Importantly, there have been no prospective studies of multifactorial risk factors for falling specific to the population with dementia. Previous studies have identified multiple risk factors for falls in the older population as a whole, in a variety of settings. In multivariate studies significant risk factors include fall history, gait, balance and mobility impairments, visual impairment, cognitive impairment, fear of falling, environmental hazards, muscle weakness and incontinence.[9] In community settings, impairment on various cognitive measures has been shown to be a risk factor for falls, although dementia itself has only been examined and shown to be a risk factor for falls in community dwelling people with Parkinson's disease,[10] and in residents of extended care settings.[7], [11] The identification of potentially modifiable risk factors has been critical in the development of effective multifactorial falls intervention programmes, particularly for older people at high risk of falling.[12] It is likely that falls in dementia are also multi-factorial in origin, possibly with risk factors similar to those identified in the general older population, but there may be other potentially modifiable factors specific to dementia. It is thought that those with Lewy body (LB) dementias are at particular risk, and a history of recurrent falls is accepted as a supporting feature for the diagnosis of dementia with Lewy bodies (DLB).[13]

Only one previous multifactorial intervention study has exclusively recruited participants with dementia.[14] Interventions were based upon those used in successful trials in people without dementia, and the study did not show a significant reduction in falls or number of fallers. However, 70% of the participants in this trial lived in institutional settings, most had moderate to severe cognitive impairment and all had already presented to hospital with a fall. A recent systematic review of interventions to prevent falls in hospitals and care homes suggested that effect size was smaller in trials where the prevalence of dementia was high, but meta-regressions were limited by incomplete reporting of dementia prevalence.[15] It remains possible that multifactorial interventions would be successful in patients with less severe dementia, but we suggest that interventions in future trials should be tailored to potentially modifiable factors which can be shown to predict falls in community dwelling people with mild-moderate dementia. There is therefore an urgent need to understand risk factors for falls in dementia to enable a more tailored and effective intervention to be developed.

To our knowledge, this is the first study which has identified predictors specific to dementia, including the identification of non-modifiable predictors such as a diagnosis of Lewy body disorder, longer duration of dementia and previous history of falls or recurrent falls. These factors will be useful in identifying individuals at particular risk, who may benefit from further assessment and intervention. Even more importantly, a number of the predictors identified are potentially modifiable, and should be included as key elements of a multifactorial intervention. These factors included use of cardioactive medications, autonomic symptoms, symptomatic orthostatic hypotension, depression and limitation of physical activity. We suggest that interventions targeted towards these predictors could reduce the burden of falls related morbidity and mortality in community dwelling people with mild-moderate dementia.

Our study has a number of strengths which reinforce our findings. This was a prospective study and the majority of participants had mild-moderate dementia and were residing in the community. The incidence of falls was determined by completion of daily diaries, with the support of a caregiver where appropriate, with an adequate follow up period. The use of this method is particularly important because it has the highest sensitivity for accurate recording of falls. The number of participants in our study was high (179); 80% of those approached agreed to take part and compliance with fall diary returns was also high. Statistical analysis was strengthened by the use of Cox regression models and loglinear analysis of incidence densities, thus taking account of censoring of falls diaries in some participants. All analyses included adjustment for age and gender and the inclusion of multivariate analyses enabled efficient estimate of significant associations while adjusting for a number of confounding factors simultaneously.

There are some potential limitations to this study. Ideally, we would have recruited both groups using random sampling from a community population. Unfortunately the scale of such an exercise would have been well beyond the resources available for this study. Healthy older people volunteering for participation in research in response to advertisements are often fitter than the older population at large, but interestingly the incidence of falls in our control group was very similar to that found in most community based studies, suggesting that the control group was reasonably representative, despite the limitations of using this recruitment method. The recruitment of the patient participants from secondary care clinics rather than from primary care may have biased the sample towards those with more progressive disease. However, in the northern region of the UK, suspected cases of Parkinson's disease are routinely referred to a secondary care physician specialising in movement disorders (either a Neurologist or Geriatrician). These patients should, therefore, be reasonably representative of all patients presenting to primary care physicians with symptoms suggestive of PD. There is a greater likelihood of patients with AD or VAD not being referred to secondary care as dementia is often undetected, or not managed by a specialist in memory disorders. The reader should therefore bear in mind that participants in this study may have had more severe or progressive disease than those generally seen in primary care. In interpreting our results, consideration should be given to the need to stratify for dementia subtypes in univariate and multivariate analyses. Some of the predictors were not significant when analyses were stratified by dementia subtype, probably because these predictors were actually surrogate markers of a diagnosis of a Lewy body disorder, which itself predicted falls. In comparison with the proportions in the general population the LB dementias were over represented in our study, in order to make valid comparisons between dementia subtypes.

Prior to commencement of our study we identified only two studies of falls in dementia which included a fully multifactorial assessment of risk factors, the first of which had an inadequate follow up period,[34] and in the second the primary outcome was fall related serious injuries rather than falls.[35] Both of those studies included fewer than 100 participants, and neither included a control group. Two studies published since we commenced our study have included multifactorial baseline assessments, but the first of these was small (only 42 participants with advanced AD).[36] The second was a high quality study of 124 participants in which neuroleptic drug use and grade 2 white matter lesions predicted falls in multivariate analyses, but only participants with AD were included, and there was no control group[37]. The design of our study addressed these issues; to our knowledge, it is the first to examine falls in both AD and non-AD dementias over an adequate follow up period.

Exclusive to our study was the phasic measurement of blood pressure using beat-to-beat recordings, thus enabling detection of both magnitude and timing of blood pressure changes. Symptomatic OH and the length of time blood pressure fell below baseline rather than the magnitude of the drop were predictive of falls. Our group has previously shown an association between intermittent hypotension and white matter lesions,[38] and this resonates with the findings of Horikawa et al. that grade 2 white matter lesions predicted falls in multivariate analyses[37]. White matter lesions can be associated with abnormal gait,[39] and it is possible that intermittent symptomatic hypotension against a background of gait and balance instability exacerbates the likelihood that an older person will fall. A higher level of physical activity was a protective factor in our study. The physical activity scale measured all physical activity rather than wandering, which has been shown in one study to be a predictor of falls in dementia.[40] It is likely that the more active patients were those with fewer problems with mobility, whereas the participants with disturbed mobility restricted their activity because of fear of falling, thus having fewer opportunities to fall.

We believe that randomised multifactorial intervention trials to prevent falls in mild-moderate dementia should now be made a priority. Possible management strategies could include management of the potentially modifiable factors identified in this study; for example, the use of selective serotonin reuptake inhibitors for depression, manipulation of cardiovascular medications, adequate hydration and targeted drug therapies such as fludrocortisone and midodrine for OH. Such a focus would differ from multifactorial interventions in older people without dementia, which prioritise strength and balance exercises that are more difficult for those with impairment of recall to continue following initial intervention. It is possible that encouragement of overall physical activity may be successful in prevention of falls in dementia. However, such an approach may increase opportunities for falling in individuals at risk; similarly, aggressive treatment of motor features in LB dementias might increase activity related opportunities to fall, and also exacerbate OH. There is also a possibility that changes in psychotropic medication might result in side effects such as hypotension or somnolence, which could paradoxically increase the risk of falls. This emphasises the importance of the conduction of randomised controlled trials to ensure that modification of the risk factors identified is the correct strategy.