Page 52 - Journal of Special Operations Medicine - Spring 2017

Categorical data (i.e., hemorrhage control, casualty sta- composite being satisfactory; 44 tests had a bad out-
tus, and damage in 2x2 contingency tables) were com- come in that one or more components of the composite
pared by exposure status (unexposed–exposed) with use result were unsatisfactory. From the analysis of continu-
of a Fisher exact test. Then a Cochran–Mantel–Haenszel ous variables, these 44 test results were removed from
test was used to see if such exposure effects were differ- the subset to be analyzed, but none was removed from
ent among the three tourniquet models. Likelihood ratio overall analysis and from categorical data analysis.
p values were reported. Descriptive statistics were used
to portray results. Effectiveness Results by Tourniquet Model
The C-A-T’s effectiveness percentage was 91% (91 of
Continuous data (e.g., time to stop bleed, pressure, 100 tests); unexposed C-A-T devices had 100% effec-
blood loss) were summarized by mean ± standard error tiveness (50 of 50 tests), whereas exposed devices had
of the mean values and analyzed by using a mixed-model 82% effectiveness (41 of 50 tests; p = .003). The RMT’s
analysis of variance (ANOVA). Components within the effectiveness percentage was 98% (98 of 100 tests); un-
statistical model, including the user as a random effect, exposed RMT devices had 100% effectiveness (50 of 50
were presented as a percentage for each continuous pa- tests), whereas exposed devices had 96% effectiveness
rameter to estimate their restricted maximal likelihood (48 of 50 tests; all p = .495). The SOFTT-W tourniquet’s
variance. Fixed-effect tests were made by exposure effectiveness percentage was 94% (94 of 100 tests); un-
group, model of tourniquet, and groupxmodel interac- exposed SOFTT-W devices had 98% effectiveness (49 of
tion. Group means were compared by using a Tukey ad- 50 tests); whereas exposed devices had 90% effectiveness
justment within the mixed model. (45 of 50; all p = .204). The Cochran–Mantel–Haenszel
tests showed the C-A-T was significantly more suscep-
Additionally, if the tests of tourniquet failed to stop bleed- tible to exposure than the RMT or SOFTT-W (p ≤ .001).
ing, the skewing of continuous data was known to be
severe; for example, bleeding would only end upon the Casualty Survival Results:
death of the casualty at 2,500mL, compared with 150mL Overall and by Tourniquet Model
routinely measured with success in bleeding control. Fur- For the overall study, survival rate (alive–dead) was 95%
ther skewing from nontreatment effects, such as the ef- (284 of 300); survival of casualties was 100% (150 of
fect of the user, has been known by the investigators from 150) when unexposed devices were used and 89% when
their previous studies when users have different caregiv- the exposed devices were tested (134 of 150 tests; p <
ing strategies, skill levels, or experience levels. Given this, .001). The Cochran–Mantel–Haenszel tests showed the
a contingency was planned to analyze the subset of data C-A-T was significantly more susceptible to exposure
that were reliable. The plan was as follows: if a test of than the RMT or SOFTT-W (p ≤ 0.001).
tourniquet had a bad composition (i.e., the second com-
posite outcome was a bad result), we then removed all Major Damage Results:
such tests in which bleeding was not stopped (i.e., the Overall and by Tourniquet Model
subset analyzed had its tests with a result of composite For the overall study, major damage (yes–no) occurred
= good) and a two-way mixed-model ANOVA was per- in 1% of tests (4 of 300). Unexposed devices had 0%
formed, with the user as a random effect in the statistical major damage, whereas exposed models had signifi-
model to compare the continuous variables by exposure cantly more, at 3% (4 of 150 tests; p = .018). The Co-
status (yes–no) by model of tourniquet. chran–Mantel–Haenszel tests showed the C-A-T to be
marginally, although significantly, more susceptible to
For pairwise comparison of categorical data of tourni- exposure (p =.044).
quet models, a nonparametric Wilcoxon method was
used. For pairwise comparison of means of tourniquet Time to Stop Bleeding Results:
models, Student’s t test was used to stratify results. Sig- Overall and by Tourniquet Model
nificance for results was established when values were For the overall study, 14% of the variance of the time to
p < .05. All statistical analysis was conducted by using stop bleeding results could be attributed to the user. Over-
SAS software (SAS Institute; http://www.sas.com) and all, the mean time to stop bleeding results for all three
MS Excel 2003 (Microsoft; www.microsoft.com). models of tourniquet was 47 seconds. However, these re-
sults included all 44 tourniquet tests that had a bad com-
posite score. Excluding those 44 tests, the mean time to
Results stop bleeding was 27 seconds; for unexposed devices, the
mean was 26 seconds; and for exposed models, 29 sec-
Overall Results onds (p = .01). Results of mean time to stop bleeding by
Of the 300 tourniquet tests, 256 had a good outcome user were two tiered: user 1 (scientist) was slow (30 sec-
in that the composite result had every component of the onds) and user 2 (cadet) was fast (25 seconds; p < .001).

30 Journal of Special Operations Medicine Volume 17, Edition 1/Spring 2017