In South Korea, all males aged 18 years or older are obligated to perform military service for a certain period. New recruits should be able to withstand the psychological stress and pressures of rapid adaptation to the unique and new environment of the military [1,2]. The military authority is making efforts to continuously evaluate and manage the candidates who are ineligible for military service in each stage (i.e., conscription, training, and active) [3]. Moreover, the recruits with psychiatric issues who need more-precise clinical evaluations are referred to a hospital.

The Minnesota Multiphasic Personality Inventory-2 (MMPI-2) has been used to identify and predict military maladjustment, with a long history in military settings [4]. However, few studies have used this tool to investigate the psychological characteristics of recruits with a risk of maladjustment to military service. A longitudinal study in China demonstrated that MMPI-2 is a valid tool for predicting adaptation to military service by recruits [5]. That study found that higher MMPI-2 scores worsened the early adaptation of recruits. Bases on MMPI-2 score, recruits deemed mentally healthy would have fewer problems during the subsequent training process compared with those who had worse mental health.

Another study using the MMPI-2 Restructuring Form found that patients with psychiatric issues attempting to obtain a medical certificate for military service have somewhat different psychological characteristics to the control psychiatric patient group without military issues; that is, emotional distress, helplessness, lower self-confidence, interpersonal discomfort, and avoidance were more evident and severe in these than in the patients in the control group [6].

Furthermore, clinicians often find empirically that patients have worse therapeutic responsiveness and prognoses even though they complain of similar or identical symptoms. This suggests that there are heterogeneous and discriminatory features within those patients in addition to their common psychological characteristics. Classifying a group of recruits or soldiers into different subtypes may lead to different treatment plans and prognoses for each type, and make it possible to manage maladjusted soldiers more systematically. A recent cross-cutting study by Lim et al. [7] identified four potential subgroups with distinct clinical and psychological characteristics: Class 1 (noncli-nical), Class 2 (internalized), Class 3 (externalized), and Class 4 (confused). These latent subgroups were expected to have substantial differences in clinical prognosis and progress, but no longitudinal studies have previously been performed [7].

The purposes of this study were as follows: First, it was conducted to investigate the potential subtypes of recruits with mental health problems to obtain a psychiatric certificate for military service suitability based on clinical symptoms identified in the MMPI-2 using a person-centered approach such as latent profile analysis (LPA). Second, we aimed to determine whether there is a significant difference in treatment responsiveness and prognosis among those latent subgroups.

Participants

The subjects were male patients aged 18−28 years who visited the department of psychiatry at a university hospital for a mental health evaluation related to military service. This study included inpatients and outpatients who met the following criteria: (1) referred by the Military Manpower Administration due to potential unsuitability for military service, (2) referred for a psychiatric evaluation due to complaints of psychiatric symptoms during military recruit training or service, (3) scheduled to be examined for military service, or (4) had received psychiatric treatment for at least 6 months. The exclusion criteria were as follows: (1) a history of traumatic brain injury, neurological disorder, intellectual or developmental disorder, bipolar spectrum disorder, schizophrenia spectrum disorder, or substance use disorders excluding alcohol use disorder, or (2) fetal clinical or medical conditions that would hinder participation in the study until the endpoint.

Patients were diagnosed according to the diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders 5th edition by experienced psychiatrists. They underwent a comprehensive psychological evaluation by trainee psychologists under the supervision of a qualified clinical psychologist, and the MMPI-2 data from the evaluation is used for the study. Only clinically valid MMPI-2 results were included in the statistical analysis (Butcher et al. [4]).

The final analysis included 92 eligible patients: 52 with depressive disorder, 29 with anxiety disorder, and 11 had other psychiatric disorders. There were 76 patients (82.6%) with at least one or more comorbid disorders in addition to the main diagnosis, and the most common comorbidities were depressive and anxiety disorders (63.2%). Table 1 lists the demographic and clinical characteristics of the subgroups at 0 months. At the start of the study, none of the demographic variables other than age differed significantly between the three groups. There was no significant difference between the three groups in clinical variables, but there were suicide attempts in Classes 2 and 3 but not in Class 1. The clinical response, function, and medication of subjects were assessed at 0, 3, and 6 months.

This study was conducted using a retrospective medical records analysis from January 2016 to May 2021 and was approved by the Institutional Research Ethics Review Committee of the hospital (reg. no. 2019-04-034).

Measures

The MMPI-2 is a tool that has been the most widely used to assess personality traits and psychopathology in adults [4]. The present study used the 10 clinical scales from the Korean version of the MMPI-2 in the LPA [8]. Scores higher than T65 on the MMPI-2 suggest clinically significant psychopathology. This test was only performed at baseline (0 month).

The Clinical Global Impression (CGI) is one of the most widely used evaluation tools in mental health medicine, and provides a brief, global assessment of the functioning of a patient prior to and after initiating psychotropic medication [9]. The CGI-Severity (CGI-S) scores current symptom severity on a seven-point scale, with 1 point for the lowest symptom severity and 7 points for the highest clinical severity [9]. We assessed data at three time points (at 0, 3, and 6 months).

The Global Assessment of Functioning (GAF) scale stand-ardized in South Korea was used in this study [10]. This tool is utilized by clinicians to assess the extent to which symptoms affect the day-to-day life of a person on a scale from 1 (severely impaired) to 100 (extremely high functioning), while 0 for inappropriate information. This scale was also administered three times over a 6-month follow-up.

We conducted a LPA using ten clinical scales of the MMPI-2 as indicator variables to investigate the subgroups of subjects. LPA is a person-centered analysis method that classifies potential subgroups according to individual symptom severity patterns [11].

Based on previous studies, the following three criteria were applied to compare the models by increasing the number of latent classes one by one: (1) Akaike informa-tion criterion (AIC) [12], Bayesian information criterion (BIC) [13], and sample-size-adjusted Bayesian information criterion (SABIC) fit indices [14], with smaller values indicating a better model [15]; (2) significance of the bootstrapped likelihood ratio test [16], which is a significance test that evaluates whether k–1-latent-layer models are rejected to support k-latent-layer models, and a k-latent- layer model is selected if the pvalue is significant in the test; and (3) quality of entropy classification, which ranges between 0 and 1 to indicate the accuracy of the case allocated to the extracted latent profile, and a value higher than 0.8 indicates a good classification [17]. These three criteria were combined to determine the number of latent classes. The final model was then determined by considering the theoretically consistent degree of the number of latent classes and the possibility of interpretation [18,19]. Clinical features were compared among latent groups using analysis of variance. All post-hoc comparisons were conducted using the least-significant-difference test.

Linear mixed model (LMM) analysis was conducted to examine changes in the severity of clinical symptoms and functional level according to the treatment period of the latent classes derived from the LPA. CGI-S and GAF scores were outcome variables, while time, class, and interaction were entered as fixed effects, and the subjects were entered as random effects. Lastly, a chi-square (χ²) test was performed to identify differences in the number of medications prescribed among the three subgroups. R statistical software was used for the LPA, and SPSS (version 27; IBM Co.) was used to conduct other data analyses.

LPA to Determine the Numbers of Latent Subgroups

To determine the most-appropriate number of subgroups to classify the participants using the ten clinical scales of MMPI-2, we increased the number of classes by one from the one-class model and choose the number of classes based upon the model that provided the best fit to the observed data. We used a combination of three methods to decide on the number of classes: (1) informa-tion-theoretic methods (AIC, BIC, and SABIC), (2) likelihood ratio (LR) statistical tests (BLRT), and (3) entropy- based criterion. The results are listed in Table 2.

Regarding the fit of each model, it can be seen that all fit indices including the AIC, BIC, and SABIC decreased from the one- to four-class model. For BLRT, which evaluates the relative adequacy of a (k–1)-class model compared with a k-class model, all results were significant in the two-, three-, and four-class models (all p < 0.001). Entropy, which evaluates the quality of classifications, was observed to be close to 1. When combining the above three criteria, both three-class and four-class models met the criteria in that the fit indices were small, the LR tests were significant, and the entropy indices were large.

There was no problem with the ratio of the number of cases because the minimum proportion in both models exceeded 5%. Additionally, to examine theoretical interpretability, the plots of the three-class and four-class models were compared. As a result, in the four-class model, it was found that the two subclasses were differentiated according to the overall severity. It was therefore difficult to determine whether the heterogeneous clinical aspects of the patients were sufficiently reflected. If the patients were classified only by symptom severity, it was judged that the latent class classification had no clinical meaning. In addition to the statistical criteria, the model with three latent classes was finally determined to be the most suitable by considering both the ratio of the number of cases per group and the interpretive meaning [18,19].

