The development and implementation planning of an AI-powered Palliative Care Assessment Tool (PCAT) for aged care: a Triple C model approach

In the quiet corridors of residential aged care, a critical challenge often goes unnoticed until it is too late. Many elderly residents live with conditions that will eventually shorten their lives, and for these individuals, the goal of care shifts from curing illness to ensuring comfort, dignity, and quality of life. This approach is known as palliative care. Ideally, this support begins early, allowing families and medical teams to manage symptoms and plan for the future while the person is still well enough to participate in decisions. However, in the busy reality of nursing homes, these needs are frequently identified late. Staff members, often stretched thin and managing complex schedules, may miss the subtle signs that a resident is entering a phase where specialized support is needed. While doctors and nurses have long used checklists and structured questions to spot these needs, these tools rely on a human remembering to use them at the right moment, and they usually focus on one person at a time rather than looking at the whole community of residents.

To solve this, researchers have begun exploring how artificial intelligence—computers that can learn from data to recognize patterns—might help. The idea is not for a machine to replace a doctor's judgment, but to act as a tireless assistant that scans information about every resident every day, flagging those who might need a review. This paper describes the careful creation of such a tool, called the Palliative Care Assessment Tool, or PCAT, designed specifically for aged care in Victoria, Australia. The team did not simply build software and hope it would work; they followed a specific, human-centered roadmap called the Triple C model. This approach prioritizes talking to the people who will actually use the system, working together to shape the design, and ensuring the tool can survive and thrive in the real world long after the initial project funding ends. The result is a digital system that is ready to be tested, built on a foundation of safety and practicality rather than just technical capability.

The journey began with a simple but difficult question: how do you build a digital tool that fits into the chaotic, high-pressure rhythm of a nursing home shift? The researchers knew that if the tool was too complicated or slowed down the staff, it would be ignored. They started by gathering a wide group of people, including nurses, care home managers, and specialists in palliative care, for a series of six structured conversations. These were not formal interviews but collaborative workshops where the team listened to the daily struggles of the staff. They learned exactly where the gaps were in current care, what prevented early identification of needs, and where a digital prompt would need to appear to be seen and acted upon. This phase, known as consultation, ensured that the tool was designed around the reality of the work, not an idealized version of it.

Once the problems were clearly defined, the team moved into the collaboration phase. Here, the focus shifted to turning those insights into a working design. The researchers worked closely with the technical team to build a system that could take the complex logic of existing medical guidelines and translate them into a simple, clear interface. A key principle was that the tool must never remove the human's ability to make the final call. The system was designed to offer a risk rating—a simple indication of how likely a resident is to need palliative support—but it always included a clear path for a clinician to override that suggestion if their own professional judgment differed. This "human in the loop" feature was crucial, ensuring that the technology supported rather than dictated care.

Before the tool could ever be shown to a patient or used in a real nursing home, the team subjected it to a rigorous safety check known as a heuristic evaluation. This is a method where experts review a system against established principles of safety and usability, looking for potential pitfalls before they cause harm. The evaluation uncovered several critical issues that needed fixing. For instance, the initial design allowed a user to accidentally delete a resident's record, a risk that was immediately removed. The team also found that the system sometimes showed a risk rating that contradicted the recommended action, which could confuse a tired nurse. They also noticed that the tool did not explain why it gave a certain rating, which made it hard for staff to trust the advice.

The team addressed these issues with specific, practical changes. They added a "why this rating?" feature that explains the reasoning behind the computer's suggestion in plain language. They introduced a visual indicator to show if a resident's condition was getting worse, staying the same, or improving, helping staff see the trend over time. To reduce the mental load on nurses, the tool was reorganized to show the most urgent residents first, rather than listing them alphabetically, and it was programmed to hide triggers that did not apply to a specific person. Color was paired with text and icons so that the meaning was clear even if someone could not distinguish the colors. Perhaps most importantly, they built in a documented pathway for a clinician to disagree with the system, ensuring that the final decision always rested with a human professional.

The paper concludes by outlining how this tool is now prepared for a staged trial across several aged care sites. The researchers plan to introduce the tool to nursing staff who will use it as part of their normal daily routine, starting with a small group of residents. The goal of this next phase is not to prove that the artificial intelligence is perfect, but to see if the tool is usable, acceptable, and feasible in the real world. The authors are careful to state that this paper only covers the design and preparation stages; the actual results of the trial, including whether the tool successfully improves care, will be reported in a future study. By front-loading the work of listening to users and fixing safety issues, the team has created a system that is ready for the real world. This approach suggests that for technology to succeed in healthcare, it must be built with the same care and attention to human needs as the medical treatments it supports. The success of such a tool depends less on the sophistication of its code and more on its ability to fit seamlessly into the lives of the people who use it, ensuring that no resident is overlooked simply because the system was too difficult to use.