Page 20 - JSOM Fall 2018

of case order might affect performance. Case 3 had uniformly 4, and 5 averaged 2.5, 3, and 3.3, respectively. Turn number
satisfactory results, although a cadet and the fellow paused a appeared to be associated with patient build, because turn
couple of seconds in their first test at the moment the word number was increased with greater limb girth (Figure 2). The
“child” was heard. Case 3 was to be mechanically easy but lowest average was for the patient with a small build, and the
psychologically surprising, and the two users slowed their highest average was for the patient with a large build and the
movements briefly as they mentally figured out that there was need for use of a second tourniquet side by side with the first.
nothing extra to do. Case 4 had 75% (nine of 12) satisfactory
results; the three unsatisfactory tests resulted from an incor­ FIGURE 2 Results of windlass turn number by build of patient.
rect tourniquet placement, but none ended in a bleeding sta­
tus. Case 4 was to be moderately difficult because it simulated
care under gunfire, but such was a common way users had
learned and practiced previously. Case 5 had 58% (seven of
12) satisfactory results; five unsatisfactory tests were from an
incorrect second tourniquet placement (one of which also had
a first tourniquet incorrectly placed), and three of these five
tests ended in a bleeding status. Case 5 was designed to be
most challenging due to its extra steps and decisions.

User performance varied. Users 1, 2, 3, and 4 had unsatisfac­
tory test counts of four, three, two, and one, respectively. Plots The chart shows the mean number of windlass turns by build of pa­
tient with case information. On average, turn number appeared to be
showed two patterns of learning (Figure 1). By block, satisfac­ associated with limb girth.
tory results were 50% (six of 12) for block 1 and 83% (10 of
12) for blocks 2 and 3. As a group, users learned quickly from Discussion
the first to second block but then plateaued.
Our key finding in this study was that the judgment results
FIGURE 1 Performance and prediction results by user.
have plausible applicability to first aid. This finding served
our purpose because we apply tourniquet science to medical
readiness. Our preliminary look at how users of limb tour­
niquets made judgments was in line with our expectations.
The results of the answers and the performances were anal­
ogously similar to those seen in previous studies of judgment
science. Such plausibility may indicate that development of
feasible techniques and methods may eventually lead to de­
veloping valid tools for first­aid caregivers in the operational
health community. Several topics may be worthy of research
hypotheses (Table 3). We usually have studied simulated care­
giving, but in this study, we specifically changed our view to
The graph depicts performance results from the manikin for the four simulate the simulation of caregiving. By simulating the train­
users. There are 12 tests per user because the three tests without bleed­
ing dropped out of this analysis; we lettered the 12 to differentiate these ing of caregivers with different clinical cases, we took a direct
results from those of all 15 tests per user. These results were accrued look at how tourniquet users actually perform when the sit­
test by test so that the y­axis is a percentage (number of satisfactory uation has a moderate degree of uncertainty. In the past, we
tests divided by number of accrued tests). The reliability threshold is had repeatedly tested only one case to look at topics other
drawn at 80% (black dashed line), and two users (users 3 and 4, green than judgment. There is not much uncertainty in test numbers
line and triangle and red line and square, respectively) attained it. By
user, their self­predicted point of becoming reliable is overlaid by an 2–140 if the user remembered the case information test to test.
oversized marker. The three users (users 1 through 3) with less experi­ The within­block randomization led to a moderate degree of
ence started with an unsatisfactory result. The two users (users 1 and uncertainty for the first block; and the second block had less
2, violet line and circle and blue line and cross, respectively) with least uncertainty than the first, because all five cases were repeated;
experience had similar results throughout as most (75%; nine of 12)
of their data points were identical, and these two were the only users and the third block had the least uncertainty, because the user
who did not attain reliability. These plots are a type of learning curve, became more familiar with the cases as experience was ac­
and the expert, user 4 (red line and square), shows a nearly flat per­ crued. We do not recommend testing user judgment in this
formance, an effect commonly seen after prelearning. The other three way for novice users, because the skills required are multiple
users showed a typical pattern. The metric here starts (at the left) as a and complex, but we found that the advanced beginners here
binary measure (either 0% or 100%), because the first try is either sat­
isfactory or unsatisfactory. Such small numbers in the numerator and (two cadets) and the competent user (the fellow) realized the
denominator greatly affect the percentage. At the end, the percentages testing to be worthwhile and hard. The expert found the test­
tend to change little, and in the middle, changes from test to test are ing to be a good drill. It also changed how he thought about
moderate. The area under these curves distorts the amount of success matching types of testing to the skill levels of users. Further­
for each user, but the curves show trajectories of reliability by user.
more, the change by block in performance for the group of
users may indicate a learning curve, because performance im­
When ease of use was parsed by results among 48 tests with proved quickly then flattened.
bleeding, satisfactory results averaged 4.74 (very easy was 5)
and unsatisfactory results averaged 4.05 (easy was 4). The first minor finding was a surprise knuckleball, because
we had designed a new challenge of judging when and how to
When turn number of windlass use was parsed by cases among use two tourniquets side by side, but we chose a setting of care
48 tests with bleeding, case 2 averaged 3, whereas cases 3, under gunfire. However, after we started collecting data, such

18 | JSOM Volume 18, Edition 3 / Fall 2018