Evaluating the Relationship between Psychological Traits and Resilience to Musculoskeletal Injuries in Combat Control Graduates by Mendelian Randomization

Psychological trait identification

A literature search was performed to identify psychological traits relating to successful completion of the training for combat control (CCT). Specifically, we searched for publicly accessible papers using PubMed. The search criteria used was “combat control” or “special tactics” and “graduation” or “training” and “psychological” or “selection” or “behaviors.” These journal articles were supplemented with limited access studies published in the Defense Technical Information Center (DTIC).

We included all studies that reported psychological traits associated with the outcome of interest, whether quantitative or qualitative in design. We excluded studies that reported solely cognitive (e.g., scores from the Armed Services Vocational Aptitude Battery) or physical (e.g., 1.5 mile run time) traits. Since CCT is a unique career field within the U.S. military, we excluded studies performed in non-U.S. service-members. We included all traits reported from the studies regardless of if additional studies support or rebut their inclusion. Additionally, we considered CCT graduation and emotional reactions to killing independently, meaning that if a trait was related to both, then we included it in both lists.

Explicit inclusion criteria for studies related to completing CCT training:

1. Study must be primary literature (reviews, books, meta-analyses, editorials, commentaries and similar were excluded);

2. Participants were US military service members who had entered CCT training (other special warfare career fields and non-Air Force special operations forces were excluded as this study’s focus is on the CCT operator);

3. Participants completed a personality inventory or emotional intelligence questionnaire;

4. Psychological traits identified were reported alongside the inventory or questionnaire used; and

5. Participants were classified based upon completing the full training pipeline or attrition due to either self-elimination or cadre-determined failure.

GWAS identification

We searched the IEU OpenGWAS Project [23] for each psychological trait identified and for traumatic musculoskeletal injuries (fractures, tears, sprains, strains, and dislocations), pain, and joint damage (bursitis, tendinitis, synovitis, etc.). We restricted fractures to exclude those phenotypes caused by osteoporosis; involving operative procedures; occurring to the face, skull, or head; or resulting from specific causes (e.g., “falls resulting from simple fall”). We similarly restricted pain to those phenotypes only located in joints, limbs, or muscles (e.g., no generalized, chest or abdominal pain) and excluded phenotypes involving treatments or pain medications. Finally, we did not include sub-populations studies since not all phenotypes were divided by populations. For example, the phenotype “knee pain for 3+ months” included the two studies “ukb-e-3773_CSA” and “ukb-b-8906”, and we excluded the former. The OpenGWAS Project contains the summary statistic (aggregate data) results from 40,027 GWAS datasets (as of April 2022) and uses a standardized format for storing the data [24]. The MSI phenotypes used are provided in Table 1 and study identifiers are included in Supplementary Table 1.

Mendelian randomization

We performed two-sample Mendelian randomization (2SMR) estimates using the TwoSampleMR R package [25]. We used the MR Egger regression [26], inverse variance weighted (IVW) estimator [27], weighted median estimator [28], and Wald ratio estimator [29] algorithms. Instrumental variables (IV, SNPs associated with the exposure) were extracted by P-value thresholds 5×10⁻⁵ and 5×10⁻⁸. We excluded SNPs in strong linkage disequilibrium (LD) to reduce bias and used a clumping process with European samples from the 1,000 Genomes Project (R² < 0.001, window size = 10,000). If SNPs identified in the exposure dataset were not in the outcome dataset, proxy SNPs in LD (r² > 0.9) were used as instrumental variables. For the sensitivity analysis, we performed heterogeneity testing using Cochran’s Q and I² analyses [29] and tested pleiotropy on the weighted median estimation results [28].

Literature review identifies psychological traits associated with graduating CCT training

Our literature search identified 5 studies reporting psychological traits related to graduating from CCT training (Figure 1). We identified 29 traits associated with graduating CCT training from the psychological literature (Table 2). These included the aforementioned associations of NEO personality inventory domains extraversion and conscientiousness [17, 31] and traits from the EQI [19]: general mood, intrapersonal, stress management, optimism, happiness, and adaptability. Of the 29 traits, 15 were found in the EQI and the remaining 14 were from the NEO personality inventory.

