Combat injury profiles among U.S. military personnel who survived serious wounds in Iraq and Afghanistan: A latent class analysis

Background The U.S. military conflicts in Iraq and Afghanistan had the most casualties since Vietnam with more than 53,000 wounded in action. Novel injury mechanisms, such as improvised explosive devices, and higher rates of survivability compared with previous wars led to a new pattern of combat injuries. The purpose of the present study was to use latent class analysis (LCA) to identify combat injury profiles among U.S. military personnel who survived serious wounds. Methods A total of 5,227 combat casualty events with an Injury Severity Score (ISS) of 9 or greater that occurred in Iraq and Afghanistan from December 2002 to July 2019 were identified from the Expeditionary Medical Encounter Database for analysis. The Barell Injury Diagnosis Matrix was used to classify injuries into binary variables by site and type of injury. LCA was employed to identify injury profiles that accounted for co-occurring injuries. Injury profiles were described and compared by demographic, operational, and injury-specific variables. Results Seven injury profiles were identified and defined as: (1) open wounds (18.8%), (2) Type 1 traumatic brain injury (TBI)/facial injuries (14.2%), (3) disseminated injuries (6.8%), (4) Type 2 TBI (15.4%), (5) lower extremity injuries (19.8%), (6) burns (7.4%), and (7) chest and/or abdominal injuries (17.7%). Profiles differed by service branch, combat location, year of injury, injury mechanism, combat posture at the time of injury, and ISS. Conclusion LCA identified seven distinct and interpretable injury profiles among U.S. military personnel who survived serious combat injuries in Iraq or Afghanistan. These findings may be of interest to military medical planners as resource needs are evaluated and projected for future conflicts, and medical professionals involved in the rehabilitation of wounded service members.

Introduction Diego, California, that includes clinical records of U.S. service members injured during combat deployment. Clinical records were completed by providers in-theater and provided to NHRC where they were consolidated with patients' medical records obtained from all levels of care. Patient records were retrospectively reviewed by certified nurse coders at NHRC and assigned International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes [28] and ISSs. Additional information on the EMED is available elsewhere [29]. The ISS is calculated as the sum of the squares of the highest Abbreviated Injury Scale [30] severity score in each of the three most severely injured body regions and quantifies overall injury severity for each casualty [23][24][25]. Only those with serious or greater injury severity (i.e., ISS � 9) who survived their wounds through all levels of care were included. The study population included 5,227 casualty events that occurred during combat operations in Iraq and Afghanistan from December 2002 to July 2019. This study complied with all federal regulations governing the protection of human subjects in research and was approved by the Institutional Review Board (IRB) at NHRC. The approved IRB protocol (NHRC.2003.0025) issued a waiver of informed consent for this study.

Variables
The Barell Injury Diagnosis Matrix, a two-dimensional table which categorizes injuries by body region (or site) and nature (or type) of injury, was used to classify injuries for each casualty [31]. In the Matrix, TBI is categorized as: Type 1 TBI (i.e., moderate-to-severe brain injury as indicated by an extended loss of consciousness and/or amnesia of the injury event); Type 2 TBI (i.e., mild brain injury as indicated by brief loss of consciousness or altered mental status); and Type 3 TBI (i.e., skull fracture without specification of intracranial injury). The Matrix has 36 rows that represent body regions and 12 columns that represent injury types. In this study, the "fractures" injury type was expanded to include "open" and "closed" fractures, which resulted in an additional column. Binary (1 or 0) injury variables were coded to indicate the presence or absence of the specific body region/injury type combination in each cell of the Matrix, and only populated cells in the Matrix were examined. Overall, 181 binary injury variables were derived for each combat casualty.

