Fall Risk Model Shows Promise for Working-Age Adults in Mental Health Services

Most tools designed to predict who will fall focus on the elderly. A working-age person with depression or a learning disability who takes a tumble is often treated as an outlier, a statistical afterthought. Yet falls happen across all ages, and in mental health services they happen with enough regularity that clinicians need better ways to spot who is at risk. A team in northern England set out to test whether an existing prediction model called eFalls, built for older populations, could work for younger adults receiving mental health or learning disability care.

The researchers pulled data from 32,410 working-age adults—people between 18 and 65—treated within a large NHS foundation trust. Over the study period, about 2 percent of them fell and sustained a fracture. The team applied the eFalls model to this group, using the published mathematical formula to calculate each person's 12-month risk. On its face, the model performed well. It correctly distinguished between people at high and low risk with a discrimination score of 0.777, which in predictive modeling terms is considered good. But when the researchers looked more closely at the numbers, something was off.

The model was overestimating risk across the board. It predicted that more people would fall than actually did. This kind of miscalibration matters in clinical practice. A tool that cries wolf too often gets ignored. So the team applied a simple mathematical adjustment—a recalibration that reweighted the model's predictions without changing its underlying logic. The fix worked. After recalibration, the model's predictions aligned with what actually happened in the data. The discrimination remained strong, and clinicians could now trust the numbers.

But the story became more complicated when the researchers broke the data down by subgroup. For people with mental health conditions alone, the recalibrated model performed nearly as well as it did for the overall population. For women and men separately, performance held steady. Then came the learning disability group. Here, the model's ability to distinguish high-risk from low-risk individuals dropped noticeably. The discrimination scores fell to between 0.696 and 0.739. More troubling, the model systematically underpredicted risk in this population—it missed people who were actually at higher danger of falling.

This gap matters because people with learning disabilities face compounding vulnerabilities. They may have balance problems, take medications that affect coordination, or have difficulty reporting symptoms. A prediction tool that underestimates their risk leaves them unprotected. The researchers tested whether the recalibrated model would be useful in clinical decision-making across different risk thresholds, and it showed promise at thresholds between 10 and 25 percent—the range where clinicians typically act. Yet that promise came with a caveat: the learning disability subgroup remained a weak point.

The conclusion was measured but clear. The eFalls model, after recalibration, works reasonably well for working-age adults in mental health services. It can help clinicians identify who needs closer monitoring, who might benefit from fall prevention programs, who should be assessed for balance and gait problems. But for people with learning disabilities, recalibration alone is not enough. The model needs deeper refinement, tested specifically in this population, before it can be trusted to guide care. The researchers have shown that a tool built for one group can be adapted for another—but adaptation has limits, and those limits matter most for the people who are hardest to predict.