AI-driven decision support for the elderly
Populations in Europe and across the world are ageing, leading to a growing number of older people living with multiple chronic conditions. These individuals often experience declining function, frailty, and the need for complex care in nursing homes or home settings. Clinicians must make daily decisions about treatments and interventions, but conventional clinical trials rarely capture such complex populations. This leaves a gap between available evidence and the needs of real-world patients.
Building predictive models from real-world data
The EU-funded I-CARE4OLD(opens in new window) addresses this challenge by designing a decision-support tool to aid healthcare professionals and policymakers. “This decision-support tool was designed to predict changes in functioning, the occurrence of critical events, and the individual impact of starting or stopping specific interventions,” explains project coordinator Hein van Hout. The consortium obtained real-world data from multiple countries from interRAI assessments(opens in new window), a suite of structured instruments used worldwide to monitor health, functioning, and service use in older populations. These assessments are standardised across countries, ensuring consistent and comparable data. I-CARE4OLD trained AI models using these data to predict both short- and long-term outcomes with high accuracy. Prediction accuracy reached 80 % and researchers validated the robustness of the AI models across datasets from different countries. Furthermore, 150 professionals in seven countries have tested the I-CARE4OLD decision-support tool and have found it helpful and easy to use. “Our AI models not only allow better prediction of disease progression but also of treatment outcomes, bringing personalised care for older adults a step closer to everyday practice,” highlights van Hout.
Key findings
Among the most notable insights from the analysis was the effect of discontinuing certain psychotropic medications(opens in new window). Stopping antipsychotics reduced hospitalisations and improved overall outcomes in many care recipients. The analyses also showed how non-pharmacological interventions could deliver benefits equal to, or greater than, medication. Importantly, the models identified which older adults were more likely to benefit from specific drugs, such as antidepressants, and which were at higher risk of harm from treatments like those with anticholinergic effects. Beyond treatment impact, the project also studied the effects of the COVID-19 pandemic(opens in new window) on older persons receiving care. These analyses highlighted the vulnerability of this group during health crises and reinforced the importance of predictive tools for anticipating risks and guiding preventive interventions.
Implementation and scale-up
The most significant achievement of I-CARE4OLD is its demonstration that AI can generate personalised predictions from large-scale, real-world data and make them available to clinicians through a practical tool. This innovation informs clinical practice to optimise targeted interventions for the most complex and vulnerable older populations. The data may also inform policymakers on the presence or absence of (in)appropriate treatments in their jurisdiction. The consortium now aims to expand the scope of predictive algorithms to cover more treatments, both pharmacological and non-pharmacological, and validate them in additional datasets. Importantly, to scale up data collection, the team is working with interRAI software providers across the world to integrate the decision support into existing assessment systems. Deployment of this tool across health systems will help reduce inappropriate prescribing, improve quality of life, and support healthier ageing.
