Data were collected in 2006. The primary outcome of interest was the number of falls in the six months after the initial mobility assessment. The definition of a fall used was ‘a person unintentionally coming to rest on the ground’ (Jensen et al 2002, Vu et al 2006). Participant medical notes and incident reports were audited find more at two-monthly intervals by the research physiotherapist for entries relating to falls. The putative predictors assessed were the individual items and total score of the Physical Mobility Scale (Nitz et al 2006).

The Physical Mobility Scale includes nine mobility tasks ranging from bed mobility to ambulation, which are scored on a six-point scale from full dependence (0) to highest independence (5). Item scores are summed to give a total score (0–45) representing overall mobility, with lower scores indicating greater mobility impairment. Physical Mobility Scale assessments were carried out by physiotherapists who were independent of the staff employed by the residential aged care facilities. Physical Mobility Scale assessments were completed at three time check details points: baseline, and at two and four months after the baseline assessment. Thus, multiple Physical Mobility Scale assessments and fall data were included for each resident. The association between Physical

Mobility Scale total score and item scores, and risk of falling was assessed using Prentice, Williams, and Peterson conditional risk set survival models for recurrent events (Prentice et al 1981). An advantage of these models over traditional survival models is that they can be applied to data that include multiple observations for each participant, eg, multiple risk factor assessments and multiple outcome events. The recurrent event models used in this analysis were based on data that included up to three Physical Mobility Scale score observations for each resident corresponding to the baseline, two, and four month assessments and additional observations for each fall event that occurred. Total scores were coded into a priori specified

score categories to allow non-linear associations to be explored. Five score categories were selected to ensure an adequate number of observations these in each category. Too few observations in categories can lead to predictive models that are unstable and may provide imprecise and inaccurate associations. Each Physical Mobility Scale total score category was entered in a univariable model to establish the risk, reported as a hazard ratio, of sustaining a fall for each Physical Mobility Scale total score category. The ability of the Physical Mobility Scale items and total score categories to discriminate fallers from non-fallers was also explored through Prentice, Williams, and Peterson conditional risk set survival models for recurrent events (Prentice et al 1981).