1. INTRODUCTIONFall is a major source of morbidity and mortality in older people. Preventing falls is a key health priority. Fall is the second leading cause of accidental or unintentional injury deaths worldwide [1]. Injuries from falls contribute to increased disability and mobility limitations for older people, and therefore, as the proportion of the aged population increases, the research of the risk factors for falls is becoming increasingly important. The cost of hospitalizations due to fall related injury for people aged 65 and over is projected to increase to US $240 billion by the year 2040 [2]. Due to the complexity of risk factors for falls, it is crucial to refine conceptual and methodological frameworks for understanding and predicting falls in the population to aid the formulation of more effective fall interventions. One such framework which may be applied to understanding and predicting risks of falls is the International Classification of Functioning, Disability and Health (ICF) [3].
The ICF is a classification system that documents the complex interactions of a person with his physical, social and psychological environments, and how this interaction affects his health status [3-5]. Together with the ICD10 [6], the World Health Organisation’s diagnostic epidemiological, clinical and health management classification system, the ICF can describe health for an individual or a population and identify relationships between factors that are known to contribute to falls, or assist in detecting those most at risk [7]. The level of physical functioning experienced by an individual is an important factor in falls. The ICF’s emphasis on personal functioning—on how an individual engages in activities and participates in society in the context of environmental and personal factors [3], and the dynamic interaction between health states, the person and his/her environment [3], makes it particularly suitable for extending understandings of falls and falls risk. As a classification framework for fall prediction and intervention, the ICF can conceivably provide an inter-professional scientific basis for understanding and studying fall behaviour and risk, and ensure that important concepts are identified and measured.
The ICF allows a comprehensive study of fall risk that includes medical, social and psychological risk factors. As fall risk is multidimensional, the ICF is appropriate to systematically investigate falls in the clinical practice [8]. Further, the ICF has already been embraced across a number of falls-related disciplines, such as occupational therapy and rehabilitation [7,9], but has not been specifically related to fall risk. The interactions between the ICF components are reciprocal. Therefore, the ICF accommodates the dynamic interactions underpinning fall risk, so that any change in one of the ICF components can influence other interactions within the framework. In a study of the ICF and clinical assessment measures in relation to falls following stroke [7], the ICF was found to match the multidimensional nature of fall risk. However, this relationship was not empirically tested. Furthermore, little evidence exists on how to reliably map the ICF to fall risk, or whether an analysis of fall risk using the ICF components is able to confirm the theoretical assumptions underpinning the ICF [10].
The Australian Longitudinal Study on Women’s Health (ALSWH) is well positioned to explore the prevalence of falls and serious injury among a large sample of older women with varying degrees of health, wellbeing, mobility and functional capacities or limitations and disability within the Australian community, and to observe and study the antecedents and outcomes of falls over the course of the study [11]. An exploratory cross-sectional study of the baseline older cohort (n = 12,000) of the ALSWH demonstrated that serious falls were significantly associated with the physical component score of the SF36 (Short-form-36 Health Survey) [12], taking drugs for nerves, having had a serious life event other than a fall in the previous year, and feeling dejected [11]. These results suggested that psychological, social, and occupational factors consistent with the ICF were important to understand serious falls in older women. However, previous studies have argued that the ICF is limited by its failure to meaningfully address the influence of personal factors such as socioeconomic status and gender which are critical to understanding the lives of individuals [13], and by inherent difficulties such as those posed by a lack of differentiation between conceptual boundaries, for example, the boundaries between the concepts of Participation and Activity [4,14]. Further, it is argued that it is necessary to clarify these boundaries for the ICF to be scientifically useful for the empirical research [4]. We sought to map risk factors for falls (measured in the ALSWH) to the ICF, and to determine whether this conceptual model predicted falls statistically significantly.
2. METHOD2.1. Data CollectionParticipants in this study were drawn from the Australian Longitudinal Study on Women’s Health (ALSWH), a population-based study of changes in health and wellbeing of three different age-cohorts of women living in Australia. The sample was randomly selected from the Medicare Australia database, which is the universal provider of health insurance in Australia [15], and surveys have been conducted at 3-yearly intervals since 1996. The study was approved by the University of Newcastle Human Research Ethics Committee. Further details on the ALSWH are available from the website, www.alswh.org.au. For this study, 650 women were randomly selected from ALSWH participants who were born in 1921-1926, and who had completed Survey 3 in 2002.
Women in this sub-sample were invited to complete a postal survey collecting additional falls information including the Modified Falls Efficacy Scale [16], Fear of Falling scale [17,18], the Home Falls and Accidents Screening Tool (HOME FAST-SR) for measuring home hazards [19], and use of walking aids, and activities of daily living (Lambeth Disability Scale) [20]. This survey supplemented the data collected within the main ALSWH surveys known to be significantly related to falls. The outcome of falls was assessed in the sub-study and again in Survey 4 (2005) of the main study from the question: In the last 12 months have you had a fall to the ground (Yes, No). This self-report item has been shown to have reasonable sensitivity and specificity when compared with falls calendars [21].
2.2. Mapping to the ICFItems included in the four main surveys and sub-study were mapped to the corresponding ICF components using ICF definitions (but not the levels of classification) [3] and the linking rules developed by Cieza [22,23]. These linking rules have been used in many studies [24-27] and were the only available tool to use in assigning the ALSWH survey items to the ICF components. Initial mapping was done by two of the authors who were occupational therapists (AM, LM) according to their interpretations of the meaning of the ICF and the linking rules. Demographic information, use of health services, lifestyle factors and living arrangements were mapped to personal factors, despite these being unclassified within the ICF [3,28].
2.3. Longitudinal AnalysisUnivariate analyses (chi-square and t-tests) were applied to assess associations between the variables aligned to the ICF and self-reported falls at Survey 4. Explanatory variables were selected as those items measured on the sub-study or on the most recent survey prior to the sub-study. Most variables were measured at Survey 3 but some variables were measured at Survey 2 and education and country of birth were measured at Survey 1 (See Table 1). Correlation matrices [29] were constructed to identify highly correlated explanatory variables. Where variables were highly correlated, the variable most strongly associated with falls at the univariate level was retained for inclusion in the multivariate modelling.
Logistic regression was used to construct sub-models of factors associated with falls at Survey 4 for each ICF component. Variables with P ≤ 0.3 at the univariate level were included in multivariate analyses for each submodel, with backward stepwise removal of variables [30]. A final composite model was constructed by systematically combining each sub-model into a forward stepwise regression model-building process in order of the strongest Akaike Information Criteria (AIC) [31] for each submodel. The composite model examined statistically adjusted effects of the sub-models on the log odds of falls. A total of eight steps were conducted to create the final composite model from the sub-models. Statistical interactions between the strongest variables from each submodel were also explored. All these analyses were performed by Stata v8.2 [32] (StataCorp, 2004) and SAS v9.1 [33]. (SAS Institute, 2007).