Study design
This study (“LiFE-is-LiFE”) was a multi-center, single-blinded, randomized non-inferiority trial conducted in Heidelberg and Stuttgart, Germany. The full study protocol is available elsewhere [19]. The study was preregistered under clinicaltrials.gov (identifier: NCT03462654) on March 12th 2018. Reporting in this article is aligned with the CONSORT extension in non-inferiority trials [18] [see CONSORT checklist, Additional file 3].
In addition to baseline assessment, follow-up assessments were carried out six and twelve months after intervention start (reference was the date of the first (g)LiFE session), with a tolerance of ± 2 weeks.
Participants and eligibility criteria
For recruitment purposes a list of all persons aged 70 + was drawn from municipality registries in both cities. Persons were drawn consecutively in waves of between 250 and 1.000 persons and contacted between April 2018 and July 2019 by mail. If interested, participants could contact the study sites for a first eligibility screening by telephone. In case of a positive telephone screening, an inhouse screening was scheduled. Participant flow is depicted in Fig. 1. To be included in the study, participants had either a) experienced at least one injurious or more than one non-injurious fall in the year prior to study participation according to self-report, or b) were designated as having high risk of falls when indicating self-perceived balance decline and needing ≥ 12 s for the “Timed Up-and-Go” (TUG) [20] test. Those who already exercised more than once per week or indicated to carry out more than 150 min of moderate to vigorous PA per week were excluded. A detailed list of further exclusion criteria is provided in the study protocol [19].
Randomization and blinding
Participants were randomized after baseline assessment into one of the two intervention arms through block-randomization. Apart from block sizes, randomization was concealed, i.e., staff was not aware of the sequence before randomization. Randomization and group assignment was carried out by the study site coordinators (CPJ, CN) in an externally managed database without possibility to alter group allocation afterwards. In case of withdrawal from intervention, participants were still eligible for follow-up assessments. Assessors were blinded to group allocation at all times.
Intervention programs
A detailed description including a TIDieR checklist of both intervention formats is included in the study protocol [19]. In the LiFE program, balance and strength activities as well as general PA promoting activities are embedded into everyday tasks and routines, with the overall aim to integrate them in a way that these activities can be performed multiple times a day [14]. As there was no standardized group format of the LiFE program available, gLiFE had been developed according to Medical Research Council guidelines [21] and piloted in advance to the intervention start [17]. In both intervention arms, intervention components were taught in accordance with the LiFE trainer’s manual [15], including strength and balance activities as well as strategies to enhance physically active behavior and to habitualize activities as part of individual daily routine. LiFE and gLiFE were delivered in seven sessions within eleven weeks, either in a group (gLiFE) or at the participant’s home (LiFE), followed by two booster phone calls in week four and ten after the last intervention session. During the intervention sessions a total of seven balance activities, seven strength activities for the lower extremities, and two PA promoting activities were delivered. To help participants establish a LiFE routine as part of their daily life, they learned how to independently select, execute, and adapt intensity of activities, and how to identify appropriate daily situations in which LiFE activities can be integrated. gLiFE group sessions were scheduled for two hours and held by two trainers with a maximum of 12 participants; LiFE sessions lasted approximately one hour and were delivered by one trainer. Trainers were either physio therapists, sports scientists, health psychologists, or occupational therapists who had attended a two-day workshop prior to the start of the intervention delivery, including a certification test.
Outcomes
Primary outcomes
Combined endpoint: Falls adjusted for PA. To measure PA, “activPAL4™ micro” accelerometers (PAL Technologies Ltd., Glasgow, Scotland) were attached to participants’ central front thigh at baseline, 6-, and 12-month follow-up to continuously measure PA under “free-living” conditions for seven days (24 h), i.e., activPALs were posted back to the respective study centers no earlier than the start of the ninth day of measurement. The sensor was wrapped in a nitrile finger cot fixed with a waterproof, adhesive, transparent film. The device has shown good to excellent reliability and validity [22]. If the device was removed earlier, data were used if at least two weekdays and the Sunday of the respective week were fully captured [23]. Given that walking activity can be seen as the most hazardous PA when it comes to risk of falling [9, 10], PA exposure was operationalized as mean steps/day.
Falls were defined as “an unexpected event in which the participant comes to rest on the ground, floor, or lower level” [24] and were recorded using a monthly falls calendar sent back by use of preaddressed and prestamped envelopes. In case of a fall, information on location, date, time, injuries, subsequent treatment related to the fall, and movement during which the person has fallen had to be provided on the calendar sheet. Following recommendations of Gillespie et al. [25], falls were followed-up via telephone calls to ascertain additional information and to determine the current health status of the person.
Intervention costs. Intervention costs were calculated as costs per participant for each group (LiFE/gLiFE). Personnel and material costs, trainers’ and participants’ travel expenses, and room rent were taken into account. The average group size of gLiFE sessions was 7.9 persons. The duration of the sessions (including time for travel and preparation) resulted in 1.8 (LiFE) and 3.0 (gLiFE) personnel hours per session. Personnel costs per hour were derived from the German wages agreement for civil services 2018 (“TVöD” salary level E13 and E10). Costs for materials, manuals, and working books were considered. Moreover, a room rent of €50 per day for the trainer workshop or per gLiFE session in one of the study centres was also taken into account. In the other study centre, a suitable room was available on site for the gLiFE sessions, therefore no room rent was incurred there.