Clinical Features for Each Latent Class of the Final Three-class Model

Each latent class of the final three-class model was indicated to have clinically distinctive features. Figure 1 presents the clinical profiles of the three classes based on the MMPI-2. Each class was named as follows according to its clinical features: Class 1 (mild maladjustment; n = 14, 15.2%), Class 2 (neurotic depression and anxiety; n = 36, 39.1%), and Class 3 (highly vulnerable and hypervigilant; n = 42, 45.7%). Subjects in Class 1 presented relatively minor or nonclinical psychiatric symptoms, which acquired low scores of around 50 on all of the MMPI-2 clinical scales. Subjects in Class 2 presented a clinically high score distribution for internalizing problems such as depression, anxiety, helplessness, social discomfort, and avoidance, showing scores higher than 65T on the 2, 7, and 0 scales. Subjects in Class 3 presented internalizing problems more severe than Class 2 and also tended to experience paranoia, anger, hostility, confused thought process, and social alienation according to not only higher scores than for Class 2 on the 2, 7, and 0 scales, but also T scores of 70 or higher on the 6 and 8 scales. There were significant differences in the clinical scale scores among subclasses (Wilks’ L = 0.096, F[2,89] = 12.071, p < 0.001, partial η² = 0.598). In the post-hoc analysis, most of the clinical scales showed higher scores in the order of Class 3, Class 2, and Class 1.

Changes in Clinical Symptoms and Functional Levels between Latent Subclasses

Table 3 lists descriptive statistics on clinical symptom severity and functional levels across time for the treatment among the classes. Subjects in Class 3 had more severe clinical symptoms and worse functional levels than those in Class 1 and Class 2 at all three treatment time points. We also conducted LMM analysis to investigate the longitudinal changes in clinical symptom severity and functional level according to the treatment period for each latent subgroup. The results of the LMM analysis are listed in Table 4. The main effects of group (F = 21.62, p < 0.001) and time (F = 98.40, p < 0.001) and the group- by-time interaction effect (F = 5.62, p < 0.001) were significant on clinical symptom severity according to CGI-S scores. The main effects of group (F = 15.67, p < 0.001) and time (F = 105.81, p < 0.001) and the interaction effect between group and time (F = 5.38, p < 0.001) were significant at the functional level according to GAF scores.

The CGI-S scores were significantly higher in Class 3 than in Class 1, significantly lower at follow-up visits compared with at 0 months (all p < 0.0001), and there was a significant interaction between the group and visits at 3 months (p < 0.05), with higher CGI-S scores in Class 3 than in Class 1. Meanwhile, the GAF scores were significantly lower in Class 3 than in Class 1 (p < 0.01), significantly higher at follow-up visits compared with at 0 months (all p < 0.0001), and there was a significant interaction between the group and visit at 6 months (p < 0.05), with higher GAF scores in Class 2 than in Class 1. In an additional analysis, significant differences were also indicated among the three subgroups in the number of medications prescribed. Subjects in Class 3 received more antidepressants (χ² = 18.01 and p < 0.01 at 3 months, and χ² = 15.02 and p < 0.05 at 6 months) and anxiolytics (χ² = 12.26 and p < 0.05 at 3 months, and χ² = 10.46 and p < 0.05 at 6 months) than did those in Class 1 and Class 2 (Table 5). Antipsychotic agents and a combination of three antidepressants were prescribed more frequently in Class 3 than in Class 1 and Class 2. Taken together, subjects in Class 3 presented more comorbid and severe clinical symptoms, worse functional levels, and received more medications than did those in Class 1 and Class 2 at all three treatment time points.

This study attempted to identify distinct subgroups of patients who visited psychiatric clinics to obtain a medical certificate for military service suitability and further to determine whether there is a difference in treatment responsiveness and prognosis among these subgroups. The main results of this study are as follows: The three classes derived from the LPA showed distinguishable clinical features. The patients in Class 1 were those who did not complain of significant psychiatric symptoms in the clinic. The patients in Class 2 tended to complain of internalized problems such as depression, anxiety, helplessness, social discomfort, avoidance, and self-criticism. The patients in Class 3 tended to experience not only a higher level of various internalized problems but also a high level of paranoia, interpersonal hypersensitivity, anger, hostility, and strange beliefs or unusual perceptual experiences. The latent subgroups classified in this study were partially consistent with the result of Lim and his colleagues identifying the following four classes based on the profiles of MMPI-2 and Temperament and Character In-ventory: Class 1 (nonclinical), Class 2 (internalized), Class 3 (externalized), and Class 4 (confused). While Classes 1 and 2 identified in that previous study had very similar clinical features to Classes 1 and 2 in our study, the LPA results in the current study generated different numbers of classes with different characteristics. This may be due to differences in the sample characteristics. Unlike the sample in that previous study, we excluded patients with bipolar, schizophrenia spectrum, and neurodevelopmental disorders because the purpose of the current study was to ostensibly identify potential differences among subgroups that share similar clinical symptoms.

