Strategic Patient Involvement in Medical AI Co-Design: A Pilot Study in Heart Failure

In the emergency rooms of hospitals around the world, doctors face a difficult and urgent question every day: when a patient arrives with a failing heart, can they go home safely, or do they need to stay? This decision is high-stakes. Sending a sick person home too soon can lead to a return visit or worse, while keeping a stable person in the hospital wastes resources and delays care for others. To help make these choices, scientists are building computer programs that use artificial intelligence to predict a patient's risk. These tools look at medical data to guess who might get sicker. However, for a long time, these tools have been designed by doctors and engineers who decide what information matters and how the results should look. The people who actually live with the heart condition, and who have to live with the consequences of these decisions, have rarely been asked to help shape the tools themselves.

A new pilot study from a team of researchers in Spain and the Netherlands asks what happens when you flip that script. Instead of asking patients to simply look at a finished product and say if they like it, the researchers invited people with heart failure to sit down and design the tool from the ground up. The goal was to see if patients, when given real power to decide how the technology works, would create something different than what experts had imagined. The study found that when patients were treated as strategic partners rather than just testers, they completely reimagined the tool. They moved it away from being a calculator for doctors and turned it into a supportive guide for patients, one that speaks in plain language, respects their emotional needs, and refuses to give an answer when it is not sure.

The researchers brought together eleven people who had lived experience with heart failure for a two-day workshop in Barcelona. Before they met in person, the participants took an online course to learn the basics of how artificial intelligence works, ensuring everyone started on the same page. Once they arrived, they were not asked to critique a prototype. Instead, they were given the freedom to decide what the tool should be, who it should serve, and how it should behave. The workshop was structured around principles of trustworthy technology, guiding the group through activities like mapping out a patient's journey, creating personas for different types of people, and role-playing how the tool would fit into a real medical visit. The participants worked in two separate groups, and remarkably, both groups arrived at nearly the same design without talking to each other.

The most striking change the participants made was to the very purpose of the tool. The project had started with the idea of a screen for doctors in the emergency room to help them decide on discharges. The patients immediately pointed out a flaw in this thinking: if the tool is only for doctors, the patients themselves never see it, and they never get to understand the risk or the plan. They insisted that the tool should be a mobile application for the patient. They wanted to be the ones entering their own information, not just having their data collected by a machine. They argued that the tool should not just look at a snapshot of their heart function in a moment of crisis, but should understand them as whole people. This meant including details about their life, such as their diet, their religious practices, their living situation, and their ability to travel, because all of these factors change how a risk prediction applies to them.

When it came to how the tool should show its results, the participants rejected the cold, hard numbers that are common in medical statistics. They explained that seeing a specific percentage chance of death, such as "6.4 percent," was terrifying and unhelpful. Instead, they wanted the risk to be shown in broad, understandable categories, like low, medium, or high. More importantly, they wanted the tool to tell them what to do next. A risk score alone is just a number; a risk score paired with a clear, actionable step, such as "call your doctor today" or "monitor your weight," is something a person can use. They also wanted the tool to be honest about its own limits. If the computer did not have enough data to make a reliable guess for a specific person, the participants said the tool should simply say, "I cannot tell you right now," rather than giving a false or misleading number. They viewed this silence as a form of honesty and a way to build trust, rather than a failure of the technology.

The design also included a way for patients to control how much they wanted to know. The participants recognized that in a moment of acute illness, they might not have the energy to read a complex explanation. They designed a system where they could choose the level of detail. They could see a simple summary, or if they wanted to, they could click to see more technical information about how the prediction was made. This gave them the power to decide how much information they could handle at any given moment. They also emphasized that the tool should be a bridge between them and their doctors, allowing them to share their personal data with their care team and receive updates in return, creating a partnership rather than a one-way flow of information.

Despite the success of the workshop, the participants were realistic about the challenges that remain. They pointed out that turning these ideas into a real, working product is difficult. They noted that laws about data privacy, the cost of building and maintaining the software, and the way hospitals are funded often create barriers that stop good ideas from reaching the final product. One participant noted that a single workshop is not enough to solve these deep structural problems, and that the process of involving patients needs to be standardized and supported with proper funding and time. The researchers agreed, noting that while the patients produced a clear and coherent vision, the path from that vision to a tool in a doctor's hand is still uncertain.

This study suggests that when people with lived experience are given the space to design the technology that affects their lives, they do not just tweak the interface; they redefine the problem. They moved the focus from a tool that helps a doctor manage a hospital bed to a tool that helps a person manage their life. They showed that for high-stakes technology, especially when it involves the risk of death, the emotional and practical needs of the user are just as important as the mathematical accuracy of the algorithm. The work demonstrates that patients are capable of strategic thinking about complex systems, provided they are treated as partners rather than subjects. While the tool described in this study is not yet in use, the design specifications created by the group offer a new blueprint for how medical artificial intelligence could be built to serve the people it is meant to help, rather than just the systems that use them.