AI in elderly care: two problems nobody is talking about together

AI in elderly care: two problems nobody is talking about together

Artificial intelligence is arriving in care homes.

• Wearable devices monitoring blood pressure and blood sugar.

• Motion sensors that detect falls and alert staff within seconds.

• Apps like Pain Check, which uses facial recognition to identify pain in residents who can't verbalise it.

• Remote monitoring systems eliminate the need for overnight checks every two hours - allowing residents to sleep without interruptions.

The potential is genuine.

The UK has more than 10,000 care homes for individuals over 65, a workforce nearing 800,000, and demand that currently exceeds supply. Technology that boosts clinical oversight, cuts avoidable harm, and helps busy staff needs our focus. But two problems underlie this, and they compound each other in complex ways that lack enough discussion.

1. The data problem. The designers made tools for older, frail patients. These patients often face many health problems and cognitive challenges. These patients are also the ones most likely left out of the datasets used to create and test the tools. A systematic review of 65 RCTs evaluating AI prediction tools found that nearly 40% showed no clinical benefit over standard care. In geriatric medicine, validation studies show large differences. They also find few meaningful outcomes and limited real-world testing. An approved medical device algorithm isn't the same as one that works for an 84-year-old with dementia, heart failure, and polypharmacy.

2. The deployment problem. It's the high-end luxury care homes, the ones charging upwards of £5,000 a week, that are leading on AI adoption, simply because they can afford to invest. The frailest and most complex residents are last in line. They are in beds supported by underfunded local authorities. Put those two problems together and the picture sharpens. We have tools that don't fully understand their target population being rolled out first to the patients who need them least. Residents who need better monitoring, earlier detection, and smarter care coordination are at the end of both lines.

So what can primary care actually do? Three things.

1. Ask questions. When you see AI-generated data in referral letters or care home messages, like pain scores, fall risk ratings, or monitoring alerts, handle it like any other diagnostic test. Which tool generated it? In which population was it validated? What are its known limitations? That clinical scepticism is not obstructive; it's appropriate.

2. Advocate for inclusion. Older, frail, multimorbid patients need to be represented in the validation studies for these tools, not excluded from them. Clinicians who work with this population every day can help challenge that — in research talks, at the PCN level, and when engaging with tech developers.

3. Engage with commissioning. AI funding and deployment decisions at the ICS and PCN levels should include input from primary care. If the clinicians who know this patient group aren't included, the gap will grow. Patients who need help the most will keep getting it last.

The technology is promising. The evidence base and the deployment strategy need to catch up. Primary care is well placed to push both in the right direction.