A long-recommended but seldom-used method of analysis for fall injuries found a unique pattern of risk factors in the youngest-old

Background Few studies on fall risk factors use long-recommended methods for analysis of recurrent events. Previous falls are the biggest risk factor for future falls, but few fall studies focus on the youngest-old. Aims This study’s objective was to apply Cox regression for recurrent events to identify factors associated with injurious falls in the youngest-old. Methods Participants were community-dwelling residents of southern Sweden (n = 1,133), aged 59–67 at baseline (median 61.2), from the youngest cohorts of the larger Good Aging in Skåne (GÅS) study. Exposure variable data were collected from baseline study visits and medical records. Injurious falls, defined as emergency, inpatient, or specialist visits associated with ICD-10 fall codes during the follow-up period (2001–2011), were gathered from national and regional registries. Analysis was conducted using time to event Cox Regression for recurrent events. Results A majority (77.1 %) of injurious falls caused serious injuries such as fractures and open wounds. Exposure to nervous system medications [hazard ratio (HR) 1.40, 95 % confidence interval (CI) 1.03–1.89], central nervous system disease (HR 1.79, CI 1.18–2.70), and previous injurious fall(s) (HR 2.00, CI 1.50–2.68) were associated with increased hazard of injurious fall. Conclusions Regression for recurrent events is feasible with typical falls’ study data. The association of certain exposures with increased hazard of injurious falls begins earlier than previously studied. Different patterns of risk factors by age can provide insight into the progression of frailty. Tailored fall prevention screening and intervention may be of value in populations younger than those traditionally screened.

Many modifiable and non-modifiable fall risk factors have been identified in older populations, including environmental, physiological, and pharmacological factors [17,18]. One factor consistently found to be associated with falls is a history of previous falls, meaning that delaying or preventing early falls may reduce the number of future falls. There is, however, little research focused on falls in the youngest-old, with most studies of fall risk factors focusing on populations with a mean age greater than seventy [17].
Furthermore, to date, most studies on fall risk factors do not use statistical methods that take into account recurrent events, despite longstanding recommendations in the epidemiological literature against simply dividing populations into ''fallers'' and ''non-fallers'' [19][20][21]. Studying total fall burden is of great public health interest, as each fall is associated with costs to the individual and the healthcare system.
The primary objective of this study was to use Cox regression for recurrent events to determine which potential fall risk factors were associated with injurious falls in a relatively young population of individuals.

Design and participants
Using a longitudinal cohort design, we studied the association between injurious falls and selected exposure variables including symptoms, diagnoses, physical performance, and medications in the youngest-old. Participants in the current study were those from the 60-and 66-year-old cohorts (ages 59-61 and 65-67, respectively) from the larger cohort study Good Aging in Skåne (GÅ S). GÅ S is a longitudinal cohort study of 2,931 individuals 60-93 years of age from five rural and urban municipalities in southern Sweden who were randomly selected on the basis of the National population register. Skåne, which is Sweden's southernmost county, had 1,136,571 inhabitants in 2001, making up 12.8 % of Sweden's total population of 8,909,128 inhabitants that year [22].
It was decided for reasons of power, generalizability, and confounding effects to focus solely on the youngest two cohorts in GÅ S. These cohorts had higher participation rates and cohort sizes more than twice as large as those of older age groups. Furthermore, focusing on the two youngest cohorts limited birth year effects.
Study visits for the population in this study began January 8, 2001 and ended July 30, 2004. Visits took place at one of the study clinics or at the place of residence if the participant was unable to come to a study clinic. Study size was determined by the size of recruited cohorts in the greater GÅ S project. Details of GÅ S have been described previously [23].

Participant flow
During the recruitment period, there were 108,669 residents of Skåne in the target age group of youngest-old. Of the 2,233 initially invited to participate in the study, 1,382 eligible individuals completed the first study visit. The participation rates for men and women were 65.6 and 65.5 %, respectively. The participation rates for the 60-and 66-year-old groups were 62.8 and 68.7 %, respectively. Previous falls were determined in part from entries in a regional database that began in 2001, requiring the removal of 210 individuals with a baseline study visit in 2001 to be able to collect complete data on previous year fall history for the final analysis. Thirty-nine individuals were missing data for one or more covariates and were excluded from the final analysis. This left 1,133 individuals in the final analysis, 83 of whom died during the study follow-up period.

Measurements
Injurious falls were defined as date of first seeking emergency, inpatient, or specialist care for a medical event associated with an ICD 10 fall-related trauma code (W0-19 or V0-19; with 275 (94 %) W codes and 18 (6 %) V codes). Fall events were sorted by date and ICD code and only the first visit for each unique event was used; patients with more than one unique event had multiple event dates. Data on falls and deaths during the study period were taken from the National Diagnosis Register and Skåne Region's Healthcare Cost Database. Because outcome status was determined using each individual's unique identifying number in these registries, loss to follow-up is minimal. The only injurious falls not captured by these databases are those that occurred abroad or those occurring in another region and not resulting in inpatient care.
Exposure variables collected from baseline study examinations and participant surveys were chosen based on their plausible physiological mechanism for increasing fall risk. To limit multiple testing problems, the chosen variables were then grouped into covariates based on similar etiology or potential sequelae, forming 16 covariates composed of 57 sub-variables. Covariate groupings and measurement techniques are specified in Table 1. Baseline exposure to covariates is detailed in Table 2. Two additional covariates (previous falls and sex) were added to the final analysis.

