With an ageing population, increasingly complex health needs, and growing pressure on emergency and inpatient services, shifting care into the community is more imperative than ever. Akeso partnered with South Eastern Health and Social Care Trust to assess and quantify the potential to scale its Hospital at Home (HaH) service.
The HaH service performs well and delivers strong outcomes, particularly for community-based hospital avoidance, where majority of the referrals come from. However, there was an opportunity to measure true demand, by increasing the visibility of eligible patients within inpatient wards and ED who are also suitable for the service.
Identification and onboarding of appropriate patients currently present two key challenges. First, the service is consistently filled via community referrals, limiting capacity to accommodate suitable patients from ED and inpatient settings. Second, identifying and referring patients from wards and ED remains timeintensive, and there is inconsistent understanding among clinicians of the acuity and risk profile the HaH service can safely manage. As a result, clinically appropriate patients are not identified consistently or efficiently, constraining the service’s overall impact and reach. Akeso combined clinical engagement with advanced analytics and machine learning to understand the true scale of unmet demand and begin to build the evidence base for scaling the service.
Methodology
Our approach combined clinical engagement with advanced analytics across two workstreams:
1. Clinical Engagement
Working closely with clinical teams, we reviewed inpatient and ED pathways to understand how patients were currently identified and referred, and where opportunities for earlier intervention existed.
2. Advanced Analytics
In parallel, we developed a Machine Learning XGBoost prioritisation model using historic activity data to learn the characteristics of patients accepted onto HaH, resulting in a machine learned “suitable patient”. These outputs were validated with clinicians to ensure alignment with real-world decision making, before applying the model across inpatient and ED cohorts to identify patients who were clinically suitable but not referred. We then built a prototype prioritised patient list to demonstrate how the model could be used operationally, and modelled demand and capacity implications to support discussions on service scale.
Impact
The XGBoost model we developed predicted clinician referral decisions with 83% accuracy, revealing significant unmet demand across both ED and inpatient settings:
✓ 3,933 patients (3% of all ED attendances) were suitable for HaH but were not referred in 2025
✓ 3,189 patients (11% of Ulster inpatients) were suitable for HaH but were not referred in 2025
✓ 180 patients were eligible and referred to HaH but could not be admitted due to capacity constraints, representing a potential 23% increase in admissions if capacity were not limited
✓ If all suitable patients were admitted to HaH, 53 inpatient beds (10% of total inpatient beds) could be freed, making up half of the national 20% bed reduction target
Key takeaways / conclusion
This project demonstrated that a significant proportion of patients currently occupying inpatient beds or attending ED could be safely and appropriately managed through Hospital at Home, if the service had the capacity to meet demand. By combining clinical engagement with machine learning, we provided the Trust with a robust, evidence-based platform to make the case for scaling the service and realising meaningful benefits for patients, clinicians, and the wider health system.