Uncovering Fundamental Flaws in the Design of Work within Healthcare Systems

Related on The Signal Room: Good People Are Quietly Quitting. Related from HDSC: Why Healthcare Data Strategy Fails Without Operational Alignment.For 40 minutes of every shift the charge nurse tries to reconcile staff positions with census. The staffing tool plans based on midnight data. By 7 AM, 3 patients are added, 2 are transferred, and 1 is discharged. The charge nurse improvises the staffing plan based on memory, experience and phone calls. This is the work behind the work, and it happens in every unit, every shift, in almost every hospital in the US.

Before healthcare organizations massively adopted Electronic Health Records and AI systems, they had a working design problem. This is a result of regulations, technology limits, staffing, and memory systems. Over time, workarounds become standard operating procedures, and work design problems become cemented.

The American Nurses Association says that less than 40% of the time a nurse works is spent on direct patient care. Most of that time is spent on data entry, supply runs, equipment problems, and poorly organized communication. All of these tasks are necessary, and a lot of time goes to these tasks because they were poorly organized, not inefficiently carried out.

A study published in BMJ Quality and Safety has documented workflow patterns in acute care and found that clinical workflows are characterized by high levels of interruptions, task switching, and undocumented redundant activities. These activities are not captured by process documentation. Quality and safety are adversely affected by discrepancies between documented workflows and actual workflows.

The Joint Commission has cited communication breakdowns as a primary cause of sentinel events. Many of these communication breakdowns are not individual performance breakdowns, but rather breakdowns in the communication systems themselves. Handoff processes, escalation pathways, and inter-disciplinary coordination schemes have been designed to handle a volume and complexity that is not reflective of the current care milieu.

The issue with AI, when it is integrated into a healthcare workflow, is that it integrates the design of that workflow. If the workflow has redundant steps, the model optimizes around those redundant steps. If the workflow captures information inconsistently, the model learns from that inconsistency and optimizes around it. If the workflow was designed for a context that no longer exists, the model then produces outputs that are a reflection of that disconnect.

This is the primary reason why many AI implementations may generate technically accurate outputs, but those outputs are not useful to clinicians. Within the contours of the data that was generated and used to train the model, the model is indeed correct. The issue, however, lies in the fact that the data was derived from a workflow that does not capture the actual work.

Research from the Institute for Healthcare Improvement has looked at the gap between what the clinicians aim to do versus what they actually achieve by identifying gaps in the Work-as-Done versus Work-as-Imagined frameworks. These gaps remain invisible to leadership as well as the workflow monitoring data systems. Clinicians operate in Work-as-Done whereas AI tools that are trained on workflow data inherit Work-as-Imagined. This gap leads to the erosion of trust and stagnation in adoption.

The health systems that have successfully gained sustained value from the deployment of AI technologies have one thing in common; they prioritized designing the work before they automated it. These healthcare systems thoroughly mapped the actual workflow by means of direct observation of the workflow, interviews with the clinicians, and analysis of the workflow with time measurements. These systems identified steps that resulted from the limitations of the systems (as opposed to the clinical needs) and removed or restructured these steps before the AI was embedded into the process.

The Agency for Healthcare Research and Quality has documented workflow redesign as the first step that needs to be undertaken if one wishes to successfully implement health information technologies (HIT). These studies have shown that un-redesigned workflows lead to low rates of adoption, high rates of workaround behaviors, and low return on investment from the deployed technologies compared to technologies that are implemented into restructured workflows.

Lean methodology has been employed and adapted by various healthcare organizations including the ThedaCare and Virginia Mason Medical Center in order to improve the clinical and operational processes. There is substantial overlap between the principles of Lean methodology and efficient deployment of AI technologies. Both approaches require a candid reflection on the actual execution of the work (as opposed to how it is documented or believed).

Drafting and redesigning systems is a part of a design problem. There is a lot of time, analysis, and management of change needed to redesign processes, and the redesign is expensive and does not provide immediate measurable benefits. Clinicians are asked to describe how they work, and not in the theoretically correct and expected way, in order to determine if the redesign is needed. Redesigning processes, which have not changed in years, is necessary to eliminate inefficiency, a burden, and a poor use of clinical time.

Out of interest to health system leaders, the benefit of redesigning processes is not automated systems and the logic model to design processes. The benefit of redesigning processes is the work of system design, work that the system is going to touch. The leaders that automate systems and processes that are not well designed will not provide continuous improvement in value. The leaders that have honest discussions about what is being done, not what is supposed to be done, will enable the improvement of processes, and subsequently, the intelligent use of AI.

American Nurses Association has done research about Nurses and how they spend their clinical work shifts. Workflow interruption has been studied in the BMJ Quality and Safety journal. The Joint Commission has identified failure to communicate as a contributing cause of sentinel events. The Institute of Healthcare Improvement has studied the difference between work that is imagined vs work that is done. AHRQ has published work on workflow redesign as a prerequisite to implementing health IT. Lean methodology has been applied at Virginia Mason and ThedaCare for the improvement of healthcare workflows. This edition connects to the operational and workforce themes addressed in Editions H, J, and M of the newsletter.

Christopher Hutchins
Founder & CEO, Hutchins Data Strategy Consultants