Written by: Kyle Samani
The premise of clinical decision support (CDS) is simple and powerful: humans can’t remember everything, so enter data into a computer and let the computer render judgement. So long as the data is accurate and the rules in the computer are valid, the computer will be correct the vast majority of the time.
CDS is commonly implemented in computerized provider order entry (CPOE) systems across most order types – labs, drugs, radiology, and more. A simple example: most pediatric drugs require weight-based dosing. When physicians order drugs for pediatric patients using CPOE, the computer should validate the dose of the drug against the patient’s weight to ensure the dose is in the acceptable range. Given that the computer has all of the information necessary to calculate acceptable dose ranges, and the fact that it’s easy to accidently enter the wrong dose into the computer, CDS at the point of ordering delivers clear benefits.
The general notion of CDS – checking to make sure things are being done correctly – is the same fundamental principle behind checklists. In The Checklist Manifesto, Dr. Atul Gawande successfully argues that the challenge in medicine today is not in ignorance, but in execution. Checklists (whether paper or digital) and CDS are realizations of that reality.
CDS in CPOE works because physicians need to enter orders to do their job. But checklists aren’t as fundamentally necessary for any given procedure or action. The checklist can be skipped, and the provider can perform the procedure at hand. Thus, the fundamental problem with checklists are that they insert a layer of friction into workflows: running through the checklist. If checklists could be implemented seamlessly without introducing any additional workflow friction, they would be more widely adopted and adhered to. The basic problem is that people don’t want to go back to the same repetitive formula for tasks they feel comfortable performing. Given the tradeoff between patient safety and efficiency, checklists have only been seriously discussed in high acuity, high risk settings such as surgery and ICUs. It’s simply not practical to implement checklists for low risk procedures. But even in high acuity environments, many organizations continue to struggle implementing checklists.
So…. what if we could make checklists seamless? How could that even be done?
Looking at CPOE CDS as a foundation, there are two fundamental challenges: collecting data, and checking against rules.
Computers can already access EMRs to retrieve all sorts of information about the patient. But computers don’t yet have any ability to collect data about what providers are and aren’t physically doing at the point of are. Without knowing what’s physically happening, computers can’t present alerts based on skipped or incorrect steps of the checklist. The solution would likely be based on a Kinect-like system that can detect movements and actions. Once the computer knows what’s going on, it can cross reference what’s happening against what’s supposed to happen given the context of care delivery and issue alerts accordingly.
What’s described above is an extremely ambitious technical undertaking. It will take many years to get there. There are already a number of companies trying to addressing this in primitive forms: SwipeSense detects if providers clean their hands before seeing patients, and the CHARM system uses Kinect to detect hand movements and ensure surgeries are performed correctly.
These early examples are a harbinger of what’s to come. If preventable mistakes are the biggest killer within hospitals, hospitals need to implement systems to identify and prevent errors before they happen.
Let’s assume that the tech evolves for an omniscient benevolent computer that detects errors and issues warnings. Although this is clearly desirable for patients, what does this mean for providers? Will they become slaves to the computer? Providers already face challenges with CPOE alert fatigue. Just imagine do-anything alert fatigue.
There is an art to telling people that they’re wrong. In order to successfully prevent errors, computers will need to learn that art. Additionally, there must be a cultural shift to support the fact that when the computer speaks up, providers should listen. Many hospitals still struggle today with implementing checklists because of cultural issues. There will need to be a similar cultural shift to enable passive omniscient computers to identify errors and warn providers.
I’m not aware of any omniscient computers that watch people all day and warn them that they’re about to make a mistake. There could be such software for workers in nuclear power plants or other critical jobs in which the cost of being wrong is devastating. If you know of any such software, please leave a comment.