Exercise and Unintentional Injuries: Meta-Analysis of Randomized ...

Exercise and Health Exercise and Unintentional Injuries: Meta-Analysis of Randomized Controlled
Trials
George A. KELLEY, Kristi S. Kelley School of Medicine, Department of Community Medicine, West Virginia
University, Morgantown, WV, USA
ABSTRACT
Kelley GA, Kelley KS. Exercise and Unintentional Injuries: Meta-Analysis of
Randomized Controlled Trials. JEPonline 2009;12(4):8-22. One of the major
reasons given for not participating in a regular exercise program is fear
of injury. The purpose of this study was to use the meta-analytic approach
to examine the overall risk of dropouts as a result of unintentional
injuries in adult humans participating in randomized controlled exercise
intervention trials. A random sample of studies was selected from an
electronic search of 1,594 citations identified in PubMed. Data abstraction
was performed by both authors, independent of each other, with every item
reviewed for accuracy and consistency. A random effects model that
controls for heterogeneity was used to analyze the event rate (ER) for
dropping out of a study due to an unintentional injury. Heterogeneity was
examined using the Q and I2 statistic. Thirty-three outcomes representing
1,211 men and women (677 exercise, 534 control) met the criteria for
inclusion. Overall, the ER for dropouts as a result of unintentional
injuries was 0.073 (95% confidence interval, 0.037 to 0.140). No
statistically significant heterogeneity was observed (Q = 2.4, p > 0.99; I2
8 weeks, (3) adult humans > 19 years of age, (4) non-
athletes, (5) studies published in English-language only, (6) studies
published in journals between January 1, 1995 and October 31, 2005, (7) not
a cardiovascular, cardiopulmonary, or neuromuscular rehabilitation study,
(8) comparative control group that did not exercise, (9) data available on
dropouts and unintentional injuries. The inclusion of studies was limited
to randomized controlled trials because it is the only way to control for
unknown confounders. If data on unintentional injuries was not provided,
contact was made with one or more investigators to obtain such. All studies
were selected by both authors, independent of each other. Discrepancies
were resolved by consensus. Data Extraction
Codebooks were developed that could hold up to 146 items from each study.
The major categories of items that were coded included (1) study
characteristics (for example, country in which the study was conducted),
(2) participant characteristics (for example, gender, age, height, body
weight), (3) training program characteristics (for example, type, length,
frequency, intensity, duration, mode), and (4) primary outcome (number of
participants who dropped out of the exercise program as a result of an
unintentional injury). Data on participants who may have been injured
during the exercise intervention but did not drop were excluded because it
was expected that an insufficient amount of data would be available for
analysis given the expected sample size. All studies were coded by both
authors, independent of each other. The authors then reviewed every item
for accuracy and precision. Disagreements were resolved by consensus.
Inter-rater agreement using Cohen's kappa was 0.88 prior to correcting
discrepant items (8). Statistical Analyses
Calculation of study-level estimates for unintentional injuries. The
primary outcome for this study was the event rate (ER) for dropouts as a
result of unintentional injuries in the exercise groups. For each group
from each study, this was calculated as follows:
[pic] Where ninje is the number of participants in the exercise group who dropped
out of the intervention because of an unintentional injury and ntote is the
initial total number of participants in the exercise group. Control group
data were not included in these estimates because this would have resulted
in zero events for all control groups. The standard error for the ER was calculated as follows: [pic] where ER is the event rate for unintentional injuries and ntote is the
initial total number of exercise participants. For those cells that were
empty, i.e., zero (0) events, a continuity correction of 10-4 was added in
order to calculate the standard error for each ER (9). Pooled estimates for unintentional injuries. After calculating treatment
effects and variances for each outcome from each study, all results were
pooled using a random effects model, an approach that accounts for between-
study heterogeneity (10). For all analyses, the ER was analyzed using the
logit transformation and then back transformed to the original ER for
presentation purposes. In addition to using a random effects model, heterogeneity based on a fixed
effects model was examined using the Q statistic (11). An alpha value of <
0.10 for the Q statistic was considered to be indicative of statistically
significant heterogeneity. In addition, the consistency of between-study
findings for all outcomes was examined using a recently developed extension
of Q known as I2 (12). This was calculated as: