Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index

Predicting falls in older people in the community


Bandolier 85 examined predictors for falling in hospital. What many people asked for was a set of predictors for falling in the community. A study from Holland [1] goes a long way to providing that, as well as explaining how studies of diagnostic or prognostic algorithms can be done.


In Amsterdam, a ten-year study of ageing in older people (55-85 years) collected information on a random sample of people in this age range, representing the older Dutch population. A sub-sample of 1420 those aged 65 or more was chosen, of whom 1374 were willing and able to participate in a study of falling over one year.

A very wide range of sociodemographic variables were collected at baseline, as well as self-reported information, and medication information taken from containers used, and included prescribed and non-prescribed medicines. Assessment of physical functioning included questions on problems with feet and muscles, dizziness, and visual impairment. Functional limitations were considered to be present when participants had difficulty with at least two activities - climbing stairs, using their own or public transport, or cutting toenails. Poor vision was determined by ability to recognise a face at a distance of four metres, with glasses or contact lenses if used.

Falls over one year were determined by use of a falls calendar, completed weekly and mailed every three months, with telephone reminders.


Participants had a mean age of 75 years. All four three month periods were completed by 94% of participants. At least one fall occurred in 33%. A single fall occurred in 22% and recurrent falls in 11%. Fractures were more common in recurrent fallers than in single fallers or those who did not fall (Figure 1). Recurrent falls were more common in older men, but not older women.

Figure 1: Fracture rates with no falls, single falls and recurrent falls

A number of variables were associated with falls in univariate analysis, but only a small number for multivariate analysis. Regression coefficients from logistic regression were converted to scores (Table) for these factors.

Table 1: Predictive factors and scores for single and recurrent falls

  Single falls Recurrent falls
Factor Points Points
Previous falls 5 5
Urinary incontinence 2 3
Visual impairment 2 4
Benzodiazepine use 2 3
Maximum 11 15

The predicted percentage of recurrent fallers for each score is shown in Figure 2. At a score of 7 points, where 15% of recurrent fallers would be expected, the positive likelihood ratio was 2.6 and the negative likelihood ratio was 0.58.

Figure 2: Predicted levels of recurrent falls with increasing scores


There's good methodology here. In Mr Punch's words, "that's the way to do it". There is a table of sensitivities and specificities so that the impact of various risk scores can be examined, either in terms of likelihood ratios or absolute risk.

How useful is it? Firstly it is simple, with four easily-remembered keys to risk of recurrent falls. If an older person has some or all of these, then it may be worth a hard look at things that might help. For instance, an older person with previous falls, visual impairment and on benzodiazepines has a score of 12 and a one-year risk of recurrent falls of 1 in 3. That's 1 in 1 over three years. So thinking about alternatives to, or no benzodiazepines, or suggesting changes in the home or institutional environment, or alert systems in case of a fall might all make sense.

Assessing older people in the community for their risk of falling could be a useful feature of clinical governance or quality improvement. It is a big problem ( Bandolier 20 ).


  1. 1 AM Tromp et al. Fall-risk screening test: a prospective study on predictors for falls in community-dwelling elders. J Clinical Epidemiology 2001 54: 837-844.
previous story in this issue