Beyond age: what linked data shows about who falls

The role of Linked data  

To answer that question, researchers required a more complete dataset than any single clinical system could provide. No individual healthcare record captures the full trajectory of a patient's risk and outcomes: information relevant to falls risk is often recorded in primary care, while the fall itself, and its clinical consequences, are documented separately within hospital systems.  

By linking data from the Whole Systems Integrated Care (WSIC) programme and Imperial College Healthcare NHS Trust, researchers were able to follow patients across different parts of the North West London healthcare system and build a more complete picture of risk. This made it possible to look beyond individual events and understand how factors such as age, long-term conditions and deprivation interact to influence falls. 

What the data revealed 

Using this linked data, researchers applied an unsupervised clustering approach, a machine learning technique that allows patterns to emerge from the data itself, rather than relying on predefined age or risk categories.  

Some findings reinforced what clinicians already know. The oldest and most clinically complex patient groups experienced the highest risk, with approximately 11 to 12 times greater likelihood of a fall-related hospital encounter compared with lower-risk groups.  When falls did occur, these patients also tended to have longer hospital stays.  

Other findings were more surprising. Preliminary analysis of 9,684 inpatient fall events suggests that around a third of all fallers are under 70, younger than the current screening threshold. "The next step is to understand whether those patients are actually missing from the current screening approach, and whether other characteristics could be better signals of risk," says Joy Li, an iCARE data scientist working on this project. 

Among younger patient groups, those living in more deprived circumstances experienced a 29% higher risk of a fall-related hospital encounter, even when they did not have the same level of clinical complexity seen in older populations. 

Taken together, the findings suggest that falls risk may be influenced by a wider range of factors than age alone. 

The next steps: from predicting falls to preventing them 

For the researchers, identifying higher-risk groups is only the beginning. The next stage of the project turns to pre-falls assessment: improving the accuracy and predictive power of risk assessment before a fall happens, not just after. " This data can help us understand where, how and why those falls happen, so that prevention can be more specific to different patients and clinical environments." says Joy. 

The team is now working to quantify the proportion of inpatient fallers who had received a formal falls risk assessment prior to their fall, broken down by age group, risk level ward and site. A central question is whether the four faller groups identified through analysis, differ in how consistently they are assessed before a fall, and in particular, whether younger groups are being systematically missed. 

The longer-term vision is integrating this into clinical workflows, so frontline staff can use risk insights in real time. Dr Ekin Yağış Research Associate

The aim is to build an evidence base that can support revising the current age-based screening threshold, and identifying alternative triggers, such as cognitive impairment or a history of recurrent falls, that could extend coverage to patients currently excluded from the screening programme. 

"There's real momentum around operationalising this: working out how to make these models actually useful in clinical practice, not just in research papers," says Dr Ekin Yağış, data scientist at iCARE. "The longer-term vision is integrating this into clinical workflows, so frontline staff can use risk insights in real time." 

The ambition is not simply to build better risk models, but to support more targeted prevention. If different groups of patients experience falls for different reasons, then a one-size-fits-all approach may miss opportunities to intervene earlier. 

"If we can identify which patients are at highest risk, understand why, and translate that into ward-level or patient-level interventions that actually work," says Ekin, "that changes lives." 

Could this apply to your work? 

This project relied on linking data that would otherwise stay siloed, across primary and secondary care. If your team is working with fragmented healthcare data and thinks that data linkage could help, get in touch with iCARE →