Page 100 - JSOM Spring 2023

number of injuries in each category, although one study pro- of the plot, reinserting the “trimmed” studies on the other side
vided these data directly. 55 of the plot, and imputing their counterparts on the original
side of the plot. A new RR and 95%CI are estimated that in-
Methodological Quality cludes these hypothetical missing studies.
The National Institute of Health Study Quality Assessment
56
Tool for Observational Cohort and Cross-Sectional Studies Results
was used to assess the methodological quality of the selected
studies. Studies were rated on a 14-point scale that included the Figure 1 shows the PRISMA diagram depicting the literature
following: clearly defined objectives and populations, partici- search. A total of 422 titles and/or abstracts were examined.
pation rates, selection bias, sample size justification, adequacy Of these, 96 full text articles were obtained and examined in
of exposure measure (BMI), adequacy of outcome measures detail. Nine articles were found that fully met the review crite-
(injuries), loss to follow-up, and statistical adjustment for con- ria and were included in the meta-analysis.
founders. The obtained score was divided by 14 (the number
of items) and multiplied by 100%, thereby converting the Table 1 shows the characteristics of the nine selected stud-
scores to a percentage of available points. Both authors in- ies. Most investigations involved individuals in basic military
dependently rated each selected article. After the independent training, 48,49,55,57,58,65,66 although two studies involved Soldiers
evaluation, the reviewers met to examine the other reviewer’s in operational military units. 67,68 SMs were from the US, 55,57,65–
scores and to reconcile differences. The final consensus score 67 Finland, 48,58 and Israel. 49,68 There was some overlap in the
of the reviewers served as the methodological quality score. US studies because several followed basic training cohorts
during the same years. 55,57,66 Height and weight were usually
Meta-Analyses directly measured, 48,49,55,57,58,67 although it was self-reported in
The Comprehensive Meta-Analysis Statistical Package, Ver- one case, and not clear in two cases. 65,68 When unclear, the
66
sion 3.2 (Biostat, Englewood, New Jersey, US) was used to authors reported weight/height as obtained from “administra-
perform the meta-analyses. Data from the 2X4 matrices were tive records” or “medical records.” Injury case definitions
68
65
used to calculate risk ratios (RR) and 95% confidence inter- varied. Most studies included various types of physical dam-
vals (95%CI) comparing healthy weight (reference cohort) to age to the body, 48,58,65–67 two studies examined specific injury
the other three BMI levels. Forest plots were constructed to types (stress fractures and knee meniscus damage ), and one
68
57
55
show risk ratios calculated from the data in each study. Two study did not provide a clear injury case definition. Method-
studies 55,57 providing separate data on large groups of male ological quality was generally very high with scores ranging
and female Soldiers were entered as separate cohorts into the from 71% to 86% of available points and seven studies scor-
analysis. Another study analyzing overuse and acute inju- ing 86%. 49,57,58,65–68
58
ries separately was also included as two data points into the
analysis. For comparisons of injury risk between healthy weight and
underweight SMs, Figure 2 shows the forest plot, Figure 3
The meta-analysis produced a summary risk ratio (RR) and shows the funnel plot, and the second line of Table 2 provides
95% confidence interval (95%CI) representing the pooled summary statistics. Overall, underweight SMs were at higher
results from data in the individual investigations. Three me- risk of injury than healthy weight SMs, but the studies showed
ta-analyses were performed comparing the risk of injury considerable heterogeneity as indicated by the Q-statistic and
among healthy weight SMs with those 1) underweight, 2) I test. The funnel plot indicated more studies were right of the
2
overweight, and 3) obese. A random effects model was used plot, but Egger test did not suggest publication bias. On the
for all analyses. Heterogeneity was assessed using the Q- and other hand, the Trim and Fill procedure suggested two studies
2
59
the I -statistics. Heterogeneity was the degree of variability were missing, and imputing these studies modestly increased
in the analyses. the RR.
Publication Bias FIGURE 2 Forest Plot of Studies Comparing Risk of Injury in
The Comprehensive Meta-Analysis Statistical Package, Ver- Healthy Weight to Underweight
sion 3.2, was also used to examine publication bias using fun-
nel plots, the Egger test, and the trim and fill procedure.
60
62
61
Funnel plots are graphs of each study’s logarithm (log) RR
plotted aganist the standard error (driven primarily by sample
size). Studies with larger sample sizes tend to cluster near the
top of the plot and near the pooled log RR, while studies with
smaller sample sizes are generally lower on the plot. If publi-
cation bias is present, the bottom of the plot tends to show a
higher concentration on one side of the plot because studies
with smaller samples are more likely to be published if they
have larger effect sizes. Funnel plots, although useful visual
indications of possible publication bias, have been criticized
owing to subjectivity of interpretation and differences arising To compare injury risk between healthy weight and over-
from different choices of axes. 63,64 Thus, additional statistical weight SMs, Figure 4 shows the forest plot, Figure 5 the funnel
61
indicators of publication bias were examined. The Egger test plot, and third line of Table 2 provides summary statistics.
evaluates bias by the amount of asymmetry in the funnel plot. Overweight SMs were at slightly higher injury risk than nor-
The Trim and Fill Procedure adjusts the funnel plot through mal weight, but estimates showed considerable heterogeneity.
62
an iterative process, removing studies concentrated on one side The funnel plot was relatively symmetric with a similar
98 | JSOM Volume 23, Edition 1 / Spring 2023