Literature Search Strategy
Two independent researchers (A.Z., M.H.) performed a search for articles published between 1966 and February 2009 in the following databases: Cochrane Bone, Joint and Muscle Trauma Group Register; Cochrane Controlled Trials Register; MEDLINE; EMBASE; PEDro (Physiotherapy Evidence Database); and SCOPUS. The key words and phrases (in different combinations) searched were neuromuscular, sensorimotor, kinaesthetic, proprioceptive, balance, balancing, training, exercise, program, wobble board, postural control, perturbation, balance board, proprioception, coordination, agility, jump, jumping, performance, reaction, muscle, strength, sprint, and reflex. References listed in papers and cited references were also examined to identify additional studies. Both English and German trials were considered for this review.
Title, abstract, and key words sections of identified studies were examined by the 2 independent reviewers to determine whether they met the following inclusion criteria: RCT or CT, balance training of the intervention group, no balance interventions for the control group, and physically active participants up to 40 years of age without injuries or surgeries within the last 6 months or chronic instability of the lower extremities. For both research questions, the outcomes of interest were postural control, muscle strength, agility, jump performance, and sprint time for functional performance and muscle reflex activity, rate of force development (RFD), reaction time, and electromyography for neuromuscular control. Trials with a combination of balance training and other interventions (eg, multi-intervention programs) were excluded from this review. Papers with imprecise abstracts were considered for full-text analysis. Disagreements between the reviewers regarding the eligibility of studies were solved by consensus. Persisting disagreements were discussed in the monthly consensus meetings of all coauthors.
Relevant information in the selected studies was extracted by the 2 independent reviewers using predetermined extraction forms. Data of interest were research question, methodologic assessment, participants, training intervention, outcome, timing of the outcome assessment, and results. To ensure agreement between the reviewers in selecting study characteristics, we pilot tested the extraction form on 5 included articles before data extraction began. Discrepancies in data extraction were solved by discussion. The final data reports were based on consensus of the reviewers.
We used 2 methods to evaluate balance training effects. First, the individual results were summarized as reported in included trials (eg, differences in group changes over time). Second, we used Review Manager (version 5.0; The Nordic Cochrane Collaboration, Copenhagen, Denmark) to calculate standardized mean differences (SMD
the Hedges adjusted g, defined as the difference between the posttest treatment and control means divided by the pooled SD) and 95% confidence intervals (CIs) for each trial when sufficient data were available. When comparable data from multiple studies were available, they were pooled using a random-effects model. The random-effects model accounts for the heterogeneity of studies. Heterogeneity (variability in intervention effects among studies) was assessed by using χ2
statistics and 95% CIs.
Included studies used various assessment and data analysis methods for measured outcomes, such as postural sway or muscle strength, and often multiple results were presented for a single outcome (eg, different sway directions). To avoid bias, representative data were used for overall effect-size calculation. For example, if multiple variables were available for postural sway, we defined the sway path or medial-lateral sway of the dominant or right leg under the eyes-open condition as appropriate for meta-analysis. Furthermore, differences in training effects between athletes and nonathletes as well as between programs of various lengths were reported when sufficient data were available.
The methodologic quality of all eligible studies was independently examined by the 2 reviewers. For this approach, the scale of van Tulder et al9
for the assessment of internal study validity was used. Neglecting the criteria of participant and therapist blinding, we shortened the original van Tulder scale by 2 of the 11 criteria. Consequently, the modified van Tulder scale in this review included the following items: (1) acceptable method of randomization, (2) concealed treatment allocation, (3) similar group values at baseline, (4) blinded assessor, (5) avoided or similar co-interventions, (6) acceptable compliance, (7) acceptable dropout rate, (8) similar timing of the outcome assessment in all groups, and (9) intention-to-treat analysis. Adequate methods of randomization included a computer-generated random-number table or use of sealed opaque envelopes. Methods of allocation using date of birth or alternation were not accepted as appropriate.9
Compliance with the interventions, determined by training diaries or monitoring, should not have been less than 75%. A dropout rate of up to 25% was considered acceptable for follow-up of less than 6 months, and a dropout rate of up to 30% was considered acceptable for follow-up of ≥6 months. The 9 criteria for assessment of methodologic quality were scored with yes
, or (in case of inadequate reports) unclear
. For each yes
score, 1 point (on the van Tulder scale) was given. On the summary quality score (maximum of 9 points), at least 50% yes
scores were needed for high quality.9
The methodologic quality assessment was pilot tested by the reviewers for agreement on a common interpretation of the items and their implementation. The consensus method was used to discuss and resolve disagreements between the reviewers.