The reality of healthcare data is that it is messy and incomplete, it can shed some light on the activity of busy clinicians and the administration of healthcare processes but with different systems used differently by different departments, highly variable pathways and moving systems boundaries the only real certainty is that data will be different between systems and over time. Recent advances in process mining recognize this phenomenon as process evolution or “concept drift” and new techniques such as applying sliding time windows to spot changes in process are being developed with some success [15]. Our approach within the CHC project has been to combine process mining of EHR data with a systems approach to enquiry. Following GST, the starting point is to identify and define a system of study that has a clear boundary and a single clear structure. For example we have worked with a number of urgent care departments and have treated each one as a separate discrete system, resisting the temptation to aggregate urgent care data across the region because such an aggregated view would fail GST’s test of what is O.Johnson /GeneralSystemTheoryand theUseofProcessMining to ImproveCarePathways 19