Statistical analysis
Analyses were performed using SAS software, version 9.4 (SAS Institute, Cary, North Carolina) and R software, version 3.6.2 (The R Project for Statistical Computing). The R package poLCA [32] was used for LCA, a probability model-based clustering algorithm [33,34]. LCA was used to map the 181 binary injury manifest variables [34] onto classes termed "injury profiles." Injury profiles represented mutually exclusive groups of combat casualties with commonly co-occurring injuries. Several LCA models were built, with the number of latent classes ranging from 1 to 10. The best model for each group was chosen using a combination of qualitative and quantitative measures, preferring models with more coherent classes, fewer parameters, and better fit statistics [34]. Binary injury manifest variables with a conditional item probability (i.e., class-specific indicator probability) of at least 0.30 were used to identify and label the classes in the LCA models. Interpretation of injury profiles was based on LCA results and input from subject matter experts. Fit statistics, including the Bayesian information criterion (BIC), sample-size adjusted BIC (SABIC), Akaike information criterion (AIC), and consistent AIC (CAIC), were computed [35]. Casualties were assigned to each class in the LCA model using the maximum-probability assignment rule. To evaluate the likelihood of misclassification, mean classification posterior probabilities were estimated for each class. Values above 0.70 indicated well-separated classes in the model [36], and entropy values greater than 0.80 indicated "good" model classification of individual cases into classes [37]. The selected LCA model yielded latent classes (or injury profiles) that were described by injuries with conditional item probabilities above the 0.30 threshold in each group. Chi-square tests assessed the distribution of demographic, operational, and injury-specific variables across injury profiles. Multiple hypothesis tests were conducted to compare the proportions of the levels of each categorical variable over all possible pair combinations across injury profiles. P-values were adjusted using the Holm method to control the family-wise error rate for multiple comparisons. An alpha level of 0.05 was used for statistical significance. Table 1 summarizes fit statistics of the LCA models with classes ranging from 1 to 10. The BIC, SABIC, AIC, and CAIC statistics indicated the ideal model had between 6 and 10 classes. The 7-class LCA model was selected as the best model based on a combination of good fit (smallest CAIC statistic), model parsimony (fewer model parameters), and coherent, interpretable classes. The 7-class LCA model had an entropy statistic of 0.857 and outperformed most of the models in delineating the classes. Assignment of cases to unique classes using the maximumprobability assignment rule yielded mean class membership posterior probabilities above 0.90 for all seven classes ( Table 2). Posterior probabilities of membership among classes where cases were not assigned did not exceed 0.05, indicating a very low expected misclassification rate. Table 3 shows the injury types and sites with conditional item probabilities above the 0.30 threshold for each of the seven classes in the model. Injury profiles were defined as: open wounds (class 1); Type 1 TBI/facial injuries (class 2); disseminated injuries (class 3); Type 2  Characteristics of the study population by injury profile are shown in Table 4. The study population was predominantly male (98.3%), aged 18-24 years (55.9%), and in the Army (70.6%). The majority were injured while deployed in Iraq (50.2%) between 2002-2008 (53.3%). Most casualties were injured by blasts (75.9%) and sustained serious injuries (ISS 9-15; 59.3%). A slightly higher proportion of the study population was mounted than dismounted at the time of injury (42.2% vs. 39.5%). All variables except for sex and age differed significantly across the injury profiles (ps < 0.001). The burns profile (class 6) had the highest percentage of Army service members (77.

