Abstract

Introduction. Falls are frequent in older adults and may have serious consequences but awareness of fall-risk is often low. A questionnaire might raise awareness of fall-risk; therefore we set out to construct and test such a questionnaire. Methods. Fall-risk factors and their odds ratios were extracted from meta-analyses and a questionnaire was devised to cover these risk factors. A formula to estimate the probability of future falls was set up using the extracted odds ratios. The understandability of the questionnaire and discrimination and calibration of the prediction formula were tested in a cohort study with a six-month follow-up. Community-dwelling persons over 60 years were recruited by an e-mail snowball-sampling method. Results and Discussion. We included 134 persons. Response rates for the monthly fall-related follow-up varied between the months and ranged from low 38% to high 90%. The proportion of present risk factors was low. Twenty-five participants reported falls. Discrimination was moderate (AUC: 0.67, 95% CI 0.54 to 0.81). The understandability, with the exception of five questions, was good. The wording of the questions needs to be improved and measures to increase the monthly response rates are needed before test-retest reliability and final predictive value can be assessed.

1. Introduction

Falls are a common cause of accidents and they can have serious consequences ranging from fear of falls to fractures, loss of independency, or even mortality. Approximately 25% of people over 65 years of age and living at home fall each year and about 20% of the falls require medical attention [1]. Mortality after a falls-related hospitalisation is high [2] and the falls specific mortality is still rising, although the mortality due to fractures after falls is declining [3]. Forty percent of the admissions to a long-term stay in a nursing home are due to a fall. Therefore, prevention of falls or their consequences is important. There exist a plethora of known risk factors for falls [4, 5] and the risk factors generally increase with age. However, older people are often not aware of their own fall-risk [6]. They are aware of the increased fall-risk of other elderly persons, but they are often convinced that this does not apply for themselves [7]. Furthermore, some older adults are reluctant to admit that they are at risk for falls because they fear that their families might send them to nursing homes [8, 9]. Health professionals, such as nurses or physiotherapists, might play an important role in raising the awareness of the fall-risk.

Screening for falls is usually performed by a health professional. However, in the group of the “young old,” not all are regularly seeing health professionals, or they are seeking care for other health conditions and the potentially increased fall-risk is not recognised or not perceived as an issue and not targeted by them or the health professionals [10]. A self-assessment tool might increase the awareness of the fall-risk and the motivation to discuss the problem with a health professional and to start a preventive programme [11, 12].

Current self-administered predictions tools do not cover all dimensions of fall-risk, such as dual tasks, medication, diseases like diabetes, pain, stroke, rheumatic disease, fear of falling, the frequency of toileting, gait problems, balance, muscle weakness, sensibility impairments, or hearing problems [1317] (see also Table 1 for a comprehensive overview of existing tools).

Therefore, based on a search for systematic reviews and meta-analysis on risk factors for falls, we set out to (a) collect risk factors that were consistently reported in studies, (b) to extract coefficients from predictive models, (c) to devise a comprehensive set of questions, and (d) to test, in a sample of community-dwelling persons aged sixty years or older, the feasibility, understandability, calibration, and discrimination using the extracted coefficients, including the continuous assessment of falls during a six-month follow-up period. We hypothesize that (a) the monthly response rate is higher than 80%, (b) that the understandability of the questions is good, (c) that the self-predicted fall-risk is not in agreement with the observed fall-risk, (d) that the observed fall-risk is associated with the predicted fall-risk, and (e) that we can discriminate between fallers and nonfallers based on the risk score calculated with the coefficients from the literature and our self-reported questionnaire.

2. Materials and Methods

This study included several steps: (1) defining a set of predictors for falls based on published meta-analyses, (2) devising a set of questions for the self-assessment of the risk factors out of seven questionnaires, and (3) prospective cohort study to assess the feasibility and the preliminary predictive values of the online assessment of the fall-risk.

2.1. Defining the Set of Predictors

We searched in PubMed for systematic reviews and meta-analyses on risk factors in community-dwelling elderly people; search strategy: (((risk OR odds OR OR likelihood OR sensitivity OR specificity OR AUC OR ROC OR calibration OR discrimination))) AND ((((((falls [title]) OR fall [title]) OR faller [title])) AND ((meta-analysis [Publication Type]) OR systematic review [title])) AND ((elderly OR older OR aged OR senior OR seniors))). Inclusion criteria were systematic reviews and meta-analysis on prospective cohort studies including community-dwelling elderly persons. We extracted the risk factors for falls that were statistically significant in the meta-analyses. For each factor we extracted the coefficients (i.e., log of the odds ratio) for the prediction of falls in community-dwelling older adults from the meta-analysis with the most included participants or studies for the given predictor.

