Waiting for AI to automate medication dosing? Dose automation is already here.
Up to this point, electronic health records have done a remarkable job transitioning the American healthcare system off of paper charts. While there are still vestigial paper processes present in the corners of uninspired healthcare organizations, over 95% of hospitals have met Meaningful Use requirements for health IT. But, is it enough to simply transfer the same paper workflows to the computer?
Eventually machine learning will be unleashed on EHRs, finally unlocking the value of the data collected over the past decade plus. But that does not mean that we cannot take advantage of the available clinical data for real time decision support today, especially for organizations using the Epic EHR. Up to this point, we have had some token efforts at real time clinical decision support in the way of BPAs and med warnings. Unfortunately, these end up being more of a nuisance than an asset, heavily contributing to alert fatigue.
An under appreciated functionality upgrade that came all the way back in the 2017 version of Epic makes a giant leap forward in real time clinical decision support. Large prestigious organizations are dipping their toes in the water of these new features, but most organizations haven’t realized the true power of what Willow has done with rule based ERX contexts and rule based dispensable mapping.
As clinicians, we learned to love Case Studies in school. Let’s do an abbreviated one to highlight workflow:
LK is an adult ICU patient with a potassium of 3.0. The P&T approved hospital protocol states that the patient should receive a 40 mEq bolus of potassium chloride. The patient has a central line placed, so he can receive a concentrated premix bag.
Many Epic organizations have an order set for such situations. There are probably several sections, one for each arm of the potassium replacement protocol. Maybe there are even separate sections for central vs peripheral lines. Some of these order sets are probably pretty slick. As long as the ordering provider has committed the potassium replacement protocol to memory and they know off hand if the patient has a central line or not, the order set is a pretty nifty tool. If the provider doesn’t know those things, that could mean a lot of pharmacist work on the back end with the potential of a drug error in the mix.
But what if the provider could just order IV potassium and the system did the rest? Would that increase patient safety? Would that save the providers and the pharmacists some valuable time? Think about how many times a day potassium replacement is ordered.
What if LK isn’t an adult? What if LK is a child in an ED in a hospital that doesn’t see many pediatric patients? Could having the pediatric protocol built into the system possibly save that child from a potentially fatal drug error?
Potassium replacement is a relatively straightforward but also a very common example. Potassium, or any electrolyte replacement protocol, is a perfect example of how the system could make better use of data that is already being collected. Two key Willow features liberate this data and use it to present meaningful automated clinical decision support in real time. This is made possible by rule based ERX context and rule based dispensable mapping.
Here are some other common scenarios that can be solved with rule based ERX context and rule based dispensable mapping:
Dosing by renal function (Vancomycin, Zosyn, or any other renally dosed medication)
Dosing by multiple factors, such as age and weight (Pre-term neonate gentamicin dosing vs full term neonate dosing)
Products by patient location (useful for large hospital systems with a variety of sites. Smaller hospital carries Mini-bags, larger hospital has premix)
Concentration dependent (ICU vs medical floor, central line vs peripheral)
Basically any medication that has dosing defined by a discrete set of parameters
Let’s find out how all of this is done.
1. Rule Based Context
Not many organizations have made the shift from scoring system contexts to rule based contexts. Rule based contexts present a monumental shift in how contexts are defined and how they are used. Instead of contexts being defined in one system level scoring rule, contexts can be defined at the ERX level using patient context CER rules. While this may not seem like a big difference, it allows the system to evaluate patients on any number of variables that can be defined in a rule. For example, vancomycin has variable dosing based on age, weight and renal function. It is now possible to present a dosing recommendation for a preterm neonate under two kilograms and a different dosing recommendation for a full term neonate that is four kilograms. It is the system enhancement of evaluating multiple variables that unleashes the power of dose automation. With the number of variables that can be evaluated by CER rules, the opportunities for dose automation are virtually limitless.
2. Dispensable Mapping Rules
Dispensable mapping has proven to be an incredibly useful automation tool for the pharmacy. While most organizations do have a catalogue of dispensable mapping ERX records, this functionality is highly underutilized. Aside from the automation benefits, dispensable mapping records also reduce the maintenance burden associated with changing ERX records used in order sets or treatment plans. Yes, reduce maintenance burden.
Dispensable mapping rules introduce a much more exciting level of automation. Much like rule based contexts, the rule parameter available in dispensable mapping allows any patient context CER rule to be utilized in evaluating a dispensable. A common and easy to understand example would be the IV electrolyte policy. Generally, the concentration and rate of administration of IV electrolytes are dictated by a patient’s location and whether or not the patient has a central line placed. Patients in the ICU with central lines are able to receive more concentrated electrolytes administered at a faster rate while patients on the medicine floor cannot. Until now, these types of policies have been enforced manually by pharmacy and nursing staff. With dispensable mapping rules, the system can evaluate where the patient is located and if that patient has a central line, and then choose the appropriate dispensable ERX.
Combining rule based contexts with dispensable mapping rules opens the door to extensive automation. Virtually any policy, order set, or algorithm based on discrete variables could be automated using these two Willow features.
Since many organizations are strapped for time and resources, rolling out these two Willow features needs to be a strategic process.
Begin by targeting high use medications with consistent prescribing practices, similar to the electrolyte example presented earlier. Automating dosing for these medications produces the biggest impact return on build effort, while minimizing the amount of clinician time needed.
Next, focus on procedures that need to be standardized but for any variety of reasons are not. Let these “new” Willow upgrades be the excuse to define practices that have gone neglected too long. Beyond gaining efficiencies through automation, standardization reduces patient safety risks and leads to more consistent outcomes.
Assign each medication project to just one analyst as a point person. Shield them from distractions like other projects or meetings. It is important to keep the turnaround time for these projects very low in order to keep the clinical stakeholders engaged.
Dosing automation represents the next frontier in electronic health records. With automation, we are moving beyond simply transferring our practices from paper to the screen. As with any other industry, automation reduces the burden of mindless, repetitive tasks on our valuable clinicians, allowing them to focus their energy on solving more complex problems. From a patient safety perspective, automation ensures consistency across all patients, reducing the risk of needless human error. Embracing dosing automation is embracing the future of medicine.
(Another) Case Study
With so many potential applications for these features, let’s keep it simple with a common drug that most organizations have a protocol for: vancomycin.
How would you write patient context rules to capture this protocol? Keep in mind, the system will stop on the first rule that evaluates as true.