E-health StandingTall balance exercise for fall prevention in older people: results of a two year randomised controlled trial

Study design

We conducted a prospective, assessor blinded, two arm, parallel randomised controlled trial with two year follow-up in Sydney, Australia. The trial was approved by the University of New South Wales ethics committee in December 2014 (HC#14/266) and was registered prospectively in the Australian and New Zealand Clinical Trials Registry (ACTRN12615000138583) on 13 February 2015. The study protocol was published in 2015.8 The statistical analysis plan was preregistered in October 2018 through the OpenScience framework (https://osf.io/42gje/) before completion of data collection in November 2019. We used the CONSORT (consolidated standards of reporting trials) statement, ICMJE recommendations, and TiDieR (template for intervention description and replication) checklist when preparing this article.

Participants

We recruited community dwelling older people in the Sydney metropolitan area by using flyers, printed advertisements in local newspapers, presentations at residential and community senior centres, and word of mouth. Study participants lived on average 12 km (range 1.2-46.9 km) from Neuroscience Research Australia (Randwick, NSW). After initial screening by telephone, eligible people were invited to participate if they were aged 70 years or older, living in the community, independent in activities of daily living, able to walk household distances without the use of a walking aid, and willing and able to give informed consent and comply with the study protocol. People were excluded if they had an unstable or acute medical condition that precluded exercise participation, suffered from a progressive neurological condition (such as Parkinson’s disease or multiple sclerosis), were cognitively impaired as defined by a Pfeiffer short portable mental status questionnaire score less than 8,9 or were currently participating in a fall prevention programme. Eligibility was determined after informed verbal consent. People who were eligible and agreed to participate in the study were asked to provide informed written consent.

Randomisation and masking

Participants were randomised after completion of the baseline assessment. Permuted block randomisation with mixed block lengths of four and six was applied to form two groups of similar size (allocation ratio 1:1). People living in the same household were treated as one unit to avoid contamination. Allocation was performed centrally using a custom randomisation programme by an investigator not involved in participant assessments or delivery of the intervention. Allocation concealment was ensured because the randomisation code was only released after the baseline assessment was completed. Only the first 226 participants were invited for repeated physical tests to reduce costs and participants’ time. Outcome assessors were blinded to study group assignment throughout the trial. Statistical analyses were performed blinded for intervention or control group allocation.

Procedures

All participants received a tablet computer with a health promotion education programme that focused on health related information relevant to older people, in addition to usual care, for two years. This health promotion education programme comprised weekly fact sheets (104 in total) with information on healthy diet, drugs, fall risk factors, and exercise. Tablet based health education alone was chosen as the active control intervention to regulate the use of technology and allow data collection (such as number of falls during the trial period) through a tablet computer for both groups. Participants received a manual on how to use the tablet computer. After their baseline assessment, participants received guidance on the basic features of the tablet computer and health promotion education programme.

The intervention group received the StandingTall programme, with exercise equipment (foam cushion, stepping box, exercise mat), in addition to the health promotion education programme and usual care that was received by the control group. The StandingTall intervention consisted of balance exercises delivered through a tablet computer in the participants’ homes with embedded behavioural change techniques, including a weekly calendar for scheduling exercises, goal setting, and educational fact sheets. The exercises focus on standing balance, targeted stepping, and step-up (box) exercises. More information about the programme can be found in the study protocol8 and online (https://www.standingtall.org.au/). Participants were asked to exercise for at least two hours per week for the duration of the trial, in line with the international recommendations for fall prevention at the time of the study.2 The intervention was introduced gradually; participants started with 40 min/week of exercise, which was increased by 20 min fortnightly until participants reached the required amount of two hours per week in week 9.

StandingTall delivers individually tailored balance exercises that increase in difficulty over time; the programme also allows people to choose the time and duration of their exercise sessions. The intensity of the balance exercises is monitored by using a self-report modified rate of perceived exertion scale and is adjusted as performance changes throughout the trial without the need for supervision. Exercise adherence (volume and frequency) was monitored for two years after automatic data transfer to a server. During the first six months, participants were encouraged to inform the research team when they were going away or would not be able to exercise for a few weeks. Participants who did not inform the team and did not reach 100 min/week for two consecutive weeks were contacted by telephone so that the reason for non-adherence could be recorded, any issues related to the programme could be discussed, and the team could encourage adherence. These calls stopped after six months to gain a better understanding of behavioural change and long term exercise adherence.

Intervention group participants received two home visits. During the first visit, a qualified exercise physiologist instructed the participant on how to use the StandingTall programme; this visit occurred between one and three weeks after the baseline assessment and lasted approximately one hour. The second home visit after one month lasted approximately 30 min and ensured safe use of the programme and progression of training. Control group participants received two phone calls by qualified exercise physiologists at the same time points to discuss any issues with accessing the health education programme and using additional features of the tablet computer.

