Bottom-up subspace clustering suggests a paradigm shift to prevent fall injuries

Abstract

Despite over 10,000 publications since 1990, fall injury rates for older people are still increasing, over and above population ageing. Developing new ways to explore highly dimensional health data and better understand high risk individuals is imperative. The hypothesis investigated is: Falls are a complex multi-systems medical problem. And a paradigm shift in statistical methods is required before fall injuries can be substantially reduced. Here, a new bottom-up supervised subspace clustering (BUSSC) approach suggested as one alternative to conventional approaches. Pilot data were used from 96 community-living older people, 35 with a history of falling. Analysis of variance (ANOVA) found no significant group differences between fallers, and non-fallers. Conversely, BUSSC identified many significant interactions between risk factors that might cause different subgroups of older people to fall. A BUSSC model identified 100% of fallers (Kappa 0.73), comparing favourably to results published from similar populations. BUSSC's superior performance suggests developing new statistical methods should be investigated. Different to previous fall risk models, BUSSC does not require all cases to be classified; instead each cluster provides information most relevant to a homogeneous subgroup of people. However, the three interactions documented may only be small pieces of a larger puzzle. To definitively prove the hypothesis: orders of magnitude more participants should be recruited, prospective falls recorded, and interventions prescribed based on an improved understanding of the individual. A paradigm shift in statistical methods could have profound consequences for health care, allowing us to better understand the individual and focusing less on 'average' population benefit. This knowledge may help develop more individualized treatments for many conditions.