Periodic Reporting for period 3 - I-CARE4OLD (Individualized CARE for Older Persons with Complex Chronic Conditions at home and in nursing homes)
Reporting period: 2024-06-01 to 2025-05-31
Optimizing care and treatment for older persons with complex chronic conditions (CCC) is challenging. Not only do older patients with CCC present with multiple conditions and functional impairments, but these also often interact with each other, as well as with their treatments.
Patients with CCC are concentrated in home care and nursing home settings. Professionals working in these settings often lack appropriate decision support that mirrors the medical and functional complexity of these persons.
The main aim of the I-CARE4OLD project was improve prognoses and estimation of treatment impact for older care recipients with CCC at home care (HC) and in nursing homes (NH), and develop, validate and test next generation individualised decision support, using real-world high-quality data. We had the following objectives:
1. Create enriched real-life high-quality datasets on older care recipients with CCC.
2. To identify homogeneous groups of persons with CCC sharing common disease patterns.
3. Develop and validate predictive algorithms for specific health outcomes, as well as for estimating impact of (non)-pharmacological interventions (N)PIs).
4. Estimate the impact of the Covid19 care period on health outcomes of older persons with CCC at home and in nursing homes.
5. Develop and pilot test user experiences of individualized prognostications among health professionals by a e-platform.
Patients with CCC are concentrated in home care and nursing home settings. Professionals working in these settings often lack appropriate decision support that mirrors the medical and functional complexity of these persons.
The main aim of the I-CARE4OLD project was improve prognoses and estimation of treatment impact for older care recipients with CCC at home care (HC) and in nursing homes (NH), and develop, validate and test next generation individualised decision support, using real-world high-quality data. We had the following objectives:
1. Create enriched real-life high-quality datasets on older care recipients with CCC.
2. To identify homogeneous groups of persons with CCC sharing common disease patterns.
3. Develop and validate predictive algorithms for specific health outcomes, as well as for estimating impact of (non)-pharmacological interventions (N)PIs).
4. Estimate the impact of the Covid19 care period on health outcomes of older persons with CCC at home and in nursing homes.
5. Develop and pilot test user experiences of individualized prognostications among health professionals by a e-platform.
WP2 relates to data management protocols, ethical risk evaluations and approvals, metadata and linkage descriptions, and harmonisation procedures. Several revisions were made to earlier deliverables following advice received in the review meetings and we continued updating the project DMP. We prepared a public metadata repository hosted by AUMC, for archiving the collected descriptions of national data sets and data usage, as well as modelling algorithms.
WP3 identified common disease patterns and was finalized within the reporting period 1. WP4 developed and validated predictive models for six targeted outcomes: death, hospitalization, functional decline, cognitive decline, increased frailty and decline in health-related quality of life in older adults with CCC receiving HC services or residing in NH/long term care (LTC) facilities. For NH outcomes, the data were generated from the Minimum Data Set (MDS 2.0) while HC data were provided through implementation of the interRAI HC assessment tool. Nine new prediction models were developed, and three existing models were selected and refined through this effort. Logistic regression analysis was used for the prediction model development. Following initial profile development, machine-learning (ML) strategies for each profile were successfully applied to further refine the prediction models. The preliminary models met reasonable eta and C Stat association criterion, with the ML models further increasing the specific accuracy of the predictive models.
WP5 developed and validated ML based models to predict the impact of PIs on health trajectories of older adults in NH and HC. PIs were shortlisted after an international survey asking health professionals which medications pose prescribing dilemma’s.30 predictive algorithms were developed and internally validated to estimate individual treatment responses and effects on outcomes. 12/30 were also subjected to external geographic validation and 3/12 were embedded in the I-care4old tool for piloting. Specific guidance was produced to support end users in understanding and interpreting the models.
WP6 developed and validated supervised machine learning (ML) algorithms that estimate individualised impact of NPIs on health outcomes. We selected 28 (HC) and 54 (NH) (N)PIs, covering environmental interventions, physical activities, psycho-social interventions, special therapies, preventive measures, special aids, nutritional interventions, regular care and advanced care planning. For multiple NPIs, interpretable causal-effect predictions were internally validated. Overall, most group-level effects were modest, but the models were able to clearly distinguish between no or a positive impact of a treatment on specific individuals. Some of the (psycho)social interventions in the LTCF setting were externally validated. Results generally translated well between countries. We developed decision support guidance on the validated algorithms with potential end users.
Additionally, for WP4-6, the ethical implications of the AI system were self-assessed by the Trustworthy Artificial Intelligence (ALTAI) framework proposed by the European Commission, highlighting possible risks for practitioners, the importance of training to reduce bias and to ensure that the end-users are aware of the limitations of the models.
In WP7 we developed and tested a unified, multilingual e-platform designed to provide individualised risk predictions for patients with CCC. In a pilot conducted across 7 countries, we demonstrated that clinicians are willing and able to engage with AI-powered decision-support tools. We offer strong proof-of-concept for the clinical relevance and feasibility of individualized prognostications based on real-world data. The pilot validated the platform's potential to support personalized, data-driven decision-making for older patients with CCC, laying the groundwork for future refinement, scale-up, and health system integration.