Since study conditions deviate from conditions in case of an implementation in the “real world”, interventions costs were also calculated for another scenario, based on assumptions that the project team considers to most realistically represent the implementation conditions. In this “real world” scenario, it was assumed that 20 trainers with a salary according to “TVöD” salary level E8 participate in the trainer workshop. It was assumed that on average 12 persons attend the gLiFE sessions and that each gLiFE trainer pair would conduct 12 training sessions per week, while one LiFE trainer could conduct 15 sessions in the same time. The duration of the LiFE/gLiFE sessions (including time for travel and preparation) and phone calls were assumed to be 2.0 h/2.5 h and 0.5 h, respectively. For both interventions, no room rent was assumed. Furthermore, each trainer or trainer pair was assumed to have their own material set. The data and assumptions underlying the calculations of each scenario are summarized in an additional table [see Additional file 1].
Secondary outcomes
Physical activity. Mean steps/day were assessed to serve as offset variable in the primary outcome analysis to adjust falls for PA, and as PA outcome in itself.
Fall outcomes. Falls were assessed and defined according to Lamb et al. [24], that is, number of falls, fall rate per (half) person year, time to event (either fall or end of observation), number of fallers, and frequent fallers (i.e., more than one fall in the past six months). Fall consequences were categorized into minor, moderate or serious injuries according to a standardized system incorporating symptoms as well as medical care use [26].
Motor capacity. Gait performance was measured in terms of 7 m gait speed at comfortable and fast pace. The 30 s chair rise was used to evaluate functional leg strength [27]. Static balance was assessed using the adjusted eight level balance scale developed by Clemson et al. [14].
Functional (dis-)ability. The Late Life Function and Disability Instrument (LLFDI) was used to assess participants’ difficulties in performing 32 different upper and lower extremity physical activities and actions as well performance of another 16 socially defined life tasks.
Adherence. We followed the consensus agreement by Hawley-Hague et al. [28] who recommend reporting adherence in terms of completion (attendance of at least > 75% of sessions is defined as completion [28]), attendance (percentage of sessions attended out of the actual number of sessions), and duration adherence (adherence to predefined LiFE activities at home, assessed using the Exercise Adherence Rating Scale (EARS) [29]). The EARS ranges from 0 to 24.
Fear of falling and balance confidence. Participants’ fear of falling was assessed using the Short Falls Efficacy Scale-International [30], a self-rating scale including 7 items ranging from ‘not at all concerned’ (1 point) to ‘very concerned’ (4 points) and resulting in values between 7 (‘not concerned about falling’) and 28 points (‘very concerned about falling’). The Activities-specific Balance Confidence Scale (ABC) was used to measure participants’ confidence in maintaining their balance while performing certain daily activities.
For participants’ characteristics, age, sex, body-mass index, number of medications, number of comorbidities, falls in the past six months, and cognitive status (Montreal Cognitive Assessment) were assessed.
Sample size and non-inferiority margin
Sample size was calculated based on 12 month data from the original LiFE study [14]; information on this calculation can be found in the study protocol [19]. As outlined in the limitations section COVID-19-induced changes have been made to the methods used for the present analyses. We used 6-month instead of 12-month data to determine non-inferiority of the primary outcome falls per PA. However, we kept the non-inferiority margin (∆) as stated in the study protocol [19], that is, we accept a 20% difference in this outcome as a comparable reduction. As intervention costs of gLiFE are expected to be lower than of LiFE, no non-inferiority margin is defined for this outcome.
Statistical analyses
The analyses were carried out according to both the intention-to-treat principle (ITT) and the per-protocol principle (PP) to determine the robustness of the results due to missing values [31]. As dates were fixed for the gLiFE sessions, it was expected that some participants might be unable to attend all seven sessions. Therefore, attendance of a minimum of five sessions per participant was preset to assign participants to the PP sample. In accordance with the ITT principle, all randomized participants who completed baseline assessment were included, regardless of whether they had completed the intervention or prematurely dropped out of the study. In addition to missing information due to drop-out, there was occasional missing information in cases that otherwise completed the follow-up assessment. Overall, the percentage of missing values varied between 0 and 17% across different variables. As imputation of missing values is recommended for missing rates above 5% [32], missing data were imputed using multiple imputation by chained equations (MICE) with predictive mean matching as imputation method [33]. In total, 10 datasets were created based on data from baseline and 6-months’ follow-up assessments and analysed separately. Rubin’s rules [34] were applied to pool results from each dataset.
Negative binomial regression was used to compare incidence rate ratios (IRR) of falls between gLiFE and LiFE,taking into account possible overdispersion. In the model for the combined endpoint–falls per PA–mean steps/day were log-transformed and incorporated as exposure variable (offset). Confidence intervals for explorative comparison of changes between baseline and 6-month follow-up in secondary outcomes were obtained using a generalized linear model with repeated measures.
For the primary outcome, non-inferiority was indicated if the upper limit of the two-sided 95% confidence interval (CI) for gLiFE remained below the relative margin (∆) of 20% from LiFE (IRR = 1.20).
Analyses were performed using SPSS (IBM Corp. Released 2020. IBM SPSS Statistics for Windows, Version 27.0. Armonk, NY: IBM Corp). Multiple imputation of missing values was performed using STATA/SE 16.0 (StataCorp. 2019. Stata Statistical Software: Release 16. College Station, TX: StataCorp LLC).