Page 78 - JSOM Winter 2024

landings), penetrating, blast-related, and other. Participants were SOF and CF participants with similar backgrounds and TBI
classified as having mild, moderate, or severe TBI 22,23 based on characteristics. SOF and CF are different in terms of intelli-
the highest rating from available severity metrics and consistent gence, physical fitness, and resilience prior to their service. 15,16
24
with the VA/DoD classification of TBI severity. These include Therefore, we selected variables on which to match SOF and
25
the Glasgow Coma Scale, the time to follow commands, dura- CF that would show differences prior to their service, in-
tion of posttraumatic amnesia, and neuroimaging. Participants cluding demographic variables. We chose not to control for
without a severity metric and who were not hospitalized or service-related variables (i.e., variables that could affect the de-
treated in an urgent/emergency care setting at the time of index velopment of PTSD after selection for SOF). The rationale for
TBI were classified as having mild TBI. 6 this choice was to examine whether career experiences (e.g.,
number of deployments, years in service, number of TBIs, ser-
A comprehensive evaluation of lifetime exposures was conducted vice branch) were related to rates of PTSD.
to account for lifetime history of TBIs sustained within and be-
yond military service. In addition to the index TBI abstracted Propensity scores were estimated in the original SOF and CF
from medical records, total self-reported injuries and whether samples using a logistic regression model including the SOF/
injury occurred during deployment were collected via The Ohio CF group as the dependent variable and age, sex, race (Black,
State University TBI Identification Method ( OSU-TBI-ID). White, or other), years of education, TBI severity (mild, mod-
26
The OSU-TBI-ID is a standardized interview designed to assess erate, severe), and days from injury to rehabilitation admission
lifetime history of TBI based on Centers for Disease Control and as independent variables. SOF and CF participants were 1:1
Prevention recommendations for TBI surveillance. matched on the propensity scores using a caliper of 0.01. After
matching, balance diagnostics (including standardized mean
differences and variance ratios) examined the balance of de-
Time-Varying Covariates
(Collected at Every Data Collection Timepoint) mographics and baseline characteristics in the matched SOF
and CF groups. To determine whether the matching process
Problematic Substance Use changed the overall pattern of results for the SOF participants,
Participants provided information about substance use at we fit a longitudinal model to the repeated measure of PCL-C
baseline and every follow-up assessment. Participants were and compared the total SOF and matched SOF groups.
asked whether they had used any illicit substances in the past
year and about the frequency and amount of alcohol use in In the matched samples of SOF and CF, a mixed-effects lon-
the previous 30 days. Problematic substance use was identified gitudinal model was fit to the repeated measure PCL-C and
as any of the following: illicit drugs in the past year; binge included SOF/CF group (A), TBI severity (B), time between the
drinking (≥1 day in the past month that participant had ≥5 PCL-C and injury (C), two-way interactions between A, B, and
drinks); or heavy consumption (for men >14 drinks per week; C (D), and a three-way interaction between A, B, and C. Men-
for women >7 drinks per week). 27 tal health treatment and problematic substance use in the past
year were fixed effects, and matched subjects were random
Mental Health Treatment effects. We included two-way and three-way interactions in
During baseline data collection, participants were asked if they the model because some studies indicated that the trajectories
had a history of mental health treatment prior to their TBI. of PTSD symptoms differ based on TBI severitiy. Since the
12
During every follow-up assessment, participants were asked if PTSD trajectory was not expected to be linear, we applied or-
they participated in mental health treatment in the past year. No thogonal polynomials of degrees 1–3 to the time variable (i.e.,
additional details about the type, quantity, or frequency of men- time of the PCL-C measurement since injury) and included all
tal health treatment were queried or included in the analyses. orthogonal polynomials in the model. The average levels of
PTSD symptoms and their 95% confidence intervals were esti-
Main Outcome: PTSD Checklist—Civilian Version (PCL-C) mated over time based on the model for SOF/CF with different
28
The PTSD Checklist—Civilian Version (PCL-C) is a 17-item TBI severities, respectively.
standardized self-report measure that assesses PTSD symptom
severity over the previous month. Instructions ask the partici- Results
pant to rate how bothered s/he has been by each PTSD symp-
tom in the past month on a 1 (Not at all) to 5 (Extremely) As shown in Figure 1, of the 1,413 participants with known
scale. Higher total scores indicate more severe PTSD symp- SOF status, 81 participants were excluded for not having a
toms. Scores >49 indicate a probable PTSD diagnosis in mili- PCL-C at any time point, and 141 were excluded due to miss-
tary and veteran samples. 28,29 PCL-C was collected at baseline ing data on the variables used to match the samples. A total
and every follow-up assessment. of 1,191 SM/Vs who completed at least 1 PCL-C at any time
point were included in the matching, resulting in 205 SOF and
Data Analysis 205 CF participants being matched on propensity scores with
Data were analyzed using statistical software R v4.1.2 (R a caliper of 0.01. Table 1 shows the demographics and other
Foundation for Statistical Computing, Vienna, Austria). Sum- baseline characteristics of the study samples before matching
mary statistics were expressed as means (standard deviations) (SOF, n=413; CF, n=778) and after matching (matched SOF,
for continuous variables and counts (percentages) for categor- n=205; matched CF, n=205). The standard mean differences
ical variables. and variance ratios demonstrated that the matched samples
were balanced with respect to age, sex, race, education, TBI
Propensity score matching was used to create matched sam- severity, and days from TBI to rehabilitation admission.
ples of SOF and CF participants with similar demographic and
baseline characteristics such that trajectories of PTSD symp- We examined whether creating matched samples of SOF and CF
toms (i.e., PCL-C total scores) would be compared between changed the overall pattern of results for the SOF participants.

76 | JSOM Volume 24, Edition 4 / Winter 2024