2.2. Devising the Set of Questions

Based on seven existing questionnaires for the self-assessment of fall-risk [1317, 44, 45], we devised a set of questions that covered most of the fall-risks found in the previous step (review of reviews). The questions were written in German and translated into French and submitted to seven health professionals and two laypersons with the question about the understandability. Amendments were made if necessary.

We included ten questions about personal characteristics and a question about the self-perceived risk of falling, as well as the understandability (comprehension of the questionnaire) and suggestions for different formulations.

The questionnaire was implemented in an online survey system (SurveyMonkey [46]).

For the monthly follow-up we assessed whether a person fell during the last months and the number of falls. A fall is often defined as “an event which results in a person coming to rest inadvertently on the ground or floor or another lower level.” [47]. For our study, we decided to exclude falls in sports activities such as biking, skiing, or mountaineering. Based on feedback from participants at the first monthly follow-up, we added a question about the activity at which the falls occurred and two questions to assess the level of physical activity as recommended by Gill et al. [48] for the later follow-ups.

2.3. Prospective Cohort Study

The main part of this study was a longitudinal cohort study with a six-month follow-up (falls assessment and assessment of physical activity). Study participants were community-dwelling elderly persons aged 60 years or more. They had to be able to walk independently, with or without walking aids. German and French speaking participants were included if they had an e-mail address.

Participants were recruited by a snowball-sampling method [49]. This method allows the inclusion of participants that are difficult to achieve. If our hypothesis is true that our target population has a low awareness of their risk to fall, they would, for example, most probably not respond to other sampling methods such as information leaflets or advertisements in journals. Other sampling methods such as phone number lists are nowadays not valid anymore, because a large subset of the population is not listed in directories (phone books). A first set of e-mails with a link to the online survey (SurveyMonkey) was sent to acquaintances with a description of the target population (i.e., describing inclusion criteria); they were then asked to send the e-mail to their acquaintances, and so on. For six months, the monthly fall assessment was sent by e-mail via SurveyMonkey.

Sample Size. We used a convenience sample consisting of the 134 participants responding to the e-mails sent out with the snowball method. This sample size allowed the estimation of the incidence of falls and univariable association between risk factors and falls with enough statistical precision.

The project was conducted in accordance with the Declaration of Helsinki (1964) and was approved by the relevant ethical committee (CCVEM 014/14). All participants provided informed consent to the participation.

2.4. Adaptation of the Questionnaire

Based on the feedbacks on the understandability and the suggestions for alternative formulations, propositions for amendments were prepared. The final amendments will be part of a future project including a larger sample of experts including elderly persons.

2.5. Statistical Analysis

Descriptive statistics were presented as mean and standard deviation or as proportions, as appropriate. To express the association between risk factors and falls we calculated odds ratios and risk ratios and corresponding 95% confidence intervals. We used Stata Version 14.0 [50]. We calculated both risk and odds ratios because risk ratios are easier to interpret but the odds ratios allow a better comparison with published prediction tools. If a participant did not return a monthly falls follow-up, we assumed that there was no fall in this month.

To test the hypothesis that the participants are not aware of their fall-risk, that is, their self-perceived fall-risk is lower than the actual fall-risk, we calculated the proportion of fallers within each category of the self-perceived risk and calculated a chi-squared test with the null hypothesis that there is no association in the perceived fall-risk with increasing observed fall-risk.

2.6. Prediction Formula

Because our sample size was only large enough for univariable analyses and too low for the fitting of a robust multivariable prediction model, we used the coefficients published in the meta-analyses. The prediction formula consisted of a scoring function and a logistic probability function, where the scoring function reads as follows:scoring function = −4.5 + 0.1044 (age over 60/5) + 1.351 fallen last 12 months + 0.495 low spirit at some days + 0.548 incontinence + 0.62 need get up night + 0.215 rheumatic disease + 0.307 diziness + 0.779 neurological disease + 0.239 diabetes + 0.445 dichotomous pain + 0.247 high blood pressure + 0.47 heart symptoms + 0.875 fear of falls + 0.94 walk slower + 0.742 walking aids + 0.2852 perceived dual task problem + 0.859 self perceived balance + 0.457 any range of motion limitation lower extremity + 0.788 sensory deficit lower extremity + 0.399 vision problem + 0.315 do not hear good + 0.548 dichotomous home hazards + 0.718 low BMI + 0.8242 ADL need help + 0.637 fracture + 0.54 polymedication + 1.445 any medication + 0.24 postural hypotension + 0.98 difficult get up chair because of weak legs.

And the logistic probability function is as follows: .

This formula has to be considered as preliminary because the coefficients of each predictor are not adjusted for all other predictors, which leads to an overestimation of the fall-risk. The coefficients need to be adjusted, for example, by the means of methods proposed by [51]. These methods need larger sample sizes than we had in our study.