Statistical methods
We conducted a preliminary analysis (n individuals = 1,382) using v 2 tests with a dichotomous fall outcome to determine which of the sixteen potentially influential covariates to include in the final Cox Regression. This was done to reduce multiple testing problems. The preliminary analysis was carried out using IBM SPSS Statistics for Windows Version 20 (IBM Corp, Armonk, NY, USA).
The final analysis calculated hazard ratios using time to event Cox regression (Andersen-Gill model) [24][25][26] with participant age as the time scale. Note that using age as the time scale, we view the data as left-as well as right-censored. Formal tests were based on the robust variance estimators defined in Lin and Wei [27]. The six covariates that were found to be statistically significant in the preliminary analysis were included in the final analysis. Sex was added as a covariate. Potential interrelatedness of fall events was taken into account by adding a dichotomous variable for previous falls (during or in the year prior to study participation), for a total of eight variables.
All falls from the year prior to study entry until Dec 31, 2011 were included in the regression; falls in the year prior to an individual's study start date were only taken into account for the previous falls' variable. All individuals contributed time at risk from their study entry date until Dec 31, 2011, or until date of death. This analysis was conducted using SAS software version 9.2 (SAS Institute Inc., Cary, NC, USA).

Statement of human rights
Consent was obtained to use linked data from hospital registries. The study was approved by the regional ethics committee at Lund University and was therefore performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. All subjects provided written consent to participate.

Descriptive data
The participants in the final analysis had a median age of 61.

Main results
Cox proportional hazards regression of the final covariates yielded statistically significant hazard ratios for the covariates of nervous system medications, central nervous system disease, and previous injurious fall. The results of this analysis are detailed in Table 3.

Discussion
We found that taking nervous system medications, the presence of central nervous system disease, and occurrence of previous injurious fall were associated with increased hazard of injurious fall in a relatively young population. A major strength of this study is the use of statistical analysis for recurrent events, allowing us to include all injurious falls in the analysis rather than only first falls. Few studies of fall risk factors have used this or similar statistical approaches, despite the importance from a public health perspective of analyzing total falls [19,20]. This method of analysis was straightforward to use with data typically collected during fall studies and the help of widely available statistical software. Using age as time scale in the time to event Cox regression allowed us to take into account age effects over the relatively long follow-up period. Using health records rather than participant recall to determine the status of the outcome variable also eliminated the documented issue of recall inaccuracy [28][29][30], and allowed us to focus solely on injurious falls. Individuals experiencing fall injuries were effectively included more than once in the modified Cox regression, albeit with age and previous fall injury status accounted for at the time of each fall. This means that risk profiles for these individuals were more heavily weighted in the analysis, a desired effect since these were associated with a greater number of injurious falls. This is in keeping with the public health goal of studying the overall burden of fall injury.
Although healthcare usage can differ between countries, the pattern of injury severity resulting in seeking medical attention in this group indicates that the outcome is generalizable, with 226 of 293 events (77.1 %) associated with serious injury. A further 61 events (20.8 %) were associated with some other form of injury. The pattern of injuries found in this study is similar to that described in other studies of injurious falls [11,12] and the pattern of fracture injuries is also consistent with that described in slightly older populations [31,32].
A limitation of this study is the relatively low event rate in this population compared to older populations [33], which limits the power of our analysis. Additionally, a trade-off of grouping sub-variables to avoid multiple testing issues is the potential for a decreased ability to discover associations of injurious falls with individual sub-variables. Further large studies are needed to corroborate the associations we have found and potentially discover further associations. Our primary outcome focused only on falls resulting in emergency, inpatient, or specialist care. Further studies could include non-injurious falls, since these may also be associated with future injury-causing falls.
Of the covariates considered, some have been associated with injurious falls in older populations but were not found to be associated with injurious falls in our population; for example, balance impairment, female sex, and respiratory disorders [7,34]. This lack of association in our population despite the large proportions exposed to some of these covariates indicates that younger individuals may not be as susceptible to injuries associated with these risk factors. The concept of frailty, defined as increased ''vulnerability to adverse outcomes'' [35], is a possible explanation for this finding, which is in keeping with previous studies that have found that older individuals are more prone to injury after falling than younger individuals [9,36].

Conclusion
The results of the present study suggest that in the youngest elderly, taking nervous system medications, having central nervous system disease such as stroke or Parkinson disease, and having experienced injurious falls previously are associated with increased hazard of experiencing a subsequent injurious fall. These risk factors have previously been found to be associated with falls in general in older populations [17,18,[37][38][39]. This study suggests that the association of these factors with injurious falls may begin earlier than previously studied. Some risk factors previously identified in older populations were not associated with injurious falls in this study, a finding potentially explained by processes of frailty. This study also demonstrates the feasibility of using Cox regression for recurrent events with typical falls' study data and widely available statistical software. Future studies may consider the benefit of fall prevention screening and intervention in this younger population.
Acknowledgments This work was supported by the Swedish Ministry of Health and Social Affairs; the county Region Skåne; the Medical Faculty at Lund University; and the Swedish Research Council (Grant number 2013-8,604). The funding bodies had no role in study design; collection, analysis, or interpretation of data; in the writing of the manuscript; or in the decision to submit the manuscript for publication. Conflict of interest On behalf of all authors, the corresponding author states that there is no conflict of interest.
Ethical standard Consent was obtained to use linked data from hospital registries. The study was approved by the regional ethics committee at Lund University and was therefore performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. All subjects provided written consent to participate.
Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.