Assessing and reporting population-based incidence of falls is on the face of it straightforward. A number of epidemiologic studies [1–6] and a number of fall prevention trials [7] have reported fall incidence for different populations and have been reviewed elsewhere [8, 9]. Several of these studies (and their fall incidence estimates) are particularly often cited in comments, introductions or overviews [10–12]. Fall incidence, however, differs considerably between studies. In community-living persons aged 65 years and above, for example, fall incidence lower than 20% and higher than 30% has been reported [1, 13, 14]. One potential reason for these differences is that the study samples from which the incidence rates were derived represent different populations with different fall risks. They may be convenience samples, trial populations, subpopulations such as residents from nursing homes or participants of epidemiological cohorts.

In order to obtain accurate fall incidence rates, representative population-based samples with high participation rates are required. In western industrialised countries, however, high participation rates are increasingly difficult to achieve [15]. Despite considerable effort, participation rates in many studies remain unsatisfactory. This results in selection bias in participants, which is particularly problematic in the field of falls epidemiology since typical selection criteria for older people may be strongly associated with fall risk. Persons with mobility restrictions, for example, may decline participation if they have to go to a study centre. Cognitively impaired people may be generally underrepresented in studies since they are not able to respond to a study invitation or to complete a falls calendar regularly.

Furthermore, estimates of fall incidence in the literature are based on different assessment methods. A consensus expert meeting recommended prospective daily recording and adequate surveillance of documentation and ascertainment of details of falls at least once a month over lengthy periods (normally 6–12 months) [16]. But this is time-consuming and costly and often only feasible in relatively small and possibly unrepresentative samples. Therefore, surveys and epidemiologic studies frequently apply retrospective assessments, which may be biased by poor recollection of past falls. In publications fall rates are reported for different time periods (4 weeks, 3 months, 12 months), are calculated by different methods (incidence of fallers, incidence of frequent fallers, incidence rates), or are presented stratified by different age-categories (5- or 10-year strata) which further complicates comparison. For Germany, population-based data about fall incidence are rare. Only one study reported fall incidence so far, based on a retrospective telephone survey [17].

Our study presents fall incidence in community-living German women and men aged 65 years and older, stratified by five year age-group. Data were derived from (i) a national survey and (ii) a geriatric cohort study performed in southern Germany. Both studies aimed to be population-based. Additional objectives of this publication are to describe the influence of a) different study approaches and b) retrospective and prospective assessment methods on the reported incidence of falls.

The data for our analyses were derived from (i) the German health interview and examination survey for adults (DEGS1) and (ii) the Activity and Function of the Elderly in Ulm study (ActiFE-Ulm). Both studies were undertaken in Germany and are population-based.

DEGS1 is part of the continuous health monitoring system in Germany and was conducted by the Robert Koch Institute between November 2008 and December 2011 [18]. It was designed to provide nationally representative health data for the population 18–80 years of age. For Germany this was the first time that falls were assessed as part of a national health examination survey in the 65–80 year age groups.

The ActiFE-Ulm study is a cohort study of older people (65- <90 years), located in Ulm and adjacent regions in Southern Germany. The baseline recruitment took place between March 2009 and April 2010 [19].

The study protocols of DEGS1 and ActiFE-Ulm were approved by the Charité-Universitätsmedizin Berlin ethics committee in September 2008 (No. EA2/047/08) and the Ulm University ethics committees in January 2009 (No. 318/08), respectively. Both studies are more fully described elsewhere [18–20]. Here we focus on those aspects of the study designs of direct relevance for this paper.

We used two different population-based datasets to assess fall incidence in German people aged 65 years and older and living in the community. Fall incidence differed between the two studies. For ‘any fall’, the rate was consistently higher in women than men, and increased with age in men but not consistently so in women. Rates of recurrent falls were similar across the two studies and confidence intervals overlap. There was no difference between retrospectively and prospectively assessed fall incidence rates.

The descriptive data from our two datasets broadly correspond to fall incidence reported in the literature. The LASA study, for example, presented very similar age- and sex-specific fall incidence rates to those observed in the ActiFE-Ulm study [1]. In a survey from the United States which analysed a nationally representative sample of Medicare beneficiaries similar fall rates were observed to DEGS1 [24]. Recently published data from a German telephone survey presented lower fall incidence (17.9% in people aged 65 years and more) [17].

