Domo Arigato, Mr. Clinical Roboto

I’ve been told that the story of reducing clinical variation in healthcare is rooted in automobile manufacturing. It seems intuitive that reducing variation in a manufacturing process will improve the reliability of the output. In my head, it goes something like this:

Jane can’t come in today? Not to worry! Her replacement, Bill, puts on the car door in the exact same way. This means that we won’t have Jane-quality cars and Bill-quality cars; just quality cars! 

Yet, can we distill these concepts to the nuanced complexities of patients? As a clinician, I don’t view my delivery of healthcare as being similar to that of working on a factory line where all inputs are identical. Every patient is different. The “congestive heart failure (CHF) patient in room 17” is an individual with a name and a unique personal story of how they got there today. They may have chronic kidney disease which complicates CHF management, allergies to several optimal medications, and insurance inadequate enough to make outpatient management difficult. What he needs is unlikely to be standard-issue care.

So how can clinicians apply the standardization principles to the practice of healthcare without oversimplifying each unique patient? It’s certainly a complex topic. That being said, with recent examples highlighting continued quality-of-care struggles, it’s important to embrace any effective improvement strategies.

Drink a Cup of Albuterol and Call Me in the Morning

No, clinicians aren’t robotic assembly line machines. Yes, each patient is different. And we must recognize that the complexity around medicine introduces an enormous number of hazards. Rather than giving up when it comes to tackling variation, one area of focus that can deliver immediate return on investment (ROI) is creating alignment with up-to-date practice delivery.

One example that comes to mind dates back to my residency. Asthma is a very common presentation in the pediatric emergency department where a mainstay of acute management is inhaled albuterol. However, there was one outpatient pediatrician who continued to manage their patients with oral albuterol. To be honest, I wasn’t aware that albuterol was even available to be dispensed orally. While writing this piece, I looked up the evidence and found a publication that cited efficacy over placebo… from 1982. Subsequently published, there is robust evidence that inhaled albuterol is faster-acting and leads to fewer generalized side effects.

So, why was this pediatrician delivering suboptimal treatments to their asthma patients? I can’t imagine they purposefully wanted to deliver poorer care. The only plausible reason is that the training the clinician received many, many years ago became embedded in their daily practice and was simply never adjusted. I imagine that this is more common than one might think given the rapidly evolving research. One estimate suggests that the doubling time of medical knowledge has accelerated from 7 years in 1980 to around 73 days in 2020. In other words, the foundations of knowledge taught to medical students just after their white coat ceremony have significantly changed by the time they receive their diplomas.

As a patient or parent without a medical background, how can you ensure that the delivered care is optimal and aligned with best practices? It’s a scary thought that your healthcare provider might be missing a newly validated diagnostic approach, ordering unnecessary (and costly) procedures, or prescribing unsafe treatment options simply because they aren’t up to date. We must strive to ensure that care delivery is consistently aligned with established evidence and guidelines whenever possible.

The Tug-of-War Between Protocols and Autonomy

The albuterol story is a great example of an instance in which an inexcusable gap in care delivery is due to a clinician’s knowledge and established evidence. This is an area in which curating clinical workflows using tools like clinical decision support (CDS) in the electronic health record (EHR) system fills an important gap. However, EHRs are caught in the middle of the tug-of-war between customization and standardization. EHR tools, like order sets, are oftentimes structured with clinician autonomy and customization at heart. This can be at odds with the goal of protocolization tools intended to dictate alignment with best practices. 

As an example, one way of reducing variation is through the use of EHR defaults. “Nudge” teams have been established at health systems to ensure clinical workflows facilitate higher quality care. In an unsurprising emergency department example, defaulting to a smaller amount of dispensed quantity in opioid prescriptions orders led to a smaller total amount of opioids prescribed. A similar example by the same group showed over $32 million in savings by defaulting to safe, generic medications in order sets. 

While these changes take place at the health system level, you can imagine a clinician overriding these defaults with their own preferences such that the care provided may not align with best practice or optimal care delivery. Importantly, when these user overrides persist over time, despite an evolution of medical research that contradicts them, it can lead to an increasingly divergent and unsafe care delivery workflow. This ultimately transitions the responsibility of aligning EHR workflows with up-to-date best practice guidelines from members of the dedicated quality and informatics teams to individual, overwhelmed clinicians. 

Use Data To Understand Variation

So how can quality and informatics teams address the challenges of clinical variation given the highly customizable environment in which they occur? The key is to look at the stories that the data is telling. This is one of the primary pillars of value that we deliver at Phrase Health. 

One example of how we look at clinical variation is analyzing instances of ordering soon after using a curated order set. We call this “orders within 5 minutes”; we literally  systematically look at the orders a clinician places on a patient within 5 minutes of using an order set on that same patient. Conceptually, a clinical workflow employs functionality in the EHR to help complete a set of tasks, so if clinicians go around the tool and order something else, it is evidence of a potentially addressable variation. Again, the goal is not to have 0 orders coming outside of an order set; all patients are different and some variance in care provided will account for that. However, there may be orderables purposefully left off because they are contraindicated or because they reflect an outdated practice. The power of harnessing this variation is exemplified by research published by my colleagues at Children’s Hospital of Philadelphia who reduced the use of inappropriate albuterol administration in bronchiolitis patients.

Another way of understanding variation is by getting clinician feedback about their care practices. The effective yet time-intensive way of doing this is to individually interview and survey end users. However, for those without the time or bandwidth for that approach, there are clues left in the EHR by end users that can be analyzed at a system-level. For example, one of our partner sites aimed to administer flu vaccines in the emergency department to eligible patients. Feedback left in the alert frequently cited “febrile” or “fever” despite current best practice dictating this not being an absolute contraindication. This gap between clinician knowledge and best practice led to updates in the workflow and development of decision support to better lead to the desired outcome.

Bringing It Together

Beyond regulatory mandates to review protocolized ordering, health systems have significant opportunities for increasing ROI by understanding how best practices are (or, importantly, are not) being adopted within clinical workflows. This can be done at “mass production” scales by establishing system-level processes around tasks like reviewing deviations from ordering tools and scanning asynchronous clinician feedback. The EHR may enable a robust amount of customization, but it also stores the data that tells the story about reducing clinical variation.