
Home Research Community acquired pneumonia Pre-op screening Legionella detection Virtual Sentinel Node Falls Risk Assessment CV Line Cost Benefit Health Systems Analysis Topics Courses People Links

Health Informatics Unit
The University of Adelaide
SA 5005 Australia
Email

Telephone:
+61 8 8274 3713
Facsimile:
+61 8 8271 9158

Risk Assessment of Falling using Computer-Aided Modelling

Little, K., Pradhan, M. and Cleland, L.

Introduction

Falls are common in the elderly. Interventions can reduce the risk of falling (e.g. home modification, balance exercise programs) or protect against the outcomes of falling (e.g. hip protectors). These interventions should be targeted towards individuals at risk of falling in order to optimise the cost-benefit of intervention. However, an effective risk assessment tool for falls has not been developed to date.

Methods/Results

A risk assessment form was developed for a hip protector implementation project with which the authors were involved. The form was compiled from an informal literature review and consultation with an expert panel. It was designed as a risk assessment for hip fracture and therefore included falls and osteoporosis risk assessment. An arbitrary scoring system was applied to each risk factor and the scores were tallied to create a risk assessment score. Although this gave a rough estimate, a more accurate assessment is desirable.

Figure 1: Risk Assessment form

A computer-assisted Bayesian belief network, which modelled falls risk, was created (see figure 2). The model was based on the results of a formal systematic review of the literature concerned with falls risk factors. Risk factors were represented in the model by chance nodes. The arbitrary scores from the risk assessment form were compared with the evidence-based calculated probabilities from the model. Sensitivity analysis was used to examine the influence of individual risk factors. Many of the risk factors are not independent. An important difference between the risk assessment form and the model is the effect of dependencies (or interactions) between risk factors.

Figure 2: Partial Bayesian belief model for risk of falls.

Figure 2 shows a partial belief net with dependency incorporated via an intermediate node. We know these 3 risk factors interact. Almost by definition, people living in residential care are unable to attend to their own Activities of Daily Living and many suffer from incontinence. Therefore, these variables are dependent. In fact, each is an indicator of incapacity that is likely to be the underlying factor increasing falls risk. The risk assessment form's arbitrary strategy of addition does not adequately reflect this dependency. Each factor contributes one point, regardless of whether or not the other factors are present. Therefore, the form will be overestimate risk.

Another limitation of the forms additive system is an inability to adequately represent proportional increases in risk ratio. For this reason, we compared the weight associated with risk factors in the form with the risk ratio from the meta-analysis (figure 3). Figure 3 shows the weight of factors on the assessment form on the x-axis, expressed as a percentage of the maximum possible score. On the y-axis is the pooled risk ratio taken from the meta-analysis results. There are two main examples of the difference in weights given by the assessment form in comparison to the evidence. According to the assessment form the falls risk of being female is about 2X that of sedative use, however meta-analysis indicates only a small difference in the risk ratios on the y-axis. Second, a number of risk factors carry a similar weight on the form (about 11%), but their risk ratios for falling differ greatly. Therefore, the form's weighting of risk factors is not as accurate as the model's which are derived from the evidence.

Figure 3: Comparison of risk assessment form weight to pooled risk from meta-analysis