Skip to main content

Table 5 Methods to enhance pharmacist compliance with AI recommendations

From: Discrepancy between perceptions and acceptance of clinical decision support Systems: implementation of artificial intelligence for vancomycin dosing

Improvement

Example Quote

Integrate AI into EHR

ā€œI suggest that this model be integrated into Epic, and please tell me where I could access it from within an order, this will encourage me to use itā€¦ā€¦, could incorporate the data from the EHR smoothly.ā€

ā€œI guess, round of the computer system, or at the very least within the Epic, have a quick way to get into this. Myself and a lot of my colleagues don't like to have a lot of buttons to quaking to see the information, so anything that we can do on a quick basis would help us be more compliant with it.ā€

Provide a rationale for recommended dose

ā€œProvide the recommended dose and give reasons why AI recommended this dose. I think this process can make AI more efficient.ā€

ā€œBecause I want to know are we missing something like that, that the system is maybe catching and that we could then use that in future dosing.ā€

Show pharmacokinetic values

ā€œSo expected half-life for the patient, what the peak value would be, what the trough value would be, what area under the curve would be within AIā€™s scheme compared to our scheme. And think that those would both be helpful tools as well to give us some sort of objective sense that my dosing really not appropriate here.ā€

Consider trends of the data, not a snapshot in time

ā€œI want to know that it's taken into account what's happened in the past, that it can kind of foresee what potentially can happen in the future based on Experience or previous cases that can resemble it.ā€

ā€œYeah, we're starting antibiotics at the right time, and then all of a sudden they become hypotensive, and the renal hypoperfusion might not be seen for another day or two, but sometimes we have to take that into account so that we don't overdose.ā€

Synthesize lab data and provide an overview of patients' clinical state

ā€œIf AI could tell me quickly, you know, the patient's urine output, blood pressure, and their trends, all the clinical things we use to correlate the lab data to what's actually going on real-timeā€¦ā€¦So if the AI could gather that information and save me a lot of clicks.ā€

Provide a risk index

ā€œGood suggestions in this model can give us recommendations and a risk index. If your dose is different from the AI, the risk index can indicate the degree of riskĀ that your dose is overdosing or underdosing.ā€

ā€œOh, I think if there was a method by which when we go to enter the dose, there is and the alarm or a warning sign that suggests that the probability that the dose is going to be off is considerable.ā€

  1. AI Artificial intelligence, EHR Electronic health records