The Challenge

From the beginning, the team realized that the problem we were asked to solve would be far more complex than it seemed on the surface. The broad problem space in which people, even when their lives literally depend on it, are not very good about following doctors’ orders to maintain good health, or recover from a major trauma or surgery. More specifically, people have trouble being adherent to their care team’s prescribed medication plan.

To these ends, catalia health had prototyped and tested, and would now be bringing to market an answer to the medication acherence part of the equation in the near term, and perhaps an answer to the more holistic problem of maintaining good health in the longer-term. Their answer is Mabu, a bright yellow-orange, plastic robot that maintain quasi specific verbal conversations with people. It was this framework and context that the team began to work within, attempting to solve for the complexities of multiple med- ication management, via a conversational UI, which would be laid on top of a non-generalized AI, in the form of a physical robot.

Therefore our challenge became:
How might we create an easy-to-understand, simple set of conversations for Mabu to help patients manage the complex, perhaps very unpleasant, but life-saving task of managing their medications?

The Discovery

We surveyed 75+ patients asking about their medication routines, medication storage, and attitudes towards technologies used to aid their medication routines. Patients hesitated including technology into their daily life, calling it "this unknown equipment." Typically, patients used pill boxes and would personally recall by time. We also interviewed several people such as nurses and patients that would allow us to interview them about how they manage their medications and insights on how Mabu can aid them with their medications.

However, the more patients we surveyed and interviewed, the more it became apparent that complexity only increases if any of the medications: must be tied to food; taken on a schedule other than every day; are not taken at the same time as each other; have variable dosing instructions on different days; or if the patients have dementia, or pre-dementia; and the list of factors goes on. It was this level of complexity which drove the team to stop gathering data after several weeks, and start trying to create a framework of how these conversations might sound

Complexity of Conversations

This level of complexity only grew with each round of edits and user tests –we discovered that a patient does not necessarily even think of all of their medications in the same way (one is the “blue pill” and another is called by its generic proper name, as an example). This meant that no generalized rules could be created about how any single patient might continue to interact with Mabu and think of the rest of their medications, based on how they spoke about the first medication they onboard with Mabu. At this point the team realized that conversations need to be highly modular, dynamic, and account for a variety of types of behavior amongst patients and within a single patient.

Conversations started getting broken down into shorter, more modular conversations and conversation chunks, and were built to be structurally solid, and not necessarily true to the exact content Mabu might speak. One of our later user tests was conducted in Mandarin Chinese, which the team took as confirmation that the conversations we had built were sound, architecturally. It was extremely important that the conversations be built more for structure than content for several reasons, but there were three main factors which drove this strategy more than anything else.

The Outcome

In the case of this team’s work, it took no small effort to understand all the onion skin layers of complexity which exist for patients who take many medications. But as these layers revealed themselves, the team had the difficult task of making all of it simple, easy, and intuitive.

We created 11 different conversational user interfaces assisting patients in four key areas: 1) onboarding medications. 2) checking on medication adherence. 3) setting up medication groups. and 4) setting up meal reminders. In final implementation, we accomplished having the same amount of conversational text and duration to manage multiple medications versus singular medication. Simplicity out of complexity reduces clutter.

Reflection

The work we have done for Catalia Health can be used in multiple complex scenarios, as an architecture to talk about multiple disease states (concurrent diagnoses), including mental illnesses. This work will help contribute to Mabu and Catalia Health’s ability to go forth and tackle any disease state, or any combination of them, which in turn might just revolutionize assisted living and home healthcare as we know it.