One explanation for the differences in fall incidence estimates for the same target population may be a different selection of people with different fall risks. Since the studies reported here, and Schumacher et al’s [17] telephone survey were designed to be population-based, study design and the way the study population were approached, seem to influence results considerably. The DEGS1 study asked participants to visit a study centre and may therefore represent fall incidence of older people who are still mobile enough to participate; that is a more able and higher functioning section of the population. In the ActiFE-Ulm study assessments were performed at home, which made it easier to participate for persons with mobility restrictions. Even though both studies made considerable efforts to increase participation, the actual participation rates achieved in both studies may be too low to guarantee representativeness in higher ages. For example, persons from DEGS1 aged 65–79 years had a slightly less frequent rate of officially recognized severe disability than persons from the population (20.0% vs. 22.9%) [20]. And people with disabilities in activities of daily living and with care need as assessed by German long-term care insurance criteria, but living at home were underrepresented in the ActiFE-Ulm study (e.g. 0.7% of participants aged 70- < 80 years compared to the 5.1% one would expect based on age specific German population data). Their risk of a femoral fracture is far higher than that of community-living people without assessed care needs, but very similar to that of residents of nursing homes [25]. Therefore, the fall risk of older disabled people living in the community may resemble the very high risks observed in nursing home residents [26, 27] rather than that of community-living people without disabilities. This means that the ‘true’ fall incidence of community-living people could be higher than that presented here. Even though sample sizes of both studies were each clearly larger than 1,000 participants, the statistical power was still too low to compare the estimates of the different strata with each other.

One might expect that the home interview method used in the ActiFE-Ulm study could yield different incidence rates to those obtained using assessment simply by questionnaire, as in the DEGS1. But it is not clear which method would necessarily yield higher rates and which would be closer to the true rate. So the observed differences for ‘any fall’ could be an artefact of method. However, this explanation does not hold true for the assessment of ‘recurrent falls’, where the rates cannot be said to differ between assessment methods.

Collecting falls data using prospective methods is time consuming and costly. Face to face visits by research staff are required to ensure the older person understands how to record falls on a diary or calendar. Whilst returning calendars by post is a relatively modest expense, during the prospective follow up period staff is needed to follow up participants who fail to return calendars, and to contact those who report falls to obtain further data about the fall. The retrospective method can be done during a single interview, or even, as in the DEGS study, by questionnaire. Our results would suggest that one obtains as accurate an estimate of the incidence of falls using a single visit interview as one does using the more costly 12 month prospective approach, although neither can be considered a “gold standard”, and further methods for collection of falls data need to be developed.

There is a risk when retrospectively assessing falls that the data will be inaccurate because of recall biases. One review determined the effect of the recall interval on fall reporting [28]. As expected, longer recall periods resulted in poorer fall reporting particularly in persons with poorer cognitive function. In demented people, recall periods should be even as short as one week and additional information by care-givers is required [29]. Reports of falling in the previous 12 months by people who did not actually fall are rare [28]. In the ActiFE-Ulm study, however, retrospectively assessed data of any fall were somewhat higher in the oldest age-groups. This may be due to impaired time awareness amongst older people, who may mistakenly report fall events from previous years to the last 12 months. Another explanation could be that older people are more willingly to report falls from the past than those from the present since they may be afraid that report of current falls could result in unpleasant consequences like hospitalisation, institutionalisation and loss of independence.

One limitation of our analysis could be that study participants’ retrospectively and prospectively assessed falls did not cover the same time period. But the problem with having a retrospective assessment which covers the same time period as prospective assessment is that participants would be primed to recall falls over that period. Therefore, using two subsequent time periods may actually be a more reasonable way to compare retrospective and prospective assessed fall incidence, although it assumes that the incidence rate remains steady over the two time periods.

A further limitation is that both prospectively and retrospectively assessed falls relied on self-report, which was not verified by such methods as crosschecking with medical records for injurious falls. Previous studies have validated reported falls by checking for other data to confirm and suggest that the method is reliable. To date there is a paucity of data on actual falls, and the literature is dependent on self-report methods and there is no gold standard to provide criterion validity. However, this is changing fast and the use of body worn or body fixed fall sensors with high sensitivity and specificity in community based epidemiological studies is becoming a reality [30]. Other technological solutions include use of video recording and other environment sensors to detect and record falls [31]. Clearly such instrumented falls offer promise for future research, but even they depend on report methods being used at least during the methodological validation stage and more research is required to determine reliability and validity of instrumented falls alongside self-report.

In conclusion, fall incidence among community dwelling people aged 65 years and older was assessed in two population-based German studies, and this gave us the opportunity to also compare different fall assessment methods. Overall fall incidence of ‘any fall’ assessed retrospectively, but not recurrent fall rates, differed between the two studies, consistent with both the very different possibilities that true rates differed between the two populations and that they were artificial differences because of methodological differences between studies. In the ActiFE-Ulm study, retrospective and prospective methods to assess fall incidence over different assessment periods provided strikingly similar results.

Population-based studies are likely to underestimate true fall incidence among community residents as they have difficulty capturing high risk groups such as physically and cognitively impaired persons, and those depending on home care. Nevertheless, incidence rates of ‘recurrent falls’ observed for the German population residing in the community are considerable and deserve further investigation with respect to determinants and health consequences.