Discussion
The U.S. military conflicts in Iraq and Afghanistan resulted in a new pattern of injuries among combat casualties. To our knowledge, the present study is the first to describe injury profiles among combat casualties using LCA. Seven injury profiles were identified and described by demographic, operational, and injury-specific data, which reflected different periods of the OIF/OEF conflicts and highlighted ubiquitous injury types, such as TBI and lower extremity injuries [11,19]. The findings may be of interest to military medical planners who project the logistics, resources, and skilled providers required to treat combat casualties with serious injuries in future conflicts, and to medical professionals involved in injury rehabilitation, as many military personnel with combat injuries may require life-long care [27].
One of the profiles identified in the present study indicated a wide range of open wounds marked by a high proportion of service members dismounted at the time of injury who were primarily injured by blasts. This profile appears to be similar to "dismounted complex blast injury," which is characterized in the literature by extensive open wounds, including amputations and pelvic/urogenital injuries [21,22]. The circumstances surrounding dismounted complex blast injury typically involve military personnel on foot patrol when an explosive device is activated nearby [22]. Survivors of these type of injuries face quality of life concerns due to resulting disabilities, and optimal rehabilitation strategies are necessary [38]. Pelvic protection has been developed for U.S. military personnel and future research is needed to determine its utility in a combat environment [39].
In contrast to the open wounds profile, the group with disseminated injuries, including injuries to internal organs and fractures (both open and closed), had the highest proportion of service members mounted in a vehicle at the time of injury. Most service members in this profile were also injured by blasts. A unique aspect of this profile was fractures to the vertebral column, which has been identified in previous research on mounted casualties [40,41]. Though enclosure within a vehicle affords some protection compared with dismounted personnel, certain characteristics of the injury incident can increase risk for serious injury, such as vehicle rollover, and high velocity displacement [41]. A key variable missing from the present analysis was the type of vehicle mounted at the time of injury, which can impact injury patterns [40]. Over the course of OIF/OEF, vehicles have been improved to maximize operational effectiveness and increase the amount of protective armor. Humvees were widely used during the early phases of these conflicts, but were later phased out in favor of Mine-Resistant Ambush-Protected vehicles [42]. In addition, position in the vehicle can affect injury patterns, as a previous study found that gunners (i.e., operators of the weapon on top of the vehicle) had a higher percentage of extremity wounds compared with drivers and passengers [41]. As such, a more detailed analysis of mounted injuries is required to further define risk factors and develop potential preventive strategies.
The identification of TBI-related profiles was not surprising given that TBI emerged as one of the signature wounds of OIF/OEF, with an estimated 1 in 5 service members with mild TBI [10,11]. One TBI profile consisted primarily of Type 2 TBI, whereas the other predominately involved Type 1 TBI, which generally results in worse long-term outcomes than mild Type 2 TBI [43]. Of note in the present study, both TBI profiles occurred in the presence of other injuries (e.g., wounds to the face). A prior descriptive account of combat-related TBI found that a significant proportion of service members sustain concomitant injuries [10], and these other injuries can slow the course of TBI recovery [44]. Future military TBI research should address co-occurring injuries, potentially by using injury severity specific to the non-head region, such as the extracranial ISS used by Stulemeijer et al. [44]. Furthermore, efforts should continue to identify innovative methods for monitoring and mitigating TBI on the battlefield, including sensor technology and improvements in helmet design [10,45].
There were other notable findings of interest. The chest and/or abdominal injury profile was the only profile where the proportion of service members with gunshot wounds significantly outnumbered those injured by blasts. In addition, this profile was isolated to injuries to the internal organs in the mid-section, with no other injuries meeting the probability threshold. Current personal protective equipment may offer protection, but certain variables not accounted for in this study may impact its effectiveness, including overall fit and bullet/shrapnel trajectory. Another profile was predominated by burns. Most service members in this group were injured between 2002-2008 and mounted at the time of injury, which could reflect the vehicle types used earlier in the conflicts as described previously. Finally, the lower extremity injury profile was not surprising, as these injuries frequently occurred during OIF/OEF [19]. This profile also had the lowest overall injury severity, which may be indicative of lowenergy blast injuries, such as when an individual is a significant distance from a blast event, or the improvised explosive device is of a lower explosive weight [22].
The present study had several strengths. The EMED allowed for abstraction of medical and tactical information (e.g., injury mechanism, combat posture) from the point of injury, which is generally difficult to obtain in austere combat environments. Further, the Barell Injury Diagnosis Matrix is a standard injury classification method endorsed by the Centers for Disease Control and Prevention [46], and all casualty records were reviewed and validated by professional nurse coders to ensure accuracy. There are also limitations that warrant mention. The injury profiles from LCA were probability-based in contrast to other potentially more precise methodologies such as three-dimensional surface wound mapping [47]. The conditional item probability threshold of 0.30 was a subjective criterion and injuries not meeting this threshold could have contributed to the injury profile within each class. It is also important to note that the combat posture variable (i.e., mounted/dismounted status) had a large proportion of missing data. Further research of incident-related factors is needed and may require collaboration with other U.S. government agencies to obtain sensitive data (e.g., amount of explosive, distance from blast). Additional studies are warranted to explore injury profiles among casualties with minor injuries and those who died of wounds or were killed in action, as the focus of the present study was service members who survived serious injuries and findings may not generalize to these other groups.

Conclusion
The present study used LCA to classify combat injury patterns among U.S. service members who survived serious wounds from OIF/OEF. Some of the injury profiles aligned with previous research that has identified dismounted complex blast injury, as well as preponderance of TBIs and lower extremity trauma during OIF/OEF. Combat posture at the time of injury was independently associated with various injury profiles, including the open wounds and disseminated injury groups, which requires further examination as these complex injury profiles impact long-term health outcomes. Additional research may be beneficial to identify injuryrelated sequelae and outlook for recovery. As modern warfare evolves and the U.S. military prepares for the next conflict, the identification and evaluation of combat injury patterns is paramount to medical planning and resource projections, and rehabilitation of wounded service members.