A subsequent analysis was conducted to investigate changes in clinical symptoms and functional levels across three treatment time points (at 0, 3, and 6 months) in each subgroup identified using the LPA. According to the results, Class 3 did not show remarkable improvement regarding both clinical symptoms and functional level mea-sured by CGI-S and GAF scores compared with Classes 1 and 2. Furthermore, Class 3 had more antidepressants and anxiolytics prescribed than Class 1 and Class 2. More use of adjunctive antipsychotics and a combination of three antidepressants in Class 3 indicated the illness severity, insufficient treatment response, and poor prognosis. In summary, Class 1 may comprise patients with relatively light and temporary adaptation problems with regard to a negative perception of and attitude toward military service [20], while those in Class 2 may have a high motivation for treatment due to significant subjective distress and help-seeking intentions [21]. Those patients in Class 1 and 2 may therefore show relatively good therapeutic prognoses. On the other hand, patients in Class 3 had the most-severe and varied psychiatric symptoms and had the worst prognosis. Thus, patients in this class may be more likely to be discharged early due to adaptation problems even if they are sent to military service. Previous research has suggested that recruits with mental diseases tended to malfunction during military training and duties [22,23]. In a similar context, previous studies have demonstrated that recruits with higher scores on the clinical scales of MMPI- 2 have more mental health problems and difficulties in both initial and subsequent adjustment to military service and tend to be separated from the military early [5,24].

The current study may have benefited from the application of a person-centered approach such as LPA. Such an approach can be very useful to identify heterogeneous subgroups within a sample that have similar scores on several variables of interest [17,25,26], which were the clinical features measured by MMPI-2 in the current study. This study had clinical significance in qualitatively classifying different subgroups within the sample by conducting LPA using the clinical profiles of MMPI-2 and providing a basis to comprehensively understand their differentiated clinical features.

Psychological tests have been steadily used to select soldiers since the early 20th century [27,28]. Those tests are crucial for not only diagnosing current psychological problems but also predicting maladaptation problems in the military environment that could be encountered in the future. Considering that MMPI-2 is one of the most widely used psychological tools in both South Korea and other countries to evaluate mental health problems in new military personnel or prospective recruits, our results potentially have major implications for the ability to predict military life maladjustment in soldiers due to mental health problems.

Most studies related to the psychological characteristics of maladjusted recruits have had cross-sectional designs, and this study was the first attempt to investigate treatment-based changes in clinical features and functional levels using a longitudinal approach. The results have major implications for the prediction of military maladjustment and suggest the need for treatment plans and management that consider different clinical features of latent subgroups within these samples.

This study had several limitations. The ability to generalize the results is limited by it being conducted based at a single medical center with a small sample size. It is therefore necessary to replicate our results through multicenter research and to repeatedly verify the results through future studies that use larger samples. Moreover, the study included the inherent methodological limitations of retrospective chart reviews such as missing or incomplete data in clinical or demographic variables, or the risk of selection sampling bias [29]. These limitations may affect the validity and reliability of the results, so future prospective studies may be important. This study classified the latent classes with only the clinical scales of the MMPI-2, so further studies should consider the various other scales of the MMPI-2 together.

Notwithstanding these limitations, this study had the following strengths. The study is expected to provide useful guidelines for decision-making on military service suitability as well as treatment direction and prognosis predictions by identifying potential subgroups of heterogeneous and discriminatory characteristics within these groups and longitudinally tracking them. Previous studies have largely overlooked the topic of recruit adaptation, despite its importance due to the various side effects such as suicidal ideation caused by maladaptation to military life. It will be necessary to continuously conduct multidimensional studies to predict military maladjustment in recruits and to verify the effectiveness of various interventions to systematically manage those problems.

To conclude, the current study identified three latent classes of military recruits with mental health issues based on MMPI-2 clinical profiles using LPA and further found a difference in treatment responsiveness and prognosis among these subgroups. This suggests that there are heterogeneous and discriminatory features within those patients in addition to their common psychological cha-racteristics. Therefore, careful assessment and treatment planning should be conducted to address these differences.

This research was supported by Alvogen Korea. The funding sources were not involved in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.

No potential conflict of interest relevant to this article was reported.

Conceptualization, data acquisition, formal analysis, & writing-original drift: Eun-Hee Park. Conceptualization, data acquisition, writing-review & editing, & supervision: Duk-In Jon. Conceptualization and data acquisition: Hyun Ju Hong, Myung Hun Jung, Narei Hong.