We randomly recruited 10 of 15 high schools of the Calgary Board of Education to participate in the fall of 2001. Computer-generated random numbers were used to recruit schools and students and to allocate the schools to the intervention or control group. We randomly selected 2 males and 2 females from physical education program rosters in each of grades 10 to 12. If a subject declined participation or dropped out after the baseline assessment but before a follow-up assessment, we recruited another student of the same sex from the same school and grade. The study was blinded in that we randomly allocated schools to the intervention or control group following initial subject recruitment. All assessments were performed by a physiotherapist.
We included subjects if they were between the ages of 14 and 19 years, regularly attended classes and participated in physical education classes. We excluded subjects if they had a history of a musculoskeletal injury in the 6 weeks before recruitment, a previous history of a serious musculoskeletal disorder (e.g., fracture, rheumatologic disease, systemic disease or surgery) or an important medical condition (e.g., hypertension, or recurrent fainting or dizzy spells).
Each subject was asked to complete a baseline questionnaire, which included questions about previous history of injuries and participation in sports. At the initial assessment, the physiotherapist measured each participant's height and weight. Each subject completed, with their eyes closed, a timed static unipedal balance test on the gym floor and a timed dynamic unipedal balance test on an Airex Balance Pad (Fitter International Inc., Calgary). We have previously shown adequate test–retest reliability for these 2 measurements (intraclass correlation 0.7 and 0.5 respectively).24
During these tests, time was recorded when the subject's balance was lost or eyes opened, or when the maximum time allowed for each trial (180 seconds) was reached. The baseline assessment also included a vertical jump test25
to examine functional strength and the Canadian version of the 20-m shuttle run to test endurance.26
A physiotherapist taught each participant in the intervention group a progressive, home-based, proprioceptive balance-training program to be used daily for 6 weeks and then weekly for maintenance for the remainder of the 6-month study period. A 16-inch (40-cm) wobble board (Fitter International Inc.) was provided. At the 2- and 4-week follow-up assessments the program was reviewed and progressed. Progression at 2 weeks included bipedal to unipedal exercise progression and increased duration of eye-closed elements of the program. At 4 weeks progression involved wobble board adjustment to level 2, which increased the amount of wobble board instability. Core stabilization, including isometric contraction of abdominal and gluteal muscles was incorporated into the program. Each daily session was expected to last about 20 minutes, and self-reported compliance with the training program was assessed by a daily record sheet and weekly telephone calls over the 6-week training period. Each subject was retested (i.e., balance, vertical jump and shuttle run tests) biweekly over 6 weeks by the physiotherapist. For the 6-month follow-up period, each subject was asked to complete a sport participation record sheet and an injury report form as required. An athletic injury was defined as any injury occurring during a sporting activity that required medical attention (i.e., visit to an emergency department or physician's office, chiropractic, physiotherapy or athletic therapy) or resulted in the loss of at least 1 day of sporting activity, or both. Injury report forms included a section to be completed by any attending medical professional. The physiotherapist made biweekly telephone calls to all study participants during the 6-month follow-up period to ensure that all eligible injuries were reported.
The primary outcome measures included the change from baseline to 6-week follow-up in the maximum time that balance was maintained during the static and dynamic tests over 6 trials, 3 for each leg. Measurements for both legs were pooled because there is no evidence that balance differs by side.24
The primary injury outcome measure included all self-reported sports-related injuries and ankle sprain injuries.
Because of the cluster randomization design of the study, to calculate the sample size we had to take into account the possible similarity in the response of individuals within each cluster.24
We assumed an intracluster correlation of ρ = 0.01 based on a comparison with the mean ρ found by Murray and associates27
of ρ = 0.006 in examining adolescent smoking behaviour. We also adjusted for a potential drop-out and noncompliance rate in the intervention group and a contamination rate in the control group (Ro
= 0.10). On the basis of the primary outcome variable static balance, this trial was powered to detect an effect size of d = δ/σ = 0.8 (where δ = μ1
= mean [intervention group] – mean [control group] = 9 seconds; and σ = 11 seconds = the estimated common standard deviation of the timed balance test measurement in the control and intervention group), assuming a type I error (α = 0.05) and type II error (β = 0.10).
We report descriptive statistics for baseline characteristics. Baseline variables were compared between the 2 groups. We calculated the mean difference in static and dynamic balance test results from baseline to the 6-week follow-up for the intervention and control groups and compared them using both independent and cluster-adjusted t
Where the assumptions of normality and equal variance were not met, the data were logarithmically transformed.29
In this case the measure of central tendency used was a geometric mean, which was estimated by back-transformation from the mean of the log-transformed data. Our analyses were based on the intention-to-treat principal. We used multivariable mixed-effects regression analysis (i.e., allowing random effects for cluster) to examine further the effectiveness of the training program in improving both static and dynamic balance test results, controlling for other baseline covariates.28
To determine our final model, we eliminated covariates through a stepwise process, with α at 0.05.
We compared injury incidence rates in the 2 study groups by calculating the relative risk. Given the small number of injuries reported in 6 months, the intracluster correlation coefficient was calculated on the basis of the history of injuries reported in the year before study enrolment. A cluster-adjusted χ2
analysis was not warranted, because the estimated intracluster correlation coefficient was negative and hence given the value 0.28
This indicates that cluster randomization was not expected to affect the outcome related to comparison of injury rates. Stratified analysis based on previous injury was also examined using Fisher's exact methods.29
We used logistic regression to examine the effectiveness of the training program in reducing injury, while controlling for other baseline covariates.