“If you can’t describe what you are doing as a process, you don’t know what you’re doing.”
W. Edwards Deming
Edwards Deming emphasized the importance of intimately knowing a business process before attempting to improve it with a quality improvement (QI) initiative. A thorough understanding of the business workflow requires process discovery techniques that are accurate, comprehensive, and scalable. Typical methods of process discovery include observational time & motion studies, stakeholder interviews, and process workshops. However, these are often slow, costly, and difficult to scale to a national level. As a result, fewer QI interventions are attempted, and QI interventions may be less effective than expected.
In many health Information Technology (IT) deployments and policy implementations, business workflow plays a critical role. There is an opportunity to process-mine electronic health record (EHR) data to augment traditional methods of process discovery, and to identify workflow concurrence and performance issues to support QI initiatives. Process mining is non-intrusive, low-cost, and highly scalable. In 2015, WBB carried out process mining on EHR transactional data from emergency room (ER) facilities as a proof of concept. As a result, WBB observed several kinds of variations that were suitable for QI interventions that could enhance national oversight, improve EHR use, and identify potential patient safety issues.
The facilities varied in naming conventions for common ER transfer targets. For example, at one facility “Sent to Urgent Care Clinic” was used for the same clinical event as “Sent to Nurse Eval/Drop-in Clinic” at another facility. Differences in nomenclature create an obstacle to comparing processes between facilities, and can impede development of best practices. This data enabled a QI initiative to focus on reducing terminology variation and enhance governance.
WBB observed process loops in the inferred models that reflected the way the EHR application was used. In Figure 1, paths from “Admitted” to subordinate admission locations are correctly displayed. However, return paths are shown indicating that in some cases, subordinate admissions were being entered in the EHR application before the initial admission step. A functionality issue in the EHR system made it possible to select events out of sequence. A training initiative and changes to the EHR configuration could eliminate this variation and improve EHR use.
Figure 1. ED-1 Process loops
The process models also showed cases in which there were higher than two standard deviations more transactions than the norm, which could be a valuable trigger for quality audits. Where the expected average transaction count is between three and five, figure 2 shows a patient with 10 events. Cases with abnormally low or high event counts may reveal clerical errors, or process gaps that do not adequately address some patient situations. QI audits should look for potential patient safety issues as a primary objective, but also identify where process gaps or clerical error could be eliminated.
Figure 2. “Pinball” patient
By having accurate workflow models and associated EHR datasets, we can perform deeper assessments that compare the actual behavior to the desired model. The assessments can be used to support targeted QI initiatives at national, regional, and facility levels. QI initiatives may include detecting variances, initiating remedial interventions, and monitoring workflow models to identify and quantify the effects. Likewise, process mining can enable us to quickly detect effects of a QI intervention at low cost and without interfering in the clinical workflow.