Outcomes

The primary outcome measures were the rate of falls and the proportion of people who fell over the first 12 months of the trial. A fall was defined as ‘an unexpected event in which the participant comes to rest on the ground, floor or lower level’.7 Falls were monitored by using prospective weekly fall diaries through the tablet computer (completed from baseline assessment for 24 months). Fall information was automatically uploaded to a database. Research staff contacted participants by telephone at the end of each month when their fall diaries were incomplete to record missing data. The falls database was checked, reviewed, and locked before group allocation was unmasked. Falls that occurred up to one year after randomisation were included in the primary analysis. Falls that occurred up to two years after randomisation were included as secondary fall outcomes. Injurious falls were defined as falls that resulted in any injury (eg, bruises, cuts or grazes, joint dislocations, sprains or strains, fractures, pain), or falls that required medical care (eg, visit to physician or emergency department).

Secondary outcome measures were assessed at baseline, at six months to examine acute effects, and at 12, 18, and 24 months to examine retention effects. These measures included common fall risk factors: laboratory based balance and neuropsychological assessments (at baseline and at six and 12 months after baseline assessment in the first 226 participants), and remote measures (taken at home) of wellbeing, quality of life, and activity levels (at baseline and at six, 12, 18, and 24 months after baseline assessment in all participants). Physiological fall risk was assessed using the physiological profile assessment.10 Balance, functional mobility, and gait were evaluated by using tests of standing balance (standing with feet in different positions for a maximum of 30 s per condition: feet together, near tandem, and tandem on floor and foam cushion, and left and right foot on floor; sum of durations for all eight conditions), maximum forward-backwards and controlled leaning balance,1011 timed sit-to-stand12 and up-and-go tests,13 short physical performance battery,14 and self-selected walking speed over 10 m.15 Stepping performance was assessed with choice, Stroop and inhibitory stepping reaction time tests.1617 Cognitive function was measured with the Montreal cognitive assessment18 for global cognition, trail making tests19 for set shifting, and the Victoria Stroop task20 for response inhibition. Psychological outcome measures were assessed by using the iconographical falls efficacy scale (concern about falling),21 the nine item patient health questionnaire (mood)22 and the COMPAS-W scale (wellbeing).23 Health related quality of life was measured with the 12 item WHO disability assessment schedule,24 the EuroQol five dimension five level (EQ-5D-5L) questionnaire,25 and the AQOL-6D (20 item assessment of quality of life six dimensions) questionnaire.26

Detailed self-report information on frequency and duration of physical activity was evaluated with the incidental and planned exercise questionnaire.27 Daily life activity was assessed with the McRoberts MoveMonitor (McRoberts, Netherlands) as the average duration of daily walking and standing, and the number of walking and standing bouts per day28; a bout was defined as a period of consecutive activity. Because participants were instructed to remove the device before going to bed, we required a minimum wear duration of 12 hours per day on one or more days for daily life activity data to be included in the analysis. Daily life activity data were collected over a median of six days (interquartile range one day) for both groups.

Process outcome measures included exercise duration and were captured through the tablet computer. Because participants were allowed exercise breaks when they were sick or went on holiday, we averaged weekly exercise duration as median values for participants as a robust measure of central tendency. We obtained subjective user experience by assessing usability, enjoyment, and exercise self-efficacy with the system usability scale,29 the physical activity enjoyment scale,30 the exercise self-efficacy scale,31 and the attitudes to falls related interventions scale.32

All outcome measures were assessed by trained exercise physiologists or physiotherapists who were blinded to group allocation. We assessed safety in terms of adverse events, which were defined as any fall related to the prescribed exercise programme or involving the intervention equipment.

Statistical analysis

Sample size calculation—based on previous evidence, we carried out an a priori sample size calculation8 in Stata using a custom code with 5000 simulations. The calculation showed that 500 participants were required to achieve 80% power to find a fall rate reduction of 33% (incidence rate ratio of 0.67) in the intervention versus the control group that is statistically significant at a P value less than 0.05 (considering an overdispersion of 1.2, 0.8 falls/person year in the control group, and a follow-up duration of 22 months to account for 20% dropout rate). We then ran power calculations in G*Power (version 3.1.7) for our secondary outcomes (considering an analysis of variance design with four measurements and 20% dropout rate). These calculations showed that we would have 90% power to detect a statistically significant (P<0.05) small reduction (effect size f=0.15) in concern about falling in the intervention group versus the control group, assuming a within subject correlation of 0.75.8 A subsample of 200 participants with repeat physical assessments would provide us with 95% power to detect a statistically significant (P<0.05) large reduction (effect size f=0.38) in postural sway in the intervention group versus the control group, assuming a within subject correlation of 0.76.8

