Data Science In Outpatient Access Management

I work as a data scientist at Mayo Clinic in the Enterprise Office of Access Management (EOAM). Access management in health care generally is all about improving patient access to appointments and resources. Access management is also about optimizing provider schedules to meet their desired patient mix while providing timely access to patients. At Mayo Clinic, EOAM is focused on managing appointments and schedules in the outpatient setting.

What are some of the problems facing outpatient access management? There is a core function to outpatient access management related simply to balancing our capacity with demand: how many appointments have we had in the last month, and can we use past appointment volumes to predict future volumes? But at Mayo Clinic, the more important side to access management is what is best for the patient? We work to reduce appointment lag: the time between when the patient wants to be seen, and when they actually get seen. We work to reduce itinerary lengths: if a patient has to travel a great distance, we want them to be able to come to a Mayo Clinic site, have access to the necessary specialists and tests, and then leave - all within a condensed time frame.

We can solve some of the more important problems with a dashboards and reports visualizing our data, and I have a number of colleagues very skilled at SQL and Tableau that work hard to push those data products to our users.

Other problems require more indepth programming skills such as Python, which I use as a “glue” language: it is the central language for my data products, and gives me all the database access I need. A typical workflow for me is to pull data from a database by using Python to call SQL, perform a more complex computation on the data, and then push the results up to a different database that my colleagues can use to feed another Tableau extract.

Here are several projects supporting outpatient access management at Mayo Clinic that follow this basic pattern. The first project is itinerary measurement. We grab data representing patient appointments, cluster them by how close they are in time, and then determine the length of time from the first appointment in a cluster to the last. The second is template scoring. For the purposes of this paper, a template is a pre-determined pattern for a provider’s schedule: we want certain kinds of appointments to go into certain times of the day, into certain provider’s schedules. These templates in turn drive our Auto Scheduler that our schedulers use, so that when a patient calls us up and wants an appointment, Auto Scheduler is what presents the scheduler with the best options. There are such things as overrides, however, which is where template scoring comes in. Template scoring compares the actual appointment that went into a slot with what the template set up the slot to accept, and scores that slot accordingly. This is by way of measuring how well our templates are performing. Finally, there is the concept of a pathway: the appointments that a patient going through Mayo Clinic. Mayo Clinic is a network of specialists, and so it is commonplace for a patient in Neurology, say, to need a consult with a provider in Cardiovascular. If the Neurology appointment generated the Cardiovascular appointment, we can then link the appointment data to reflect that (often – certainly not always). This idea of one appointment generating another induces a Directed Acyclic Graph, such as are used in the new causal revolution (indeed, this is a causal graph!). It is very simple, in a way: if appointment A generates appointment B, you have the edge A B. The business questions you can ask of this graph are legion: where do particular surgeries come from initially? Does a particular kind of new visit tend to generate a particular test? What is the probability that a particular kind of patient will end up needing radiation treatment?

These are some of the applications of the ideas of data science to the problem of managing outpatient access.