Drug alerts and reminders
I was recently asked to look at a good article from the VA about physicians over-riding drug-drug interaction alerts, the reasons why, the measure of whether alerts were useful to end-users, etc A few thoughts:
When we developed and deployed an automated referral system for specialty care in the past, we struggled with the same question but looked at an approach such as
If an alert is to pop up during an order process, it must meet the following criteria
It must be highly useful to the person entering the order
It must be highly accurate most of the time
It must enhance efficiency on both ends (ie avoids a followup phone call back to the person entering in order to clarify, or get a different provider requested, etc)
It must be monitored for efficacy and dynamically updated based on a learning system
While doctors are not always cognizant of all the needs of “other departments’ workflow and accountabilities” , they are in fact highly accountable people generally who are dealing in levels of uncertainty around risk/benefit all the time, with varying levels of information at the moment, of varying levels of accuracy. Therefore, designing systems that are intended to assist clinicians in their outcomes and efficiencies has a primary obligation to prove that will happen before introducing more uncertainty or inefficiency into the process.
I have always believed that the future of information systems in healthcare will be realized when the systems evolve rapidly and dynamically by learning based on the behaviors and decisions of those who provide healthcare, and the first sniff test will be like the microsoft ad
:where do you want to go today?
A clinician should login to a location, and the system should be able to learn their question, their context and their task at hand rapidly
The system should provide information in a manner which has been demonstrated via real time workflow evidence to enhance the efficacy AND efficiency of the clinician/ The screens one sees should derive not from the needs of the data or the back-end departmental needs but from an intersection between what busy clinicians need in order to get the specific job done at that moment, but increasingly “made aware of” what other systems or processes appear to know, or might know, about that context
So perhaps if instead of thinking of these as alert overrides, we began thinking about any alert as never appearing until it can be shown to have passed a performance test that meets those criteria in real world settings, we would end up with alerts that had the type of content and context that users now ASK FOR, or create themselves. This would be very different from today, where users are viewed as “reacting to and “bypassing” because the alert process only represents one perspective on a subset of data about a particular patient or clinician or medication, that is highly likely to be not so aware of the rest of the context at that time. Obviously that improves as systems become more sophisticated, but starting with assumptions of the inadequacy of the data sometimes are better methods than starting with the assumption that the data or rules are consistently useful enough to begin shoving in the face of very busy people who are multi tasking!!
Very interesting subject/