Study of EHR-mediated workflows using ethnography and process mining methods.

Rapid ethnography and data mining approaches have been used individually to study clinical workflows, but have seldom been used together to overcome the limitations inherent in either type of method. For rapid ethnography, how reliable are the findings drawn from small samples? For data mining, how accurate are the discoveries drawn from automatic analysis of big data, when compared with observable data? This paper explores the combined use of rapid ethnography and process mining, aka ethno-mining, to study and compare metrics of a typical clinical documentation task, vital signs charting. The task was performed with different electronic health records (EHRs) used in three different hospital sites. The individual methods revealed substantial discrepancies in task duration between sites. Specifically, means of 159.6(78.55), 38.2(34.9), and 431.3(283.04) seconds were captured with rapid ethnography. When process mining was used, means of 518.6(3,808), 345.5(660.6), and 119.74(210.3) seconds were found. When ethno-mining was applied instead, outliers could be identified, explained and removed. Without outliers, mean task duration was similar between sites (78.1(66.7), 72.5(78.5), and 71.7(75) seconds). Results from this work suggest that integrating rapid ethnography and data mining into a single process may provide more meaningful results than a siloed approach when studying of workflow.

Process Mining and Ethnography Study of Medication Reconciliation Tasks.

We studied the medication reconciliation (MedRec) task through analysis of computer logs and ethnographic data. Time spent by healthcare providers performing MedRec was compared between two different EHR systems used at four different regional perioperative settings. Only one of the EHRs used at two settings generated computer logs that supported automatic discovery of the MedRec task. At those two settings, 53 providers generated 383 MedRec instances. Findings from the computer logs were validated with ethnographic data, leading to the identification and removal of 47 outliers. Without outliers, one of the settings had slightly smaller mean (SD) time in seconds 67.3 (40.2) compared with the other, 92.1 (25). The difference in time metrics was statistically significant (p<.001). Reusability of an existing task-based analytic method allowed for rapid study of EHR-based workflow and task.

Sharing my health data: a survey of data sharing preferences of healthy individuals.

We interviewed 70 healthy volunteers to understand their choices about how the information in their health record should be shared for research. Twenty-eight survey questions captured individual preferences of healthy volunteers. The results showed that respondents felt comfortable participating in research if they were given choices about which portions of their medical data would be shared, and with whom those data would be shared. Respondents indicated a strong preference towards controlling access to specific data (83%), and a large proportion (68%) indicated concern about the possibility of their data being used by for-profit entities. The results suggest that transparency in the process of sharing is an important factor in the decision to share clinical data for research.

A formal approach to the analysis of clinical computer-interpretable guideline modeling languages.

OBJECTIVE: To develop proof strategies to formally study the expressiveness of workflow-based languages, and to investigate their applicability to clinical computer-interpretable guideline (CIG) modeling languages. METHOD: We propose two strategies for studying the expressiveness of workflow-based languages based on a standard set of workflow patterns expressed as Petri nets (PNs) and notions of congruence and bisimilarity from process calculus. Proof that a PN-based pattern P can be expressed in a language L can be carried out semi-automatically. Proof that a language L cannot provide the behavior specified by a PNP requires proof by exhaustion based on analysis of cases and cannot be performed automatically. The proof strategies are generic but we exemplify their use with a particular CIG modeling language, PROforma. To illustrate the method we evaluate the expressiveness of PROforma against three standard workflow patterns and compare our results with a previous similar but informal comparison. RESULTS: We show that the two proof strategies are effective in evaluating a CIG modeling language against standard workflow patterns. We find that using the proposed formal techniques we obtain different results to a comparable previously published but less formal study. We discuss the utility of these analyses as the basis for principled extensions to CIG modeling languages. Additionally we explain how the same proof strategies can be reused to prove the satisfaction of patterns expressed in the declarative language CIGDec. CONCLUSION: The proof strategies we propose are useful tools for analysing the expressiveness of CIG modeling languages. This study provides good evidence of the benefits of applying formal methods of proof over semi-formal ones.