In this period, WP8 focused on translating research into impact by targeting healthcare professionals, organisations, and policymakers across Europe through tailored dissemination activities, and the development of an exploitation plan. We organised a final symposium with a global audience of stakeholders— clinicians, researchers, policymakers, vendors, and other important actors active in elderly care. By highlighting 6 key lessons learned important for a better healthcare for older persons with CCC, and moderating two panel sessions, we demonstrated the project’s relevance and reach beyond its funded lifetime.
WP9 explored whether and how the Covid19 care period impacted functional decline and changes in mood over time.
WP3 identified common disease patterns and was finalized within the reporting period 1. WP4 developed and validated predictive models for six targeted outcomes: death, hospitalization, functional decline, cognitive decline, increased frailty and decline in health-related quality of life in older adults with CCC receiving HC services or residing in NH/long term care (LTC) facilities. For NH outcomes, the data were generated from the Minimum Data Set (MDS 2.0) while HC data were provided through implementation of the interRAI HC assessment tool. Nine new prediction models were developed, and three existing models were selected and refined through this effort. Logistic regression analysis was used for the prediction model development. Following initial profile development, machine-learning (ML) strategies for each profile were successfully applied to further refine the prediction models. The preliminary models met reasonable eta and C Stat association criterion, with the ML models further increasing the specific accuracy of the predictive models.
WP5 developed and validated ML based models to predict the impact of PIs on health trajectories of older adults in NH and HC. PIs were shortlisted after an international survey asking health professionals which medications pose prescribing dilemma’s.30 predictive algorithms were developed and internally validated to estimate individual treatment responses and effects on outcomes. 12/30 were also subjected to external geographic validation and 3/12 were embedded in the I-care4old tool for piloting. Specific guidance was produced to support end users in understanding and interpreting the models.
WP6 developed and validated supervised machine learning (ML) algorithms that estimate individualised impact of NPIs on health outcomes. We selected 28 (HC) and 54 (NH) (N)PIs, covering environmental interventions, physical activities, psycho-social interventions, special therapies, preventive measures, special aids, nutritional interventions, regular care and advanced care planning. For multiple NPIs, interpretable causal-effect predictions were internally validated. Overall, most group-level effects were modest, but the models were able to clearly distinguish between no or a positive impact of a treatment on specific individuals. Some of the (psycho)social interventions in the LTCF setting were externally validated. Results generally translated well between countries. We developed decision support guidance on the validated algorithms with potential end users.
Additionally, for WP4-6, the ethical implications of the AI system were self-assessed by the Trustworthy Artificial Intelligence (ALTAI) framework proposed by the European Commission, highlighting possible risks for practitioners, the importance of training to reduce bias and to ensure that the end-users are aware of the limitations of the models.
In WP7 we developed and tested a unified, multilingual e-platform designed to provide individualised risk predictions for patients with CCC. In a pilot conducted across 7 countries, we demonstrated that clinicians are willing and able to engage with AI-powered decision-support tools. We offer strong proof-of-concept for the clinical relevance and feasibility of individualized prognostications based on real-world data. The pilot validated the platform's potential to support personalized, data-driven decision-making for older patients with CCC, laying the groundwork for future refinement, scale-up, and health system integration.
In this period, WP8 focused on translating research into impact by targeting healthcare professionals, organisations, and policymakers across Europe through tailored dissemination activities, and the development of an exploitation plan. We organised a final symposium with a global audience of stakeholders— clinicians, researchers, policymakers, vendors, and other important actors active in elderly care. By highlighting 6 key lessons learned important for a better healthcare for older persons with CCC, and moderating two panel sessions, we demonstrated the project’s relevance and reach beyond its funded lifetime.
WP9 explored whether and how the Covid19 care period impacted functional decline and changes in mood over time.
High quality individualised decision support for prognostications and estimating impact of treatment of persons with CCC is expected to improve patient outcomes, decrease adverse effects, and allocate sparse resources in a more rational and equal way. We specifically push the state of the art regarding the methodology to estimate treatment effects in observations cohort data by finding innovative ways to control for bias by indication.
In I-CARE4OLD we addressed this challenge by focussing on better evidence informed decision making for persons with CCC concentrated in home care and nursing home settings. We created a unified, multilingual clinical decision support e-platform capable of producing real-time predictions based on interRAI assessments. The development of this operational tool marks a milestone in translating high-quality research models into a scalable, user-accessible clinical solution. We anticipate our results will have direct and indirect impact on Quality of life (by optimising appropriate (N)PIs, Quality of care (Access to high quality decision support facilitates professionals and patients to make better decisions) and cost of care (High quality prognostic information may support better informed decisions that may reduce acute admissions and postpone nursing home admissions).
In I-CARE4OLD we addressed this challenge by focussing on better evidence informed decision making for persons with CCC concentrated in home care and nursing home settings. We created a unified, multilingual clinical decision support e-platform capable of producing real-time predictions based on interRAI assessments. The development of this operational tool marks a milestone in translating high-quality research models into a scalable, user-accessible clinical solution. We anticipate our results will have direct and indirect impact on Quality of life (by optimising appropriate (N)PIs, Quality of care (Access to high quality decision support facilitates professionals and patients to make better decisions) and cost of care (High quality prognostic information may support better informed decisions that may reduce acute admissions and postpone nursing home admissions).