• Download figure
• Open in new tab

Figure 1:

PRISMA flow diagram for literature review to identify psychological strengths of CCT graduates

From the IEU OpenGWAS project, we identified GWAS phenotypes corresponding to the aforementioned psychological traits for completing CCT training. We identified 31 distinct GWAS phenotypes for traits associated with the former (Table 3). We found only 14 traits with GWAS summary statistic datasets, but a total of 46 studies were identified. Nine traits were reported in multiple studies: neuroticism (14 studies); health satisfaction (5); seen a general practice doctor for nerves, anxiety, tension or depression (5); self-injury such as self-cutting, scratching or hitting (4); seen a psychiatrist for nerves, anxiety, tension or depression (4); anxiety/panic attacks (3); financial situation satisfaction (2), mental health problems diagnosed by a professional for anxiety, nerves, or generalized anxiety disorder (2), and a derived neuroticism score (2).

The causal influence of the NEO “Big 5” personality domains on resilience to injury

While 13 of the traits associated with successfully completing CCT training were identified from the NEO personality inventory and there were 26 GWAS datasets identified related to those traits, only 2 neuroticism traits and 9 GWAS datasets were found to have causal effects on musculoskeletal outcomes (ukb-b-4630, ebi-a-GCST005232, ukb-a-230, ieu-a-1007, ebi-a-GCST005327, ebi-a-GCST003770, ieu-b-4848, ebi-a-GCST002920, ieu-a-118). There were 28 significant causal effects exerted by the “neuroticism score” and 30 causal effects exerted by “neuroticism” traits (Figure 2, Supplementary Table 2). While both “neuroticism score” and “neuroticism” traits originate from different studies and datasets (ebi-a-GCST005327 and ieu-a-1007, respectively), they both are continuous variables reporting the score of the neuroticism domain. The effect sizes, measured as an influence exerted by a change of one standard deviation in the exposure, of the neuroticism score on musculoskeletal injuries ranged from -6.27 (exerted on peroneal tendinitis; SE = 2.49; OR = 0.002 (1.4×10⁻⁵ - 0.248)) to 1.05 (exerted on ankle and foot dislocations, sprains, or strains; SE = 0.41; OR = 2.9 (1.3 - 6.3)). Similarly, the greatest negative effect size for neuroticism was also on peroneal tendinitis (beta = -21.27, SE = 9.66, OR = 5.79×10⁻¹⁰ (3.5×10⁻¹⁸ – 0.096)). The largest positive effect size for neuroticism was observed on dislocations, sprains and strains at the elbow (beta = 4.31, SE = 1.52, OR = 74.3 (3.8 – 1468). Overall, back pain appears to be the most frequently associated with a causal relationship to the neuroticism measures, appearing 15 times with neuroticism (50% of relationships) and 19 times with neuroticism score (68% of relationships).

• Download figure
• Open in new tab

Figure 2:

Odds ratios for “Big-5” domains with P < 5×10-8. A) Neuroticism score vs. MSI; B) Neuroticism vs. MSI; C) MR results with high odds ratios or wide confidence intervals for neuroticism (not indicated) or neuroticism score (indicated with bracket). All odds ratios are significant (P < 0.05) and the 95% confidence intervals do not overlap 1.0 (ratios provided as Supplementary Table 2).

We used the heterogeneity test provided within the 2SMR package to evaluate the degree of heterogeneity for each MR test. The package reported that there were enough variant instruments identified for 19 of the 58 MR tests (Supplemental Table 3). Using a four-group classification scheme based upon the I² values [31, 32], we found 2 tests exhibited severe heterogeneity (I² > 50%), 11 exhibited moderate heterogeneity (25% < I² < 50%), and 3 each had mild or no heterogeneity (I² < 25% or 0%, respectively). The severe heterogeneity appeared in the MR tests for neuroticism vs. lower back pain and forearm fractures calculated with the IVW estimator (I² = 71% and 60%, and P = 0.001 and 0.02, respectively). Those with mild or no heterogeneity with neuroticism as the exposure were elbow injuries and foot fractures and were both calculated with the IVW estimator (I² = 0% for both, P = 0.66 and 0.99, respectively). The neuroticism score exposure’s influence on lower back pain calculated using the Egger regression and on low back pain, foot fractures, and peroneal tendinitis calculated with the IVW estimator also had mild or no heterogeneity (I² = 24%, 20%, 9%, and 0%, respectively; P = 0.05, 0.09, 0.28, and 0.51, respectively). In addition to heterogeneity, we observed some small, non-zero directional pleiotropy in 3 neuroticism estimates and 4 neuroticism score estimates (Table 4).

Relaxing the threshold for extracting instrumental variables to 5×10⁻⁵, we found 1,696 distinct SNPs across 14 personality GWAS datasets to use as instrumental variables compared to 210 GWAS significant SNPs in 8 personality GWAS datasets, an 8-fold increase in the instrumental variables from an almost doubling of the available datasets. Performing MR with these variables resulted in identifying nearly 100 more potential causal effects (Supplementary Table 3), and 119 estimates resulted in odds ratios not overlapping 1 (Supplementary Figure 1). There were 37 significant causal effects exerted by the neuroticism score, 86 by neuroticism, and a single causal effect exerted by conscientiousness on elbow injuries (Supplementary Figure 1, Supplementary Table 3).

The causal influence of the aspects of emotional and social functioning on resilience to injury

From the 16 psychological traits associated with successfully completing CCT training that were obtained from the EQI and were unrelated to personality, we found 20 GWAS datasets identified related to those traits and 12 with causal effects on musculoskeletal outcomes (Figure 3, Supplementary Table 4). The largest positive direction effect size was for a diagnosis of anxiety on ankle or foot dislocations, sprains, and strains (beta = 6.0; SE = 2.6; OR = 410.2 (95% CI = 2.5 – 68371)) while the greatest negative direction effect size was for a visit to a general practice physician for anxiety on general muscle strains (beta = -16.8; SE = 8.1; OR = 5×10⁻⁸ (95% CI = 6.82×10⁼¹⁵ – 0.37). As with the personality traits, back pain appears to be most frequently related as an outcome to psychological traits, occurring in 4 of the relationships.

• Download figure
• Open in new tab

Figure 3:

Odds ratios for instruments associated with non-“Big 5” personality domains extracted at P < 5×10⁻⁸. The causal trait is listed alongside brackets enclosing the affected musculoskeletal injuries. All odds ratios are significant (P < 0.05) and the 95% confidence intervals do not overlap 1.0 (ratios provided as Supplementary Table 4).

In our heterogeneity tests for the psychological traits that were not components of the “Big 5” personality domains, we observed enough variant instruments identified for 9 of the 17 MR tests (Supplemental Table 5). Using the same four-group I² classification scheme as above [31, 32], we found no tests exhibited severe heterogeneity, 2 exhibited moderate heterogeneity, 1 had mild heterogeneity and the remaining 6 displayed no heterogeneity. The greatest degree of homogeneity was found for the effect of seeing a psychiatrist for nerves, anxiety, tension, or depression upon arm fractures (Q = 4.92, I² = 39%, P = 0.18). With respect to pleiotropy, we only found a single relationship with non-zero pleiotropy: a history of having seen a general practice physician for nerves, anxiety, tension, or depression on lower leg or ankle fractures (Egger intercept = -0.0008, 95% CI = -0.0001 to -0.0015, P = 0.0357).

As with the personality-related assessments, when we relaxed the threshold for extracting instrumental variables to 5×10⁻⁵, we found an increase in the number of SNPs and GWAS datasets involved: 3,645 low threshold SNPs vs. 75 GWAS significant SNPs in 29 vs. 15 datasets, respectively. As expected, performing MR with the lower-threshold variables resulted in identifying more potential causal effects (2,344 vs. 375), and 98 estimates (vs. 58) with odds ratios not overlapping 1 (Supplementary Figure 2, Supplementary Table 6). Again, similar to the personality estimates, we found back pain to be the most frequently represented outcome with a causal relationship from psychological traits, appearing in 33 of the estimates with significant odds ratios.