Background:

Comprehensive Geriatric Assessment (CGA) is a cornerstone of patient-centred care for older adults, yet its implementation in primary care remains challenging. Results from the CEpiA study, a cluster-randomised trial evaluating the 12-month effectiveness of CGA, and other studies suggest that CGA may not be equally relevant for all patients, highlighting the need for better targeted approaches.

Research questions:

To identify frailty profiles among patients managed in primary care based on clinical data and CGA results from the CEpiA study, and to characterise these profiles in terms of prognosis and patterns of care.

Method:

This observational study included patients from the CEpiA study who underwent CGA. An unsupervised clustering approach based on Self-Organizing Maps (SOM) was applied. Clusters were characterised according to clinical features, care processes, and 12-month outcomes (hospitalisation, emergency visits, institutionalisation, mortality)

Results:

Among 369 patients, SOM analysis identified six homogeneous frailty clusters: “Relatively preserved older adults” (Cluster 1), “Autonomous men with cancer and cardiovascular comorbidities” (Cluster 2), “Neuro-psychological and pain-related frailty” (Cluster 3), “Polymedicated osteoporotic women” (Cluster 4), “Multimorbidity with social vulnerability” (Cluster 5), and “Very old adults with severe cardio-respiratory multimorbidity” (Cluster 6). A notable finding was that Cluster 3 combined high nutritional risk with no increase in nutritional interventions.
These profiles were grouped into three macro-profiles showing an increasing gradient of clinical and social complexity: a “relatively preserved” profile (Cluster 1) with low morbidity and limited care needs; an “intermediate frailty” profile (Clusters 2-4) with contrasted needs; and a “high-risk frailty” profile (Clusters 5-6), concentrating the highest levels of care use, hospitalisations, and 12-month mortality.

Conclusions:

Based on a large real-world primary care sample and a robust unsupervised clustering approach, these results support frailty profiling to personalise care. However, this study included selected patients and GPs, some missing data, and cross-sectional profiling linked to 12-month outcomes, limiting causal inference and generalisability.

Points for discussion: