A prospective, longitudinal observational study was carried out in which a cohort trial-nested case–control design was used to develop predictive models for isolated falls (one or two falls in the course of one year of follow-up) and also recurrent falls (three or more falls in the course of one follow-up year) in non-institutionalized older people with functional independence. To be included in the study, subjects were required to be aged 70 and over, with a Barthel Index score of 60 points and above. Additionally, subjects were required to be independent for walking (with possible technical aids, but not assisted by another person) and reside in the region of la Ribera – a rural area with several cities and villages totaling about 250,000 inhabitants in Valencia, Spain.

Subjects were excluded from the study in cases of associated disease conditions resulting in a life expectancy of under 6 months, blindness and deafness, serious psychiatric problems diagnosed by a psychiatrist (schizophrenia, major depression, bipolar disorder, panic disorder) or moderate to severe cognitive impairment (diagnosed by the general practitioner or geriatrician). Individuals failing to sign the informed consent document for participation in the study were also excluded.

The minimum sample size was determined via the Schlesselman method for a 35% incidence of older adults with falls and a minimum estimated relative risk of 2, with a statistical power of 80% and an alpha probability of less than 95% error. The resulting minimum sample size was 270 subjects. To compensate for potential dropouts, an oversample of 20% was adopted, bringing the final minimum sample to 324 subjects for follow up.

A publicity strategy was designed to facilitate recruitment from December 2013 to May 2014, which consisted of posters and notifications in retired people centers, meetings with the managing authorities of retired people associations and centers, and telephone calls to people inscribed in different primary care centers. Family telephone calls were also used in an attempt to avoid future dropouts in cases where individuals lacked the physical means to attend health education sessions on fall risk-free environments on their own.

The study was approved by the Clinical Research Ethics Committee of Hospital Universitario de la Ribera (Valencia, Spain), and each participant signed an informed consent document before inclusion in the study or statistical processing of the data obtained.

Geriatric assessment was carried out by a nurse in each primary care center, with collection on the baseline visit of sociodemographic variables, anthropometric data (body weight, height, and dominant grip strength measured with a dynamometer), functional variables (Barthel index, Lawton scale, Tinetti test), cognitive information (MEC-Lobo, a Spanish adaptation of the Mini-Mental State Examination (MMSE) – moderate to severe cognitive impairment being considered as MEC-Lobo <24 points), nutritional data (short Mini-Nutritional Assessment (sMNA)) scale, body mass index (BMI) (obesity being defined as BMI≥30kg/m2) and clinical parameters (comorbidities, 4 or more disease conditions). Information was also noted on the number of previous fractures, presence of ulcers, foot hygiene (correct nail cutting [short and without thickening] and correct skin hydration; if any of these were missing, foot care was considered inadequate), adequacy of footwear (adapted to foot morphology, closed, heel closed and correct heel height: no more than two centimeters; if any of these were missing, footwear was deemed inadequate), fear of falls (modified Falls Efficacy Scale, mFES), number of falls in the 12 months before inclusion in the study (a fall was considered as an unexpected event in which the person comes to rest on the ground, floor, or lower level),16 number of risk factors,17 extrinsic risks (related to the environment: lack of stair handrails, poor stair design, lack of bathroom grab bars, dim lighting or glare, obstacles and tripping hazards, slippery or uneven surfaces, improper use of assistive devices), and intrinsic risks (related to the individual: advanced age≥80 years, previous falls, muscle weakness, gait and balance problems, poor vision, postural hypotension, chronic conditions including osteoarthritis, diabetes, stroke, Parkinson's disease, incontinence, dementia, fear of falling). Visual impairment was assessed considering whether the individual could indicate that he or she had difficulty seeing (while wearing glasses if appropriate). Hearing impairment was assessed considering whether the individual could indicate that he or she had difficulty hearing. Physical activity was divided into three categories: <3h/week, 3–6h/week, >6h week.

Lastly, the use of drugs with actions upon the central nervous system (benzodiazepines, antidepressants, anticholinergic agents and medication for vertigo) was documented.

A monthly telephone call was made by the nurse who had carried out the geriatric assessment and was known to the elderly participant, in order to determine the number of falls over the last 30 days.

The study data were entered into MS Excel spreadsheets, followed by statistical analysis using the SPSS® version 19.0 statistical package (SPSS, IL, USA).

The variables were reported as proportions and/or means and standard deviations. Parametric tests (Student t-test and Fisher-Snedecor ANOVA test) were used for the comparison of means, while nonparametric tests (chi-squared test and linear trend test) were used for the comparison of proportions. The association between falls and the different risk factors, based on the odds ratio (OR) were then analyzed. A binary logistic regression model was developed to explore the importance of the risk factors, defined according to the falls observed during the 12-month follow-up period of the study. Firstly, the complete model, with all the variables in the bivariate analysis, was found to be significantly associated to the presence of falls during follow-up. In a second step, all those variables failing to produce a significant change (defined as the absence of an adjusted effect of >10%) were eliminated from the model, including those which did not result in an improved standard error when omitted. Consensus was sought among the investigators in those cases where two or more subsets of variables with the same degree of fit were obtained.

The following variables were entered into the model: age groups, sex, cohabitation, hearing problems, vision problems, inadequate foot hygiene, fear of falls, previous fractures, anxiety-depressive syndrome, use of alpha-blockers, use of benzodiazepines and beta-blockers, use of drugs with actions upon the central nervous system, multiple drug use, comorbidity, orthostatic hypotension, obesity, and physical activity.

Two binary logistic regression models, adjusted by age and gender, were generated. We used two predictive models in order to distinguish between the predictors of falls. In the first model, a binary variable (1 or 0) was used, as fallers were compared with non-fallers, whereas the variable in the second model compared recurrent fallers (3 or more falls) with non-fallers and isolated fallers (≤2 falls) over the 12-month follow-up period. Our study varied the definition of recurrent falling by varying the time unit between falls and by varying the number of falls per time unit. We collected the number of falls in the previous 12 months and the definition (at least two falls within 12 months) showed the lower association between the predictors for falls in a sample with high functionality. Previous studies by our research group,18 and other studies found in the literature,19 support a greater predictive value of fall risk profiles using the outcome variable “at least three falls (>2 falls) within 12 months”. This definition was therefore adapted for this study, with falls being stratified into three categories: non-fallers, isolated fallers (1 or 2 falls) and recurrent fallers (>2 falls).

The incidence of falls was 39.2% (95% CI: 34.1–44), and 24.1% of the subjects had suffered from falls (95% CI: 19.7–28.4). Table 3 shows the results of the analysis of the risk factors underlying isolated and recurrent falls. In the case of isolated falls, the female sex was predictive of a two-fold increase in the risk of new falls over the following 12 months among those individuals who had suffered falls the previous year.

The risk factors significantly associated to multiple falls in older adults were the female sex, living alone or with a partner, previous fractures, hypotension, fear of falls and hearing problems (Table 3).

Our first predictive model of isolated falls in older adults showed a sensitivity of 10%, a specificity of 98.2%, a positive predictive value of 75.2% and a negative predictive value of 77.5%. The model explained 47% (Nagelkerke R2) of the variance of isolated falls and correctly classified 77% of the cases. The model allowed us to predict that older adults who have suffered a fall in the last year and who are using alpha-blockers have a 24% probability of suffering another fall. The number of falls in the previous 12 months (OR=1.3; 95%CI: 1.11–1.53; p<0.001) and alpha-blockers (OR=6.72; 95%CI: 1.62–27.79; p=0.009) were predictors of falls (Table 4).

The second predictive model, corresponding to recurrent falls in older adults, showed a sensitivity of 15.2%, a specificity of 98.8%, a positive predictive value of 58.8% and a negative predictive value of 92.3%. The model explained 39.4% (Nagelkerke R2) of the variance of recurrent falls and correctly classified 91.4% of the cases. The model allowed us to predict that older adults with recurrent falls, in the presence of previous fractures (OR=9.55; 95%CI: 4.1–22.25; p<0.001), a body mass index of ≥30kg/m2 (OR=1.09; 95%CI: 1.01–1.19; p=0.035), and who are using benzodiazepines and beta-blockers (OR=2.77; 95%CI: 1.53–5.02; p<0.001), have a 99.98% probability of suffering further falls (Table 4).

Our study has a number of limitations. Firstly, the data relating to drugs associated with falls in older adults corresponded to those pharmacotherapeutic groups most commonly associated to falls, and which had been prescribed by the physician (not taken sporadically and/or upon personal decision by the patient). Secondly, the doses of the drugs with effects upon the central nervous system were not documented. Thirdly, the time, weather conditions and place of the fall were not recorded, and the follow-up period was limited to 12 months. Furthermore, monthly reporting of falls by the older adults was affected by bias resulting from a possible lack of information. Finally, adherence to drug treatment was not taken into account, and there may have been bias in the information collected from the case histories and from the patients. These circumstances may produce measurement bias in the variables that influence the risk of falls.