Based on this preliminary prediction formula, we tested the calibration of the prediction model with a calibration plot (observed versus predicted falls) and a Hosmer-Lemeshow test. The discrimination (i.e., the ability to detect fallers) was tested with a receiver operating characteristic (ROC) curve and the area under the ROC-curve.

3. Results

The systematic search for systematic reviews and meta-analysis on fall-risk factors yielded 113 abstracts from which 14 systematic reviews were included [4, 5, 40, 42, 43, 5260]. Because we extracted the coefficients from the meta-analysis with the most participants or studies included, the coefficients were taken from the newest reviews [4, 5, 40, 42, 43]. In addition, we extracted the coefficients from one single study for the variable frequent toileting [41] because we preferred this variable over the variables urinary incontinence or urinary functional sign published in the Block 2013 meta-analysis. Table 2 shows the set of extracted factors as well as its odds ratio, coefficients, and heterogeneity, if available.

3.1. Set of Items Devised for the Self-Administered Fall-Risk Questionnaire

Based on the set of predictors we devised a set of questions. Because there was considerable overlap between the predictors, we selected a subset of 29 predictors with the aim of reducing overlap. Because some constructs were covered with more than one question, our questionnaire consisted of 36 questions, including demographic characteristics. Some of the questions consisted of several response options covering different risk factors.

3.2. Characteristics of Included Participants

With the snowball-sampling we could include 134 participants. The response rate during the monthly follow-up varied from 38 to 90% (see Figure 1). The mean age of the 134 participants was 69.3 years with a standard deviation of 5.6 years. There were slightly less women than men (45% women and 55% men). The mean body mass index (BMI) was 25.95; 13% had a BMI of 30 or more (i.e., would be classified as obese). The proportion of participants who did fall during the last twelve months was 18%; only a very small proportion had consequences due to these falls. During the 6-month follow-up, 32 participants did fall at least once, we excluded seven falls (three falls on bike, one fall on ski, two falls on icy roads, and one fall during mountaineering on steep paths), resulting in 25 falls (18.7%). For each risk factor, only a small proportion of participants indicated problems which leads to wide confidence intervals in the odds ratios (Table 2) and the risk ratios presented (Table 3).

3.3. Self-Perceived Fall-Risk and Actual Falls

For the question about the self-perceived probability to fall within the next six months, 49 participants (37%) reported that they “will not fall” and 7 (14% of the 49) did actually fall; 81 (60%) reported that they “will probably not fall” and 17 (21% of the 81) did fall. Only two persons reported that they will “probably fall” and one of those did fall. Two participants did not respond to the question about the self-perceived fall-risk. There was no association between self-perceived and observed fall-risk ().

3.4. Predictive Values

After calculation of the predicted probability to fall based on the values from our questionnaire and the coefficients published in the meta-analysis (Table 2), the prediction model yielded an AUC value for the discrimination of 0.67 (96% CI 0.54 to 0.81) (Figure 2). There was statistically significant miscalibration ( value from the Hosmer-Lemeshow test <0.00001) (Figure 3).

3.5. Understandability of the Questionnaires

Ten participants stated that some questions were unclear and they provided seven specific comments, such as the following: that they were diagnosed with hypertension but had normal blood pressure under medication and did not know what to answer in the questions about present diseases; that some questions were asking about two different pieces of information and that some questions had double negations.

4. Discussion

In this longitudinal cohort study with a six-month follow-up of falls, including 134 community-dwelling elderly participants aged 60 years or more, we tested a preliminary version of an online questionnaire to assess the fall-risk. The main findings were that (a) it is feasible to do an online survey of a comprehensive set of fall-risk factors and (b) the understandability of the questions was good with the exceptions of five questions, (c) the response rate of the monthly falls assessment was too low, (d) the discrimination was moderate, and (e) the calibration was insufficient.

The strength of our study was the approach to devise a set of questions covering the whole spectre of risk factors for falls based on published meta-analyses. This study is an important first step in the development of a comprehensive self-administered questionnaire. Although we cannot present a final version of the questionnaire, this study provides important information for the future development of fall-risk questionnaires.