Analysis plan—analyses were conducted according to the predefined statistical analysis plan, as registered on the OpenScience framework (https://osf.io/42gje/). We coded data to maintain group allocation blinding during analysis. Effectiveness analyses of the primary outcome were conducted on an intention-to-treat basis by a statistician (BT or NB) and independently replicated by one of the investigators (KSvS). The α level was set to 5%. Analyses were performed with Stata (version 16, Stata Corp) and SPSS (version 25, IBM Corp).

Missing data—participants who were randomly assigned to a group were included in the analysis irrespective of their level of compliance with their group assignment, which was in line with intention-to-treat principles. The primary outcome measures (number of falls per person year and proportion of people who fell over 12 months) were analysed without imputation or adjustment for descriptive characteristics, and with correction for follow-up duration when appropriate. We assumed that the faller status of people with incomplete follow-up (n=66 at 12 months and n=188 at 24 months, distributed evenly over the two groups) was maintained during censoring. We used Little’s missing completely at random test to determine the missing data patterns of secondary outcome measures. The secondary outcome measures were imputed using estimated means single imputation if they were missing completely at random; or under the assumption of missing at random using multiple imputation to create 20 imputation datasets under joint multivariate normal imputation33 if they were not missing completely at random. Psychological wellbeing, health related quality of life, and physical activity questionnaire data were missing for 58 out of a total of 503 people at six months, for 82 people at 12 months, for 98 people at 18 months, and for 99 people at 24 months. Daily life activity monitoring data were unavailable for 21 people at baseline, 101 people at six months, 138 people at 12 months, 148 people at 18 months, and 156 people at 24 months. Clinic based balance and neuropsychological assessment data were missing for 42 people at six months and for 47 people at 12 months. These data were missing because of dropout, scheduling issues, non-adherence, or technical problems. Little’s missing completely at random test indicated that all data were missing at random with respect to participant baseline characteristics.

Primary outcomes—primary outcomes were the number of falls per person year, and the proportion of people who fell over 12 months. The number of falls per person year was analysed using Poisson regression to estimate the difference in fall rates between the two groups. Incidence rate ratios and 95% confidence intervals are reported. Poisson regression was selected over negative binomial regression (as a priori registered in our statistical analysis plan, but not in our protocol paper) to allow for a direct comparison with our planned complier average causal effects analysis because this analysis was based on a Poisson model. Online appendix 1 presents the results of the negative binomial regression. Days of follow-up was included as an exposure term in these models; that is, the natural logarithm of the days of follow-up was added as an offset. We examined the proportion of people who fell in the two groups by using modified Poisson regression models for binary outcomes. Faller status was compared (no falls v at least one fall) and relative risks and 95% confidence intervals are reported.

Secondary outcomes—secondary fall outcomes were the number of falls, the complier averaged causal effect, the number of injurious falls, the proportion of people who fell, and the proportion of people who had injurious falls at 24 months. We used instrumental variable regression to correct for imperfect participant adherence and to gain insight into efficacy by estimating the complier averaged causal effect. We used a 2000 times bootstrapped, two stage complier averaged causal effect estimator composed of a linear regression with adherence as the dependent variable and group as the independent variable to obtain an estimate for adherence. A robust Poisson regression was then performed, with falls as the dependent variable and the natural logarithm of follow-up in days as exposure to estimate the effect of the intervention among people with perfect adherence. The number of injurious falls per person year was analysed using Poisson regression to estimate the difference in injurious fall rates between the two groups. We analysed secondary non-fall outcome measures with robust generalised linear models using an exchangeable working correlation matrix and compared the change in scores over time at six, 12, 18, and 24 months between the two groups. When the residuals of the generalised linear models deviated from normality, we used a 1000 times bootstrap for each imputation dataset to obtain confidence intervals.

Patient and public involvement

StandingTall was developed using consumer design principles. A group of older people were involved during the development of the StandingTall application. They were asked to evaluate an early version on its usability and age appropriateness as a means to engage in fall prevention exercises using tablet based technology. A two week feasibility study was conducted in 10 community dwelling older people in November 2013. The average age of the participants was 77.5 years (range 67-82 years), and six participants were women. Physiological profile assessment scores ranged from mild to marked (median z score 1.68; range 0.79-2.94) and seven participants had experienced falls in the previous year. Adherence was high, with participants reporting that the programme was suitable for older people. There was no other formal patient and public involvement in this study.