Exploring the acceptability of and implementation of SUGEST: Diverse caregivers' perspectives on a machine learning-supported prognostic tool for earlier identification of palliative care needs
This study explores diverse Scottish family caregivers' perspectives on the SUGEST machine learning tool, revealing that while they view earlier identification of palliative care needs as valuable for enabling timely planning and support, its acceptability and successful implementation depend on transparent data, clinician oversight, compassionate communication, and robust pathways to action.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
As people live longer and manage more complex health conditions, the moment when medical care shifts from trying to cure an illness to supporting a person through its final stages becomes increasingly difficult to pinpoint. Ideally, this transition happens early enough to allow for calm planning, symptom management, and meaningful time with loved ones. However, in many healthcare systems, this shift often occurs only after a series of frightening emergencies, leaving families to navigate a chaotic path while the patient's condition worsens. To address this, researchers are exploring whether computers can help spot the signs of decline earlier by analyzing routine medical records. These digital tools, often called machine learning models, look for patterns in how often someone visits a doctor or hospital, hoping to flag individuals who might benefit from specialized support before a crisis hits. The promise is that spotting these patterns sooner could unlock better care, but the technology is only useful if the people it affects—patients and their families—trust it and if the healthcare system is ready to act on the information it provides.
A team of researchers in Scotland recently investigated how family caregivers feel about such a tool, known as SUGEST. This project aims to use everyday health data to identify older adults who are using emergency services frequently, suggesting they may have unmet needs for end-of-life care. Rather than building the tool in a lab and then asking people if they liked it, the researchers started by listening to the people who would be most affected. They interviewed fifteen bereaved family members who had cared for a relative through their final days. These participants came from diverse backgrounds, including communities that often face barriers in healthcare, such as ethnic minorities, refugees, and those living in poverty or rural areas. The goal was not just to see if the tool worked technically, but to understand if it would be acceptable to the people it was designed to help and what conditions would need to be met for it to be used fairly and effectively.
The conversations revealed a deep frustration with the current way care is delivered. The caregivers described a cycle of repeated crises, where their loved ones would be rushed to the hospital, sent home, and then rushed back again, with no one connecting the dots to see the bigger picture. They spoke of long waits, fragmented communication between different doctors, and a feeling that palliative care—the specialized support for serious illness—was introduced far too late, often only when death was imminent. For many, the term "palliative care" felt like a threat, signaling that treatment had stopped, rather than a source of comfort and practical help. Because of these experiences, the idea of a computer tool that could spot the need for support earlier was met with cautious optimism. The caregivers did not want a machine to predict death; they wanted a system that would trigger a conversation and a plan before the situation became an emergency.
However, the researchers found that this optimism was conditional. The caregivers made it clear that they would only accept such a tool if it was built on trustworthy foundations. They worried that the data used to train the computer might be incomplete or biased, especially for people who do not use healthcare services as often due to language barriers, poverty, or mistrust of the system. If the computer only learned from people who frequently visited hospitals, it might miss those who were suffering in silence at home. The participants emphasized that the tool must not replace human judgment. They wanted doctors to use the computer's warning as a prompt to check in, not as a final verdict. They insisted that any identification of risk must be followed immediately by real support, such as a nurse who could explain what was happening, a plan for home care, or help with practical needs. Without this follow-through, they felt that simply flagging a patient would be cruel, raising anxiety without offering a solution.
Communication emerged as another critical factor. The caregivers stressed that how the information was delivered mattered just as much as the information itself. They preferred face-to-face conversations with a trusted professional who knew their family's history, rather than a phone call or a letter. They wanted the discussion to happen in a calm, private setting, not in a busy emergency room. Crucially, they wanted to retain control over the process. They believed patients should have the choice to receive this information or not, and they wanted to decide who else should be involved in the conversation. The researchers noted that for some families, cultural or religious beliefs might make direct discussions about the end of life difficult, and the system needed to be flexible enough to respect those preferences. The tool was seen as a way to open a door for a compassionate conversation, but only if the person walking through that door was a skilled, empathetic human being.
The study concluded that the value of a machine learning tool like SUGEST lies not in its ability to calculate risk, but in its ability to create an opportunity for earlier, better care. The caregivers saw the potential for these tools to break the cycle of crisis-driven care, allowing families to plan ahead and feel less abandoned. Yet, they were equally clear that technology alone cannot solve the problem. For the system to work, it requires trustworthy data, a healthcare system with enough resources to respond to the alerts, and a workforce trained to communicate with sensitivity and humility. The researchers found that if these conditions are met, such tools could help shift the focus from reacting to emergencies to supporting people through their final journey with dignity and clarity. If these conditions are not met, the tool risks reinforcing existing inequalities and failing to deliver on its promise. The path forward, according to the families interviewed, depends on ensuring that the technology serves the people, rather than the other way around.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.