There are some limitations of our project. The understandability was assessed by semistructured interviews with experts and with an open question in the online questionnaire for the participants. We interviewed only two laypersons before we sent out the questionnaire to the participants. Interviewing of more participants before sending the questionnaire to the participants might have eliminated some problems with the understandability. It is challenging to assess risk factors with self-administered questionnaires. The different visual risk factors especially such as distant contrast sensitivity or depth perception [61] or the dual task problems are difficult to assess. Furthermore, snowball-sampling is a “biased” sampling technique because it is not random and the inclusion of the next participants depends on the previous participants (i.e., participants are not independent). An alternative would have been to search participants by the means of flyers or newspaper or radio advertisements. However, the snowball-sampling has the advantage of being nonexpensive and fast. The nonrandomness is not a large disadvantage in a feasibility study. A further limitation is the low response rate for the monthly fall-risk assessment. We did not systematically send reminders if participants did not respond. Furthermore, we did not present a fall definition to the participant, because we thought that this could confuse more than it would help. Presenting and explaining a fall definition such as the one used by Tinetti et al., “a sudden, unintentional change in position causing an individual to land at a lower level, on an object, the floor, or the ground, other than as a consequence of sudden onset of paralysis, epileptic seizure, or overwhelming external force” [62], might clarify what to report as a fall. Furthermore, the questions about the falls could include examples to illustrate what we understand by a fall. For example, some participants do not consider falling on their knees as a fall, because they were not “lying” on the ground after the fall. It is unclear whether the inclusion of photographs or graphical illustration could improve the reporting. Questions for falls could include examples of specific situations. However, our falls incidence of 19% is compatible with one-year incidences (39% for women, 30% for men), data recently published from Germany [63]; therefore we do not believe that there is an underestimation of the falls. We only assessed falls during six months; a longer follow-up would have increased the number of falls. The frequency of problems reported in the individual fall-risk questions was very low if compared to other studies on self-report fall-risk questionnaires [15, 17]. This could be due to the good health state of our participants but it could also be due to how the questions were formulated (i.e., unclear wording or wording targets only serious problems). Given the very low proportion of present risk factors, the selection of our sample could be problematic. There might be a selection bias towards a higher socioeconomic state, given the high proportion of participants with higher education. Given the low presence of risk factors we would have expected a lower falls incidence rate. Our prediction formula still overestimates the fall-risk. This is most probably due to the high correlation between the included predictors. However, our sample size was sufficiently large for univariable analyses but too low to adjust for this correlation by the means of a multivariable model. Therefore, the prediction formula needs to be adjusted with methods proposed by Steyerberg and colleagues [51] in a larger sample once the questionnaire is in its definitive version and after testing of the reliability.

If we compare our results to published studies using questionnaires for the assessment of fall-risk, we have similar values for calibration and discrimination compared to Cattelani et al. [16]. Compared to El Miedany et al. [15] we have lower predictive values; they received an AUC value of 0.89 with only five predictors. However, they included a sample where all had at least one previous fall and where 82% reported to walk slower, 65% reported loss of balance, and 55% had poor sight. Therefore, the two samples are not similar. Our AUC value is low but one has to consider that other tests widely used to predict falls, such as the timed-up-and-go (TUG) test, do not have better predictive values. A recent review on the predictive values for falls of the TUG in community-dwelling elderly people found an AUC value of 0.57 [64].

We did not find an association between self-perceived fall-risk and falls. One might expect that the self-perceived risk for falling increases fear of falls, which is known to be associated with future falls. One reason why we did not find an association is that the response options of the question for the self-perceived risk were not optimal and should be improved for future studies.

Our study has some implications for further research. The following amendments need to be done before further testing: (1) the question about past falls which should ask about the number of falls in the last year; it is recommended that persons with more than one fall in the past year should be referred to a detailed assessment [65]; (2) rewording of some questions; and (3) explication of what is considered as a fall to exclude falls, for example, due to an overwhelming external force, that is, following the falls definition used by [62]. After a refinement of the questions, test-retest reliability must be tested before the coefficients for a final predictive model should be assessed with a multivariable logistic regression based on results from a larger cohort study with a one-year follow-up in which the analyses should be separated for the prediction of one fall or recurrent falls. Furthermore, a larger sample size would allow evaluating whether some questions might be eliminated without losing discrimination or calibration of the prediction tool.

Implications for practice are as follows. Our study showed that in a sample with a relative low risk profile the incidence of falls was 19% during a period of six months and that the participants were mostly not aware of their fall-risk. Health professionals who see patients for other indications, for example, for the treatment of osteoarthritis, back pain, or neurological problems, could use this fall-risk questionnaire as a screening tool or a “flag system” and specifically test the domains where the patients report problems. The health professionals could then refer the patient to a falls-prevention group. The tool could also be used for the preparation of a visit to a medical doctor. The patients could bring the questionnaire to the medical doctor to discuss the results and possible strategies if necessary.

5. Conclusion

This study showed that fall-risk awareness is low and that even in a sample of elderly people with a low risk profile in known risk factors the falls incidence is 19% in a six-month period. The present questionnaire needs some adaptation of the wording and reliability testing before a definitive prediction formula can be developed in a large sample and with multivariable analyses. Measures need to be implemented to increase the monthly response rate for the follow-up period.

Competing Interests

The authors declare that they have no competing interests.

Acknowledgments

The authors would like to thank the experts and the participants of this study. Furthermore, they really appreciated the helpful comments from the anonymous reviewers.