|Year : 2019 | Volume
| Issue : 2 | Page : 105-109
Risk factors of psychiatric hospitalization of military service persons in Taiwan: Preliminary results from unsupervised clustering techniques
Geng-Fu Tsai E.M.B.A., A.A., 1, Yueh-Ming Tai M.D., Ph.D., 2, Sy-Ming Guu Ph.D., 3
1 Department of Psychiatry, Tri-Service General Hospital, National Defense Medical Center, Beitou Branch; College of Management, Chang Gung University, Taipei, Taiwan
2 Department of Psychiatry, Tri-Service General Hospital, National Defense Medical Center, Beitou Branch; Department of Psychiatry, National Taiwan University Hospital and College of Medicine, Taipei, Taiwan
3 College of Management, Chang Gung University, Taipei, Taiwan
|Date of Submission||25-Dec-2018|
|Date of Decision||16-Mar-2019|
|Date of Acceptance||19-Mar-2019|
|Date of Web Publication||28-Jun-2019|
No. 60, Shin-Ming Road, Taipei 112
Source of Support: None, Conflict of Interest: None
Background: To understand the discrepancies of the use of mental health providers among different military ranks and/or compulsory/voluntary military services especially of the psychiatry admission during active services. Methods: We collected military medical records of one military psychiatry teaching hospital of north Taiwan from 2012 to 2015. All 3,513 samples were divided into three groups ─ enlisted females (EFs), enlisted males, and drafted males (DMs). The outcome measurement was the time period from the date of enlisted or drafted to the first psychiatric admission (E-A period). After comparing baseline characteristics and E-A period among three groups, we applied unsupervised clustering techniques, exhaustive Chi-squared automatic interaction detector, to cluster samples based on their military ranks and compulsory/voluntary service. Results: In general, the EF group showed the longest E-A period and the DM group the shortest. The most common diagnosis was major depression followed by anxiety or other nonpsychiatric disorders. The privates and recruits showed shorter E-A periods, and the younger enlistment age of officers showed the longer E-A period if we clustered based on military ranks. Those who entered army due to obligation showed shorter E-A period and those males who enlisted voluntarily at age over 22.5 years also showed shorter E-A period. Conclusion: This study demonstrates potential clusters associating with psychiatry admission in military. But, we caution that the findings here should be treated as preliminary.
Keywords: compulsory service, military mental health, psychiatric admission, voluntary service
|How to cite this article:|
Tsai GF, Tai YM, Guu SM. Risk factors of psychiatric hospitalization of military service persons in Taiwan: Preliminary results from unsupervised clustering techniques. Taiwan J Psychiatry 2019;33:105-9
|How to cite this URL:|
Tsai GF, Tai YM, Guu SM. Risk factors of psychiatric hospitalization of military service persons in Taiwan: Preliminary results from unsupervised clustering techniques. Taiwan J Psychiatry [serial online] 2019 [cited 2020 Sep 30];33:105-9. Available from: http://www.e-tjp.org/text.asp?2019/33/2/105/261748
| Introduction|| |
When a military serviceman/servicewoman has a psychiatric hospitalization, it is quite different from a general civilian has a similar one. He/She has not only to face the consequence of a possible military discharge but also to develop a self-stigma (or internalized stigma) as being perceived as weak and shameful . One 2002 study from the United States on active duty marines and soldiers who met the criteria for one or more mental illnesses described that about half of them report remarkable guilty feeling if they sought help . A military study from the United Kingdom on about 3,000 personnel indicated that stigmatization against those who have received care for mental illnesses, impacts their seeking help from mental health providers . Studies on Taiwanese military mental health have shown the different prevalences of psychiatric diagnoses between soldiers and civilians .
Sufficient evidence is still lacking about what kind of factors impact the military hospitalization due to mental illnesses. Therefore, we did data-based unsupervised clustering analyses to address this issue. With exhaustive Chi-squared automatic interaction detector (CHAID) decision tree analysis, we intended to classify military psychiatric inpatients based on their military service time periods from enlistment to the first time of psychiatric admission (E-A period).
| Methods|| |
This is a retrospective chart review study. The experimental protocol was approved by the institutional review board at the Tri-Service General Hospital, National Defense Medical Center in Taipei, Taiwan without the need of obtaining informed consents from the study participants.
The sample comprised totally 3,513 military psychiatric inpatients in one military psychiatry teaching hospital in north Taiwan from year 2012 to 2015. Those inpatients had three main groups: (a) the all-male privates and recruits for compulsory military services (drafted males, [DMs] group, n = 2,004); (b) the male service persons for voluntary military service (enlisted males, group, n = 1,409); and (c) the female voluntary service women (enlisted females, [EFs] group, n = 100).
Enlistment- first admission time period
We defined the “enlistment- first admission time period (E-A period)” as the time period from the date of samples' enlistment to the date of their first psychiatry admission in the registry. For those who used to be hospitalized twice or more in this hospital, we used only the first-admission sociodemographic data, military ranks, diagnoses, and the date.
Exhaustive Chi-squared automatic interaction detector analysis
Differing from supervised classifications, unsupervised clustering method is a kind of data-based modeling to discover outcome measurement, like E-A period in this study . This method is linked to the potential factors, especially when nonlinear relationships between them are encountered . The exhaustive CHAID algorithm, a kind of unsupervised clustering, builds a decision tree by means of repeated partitions of each subset into two or more child nodes, beginning with the full database . This method has been widely used in many medical fields to predict pulmonary embolism , to stage a cancer , to help a clinical decision , and to promote suicide prevention .
The descriptive analyses were used to present demographic characteristics of three groups. Then, we compared parameters between groups with analysis of variance for continuous variables, and with Chi-square test for categorical variables. To determine the potential risk factors of outcome variable (E-A period), we applied the Poisson linear regression model, and the further exhaustive CHAID analysis with 10-fold cross-validation. We computed all study variables with Statistical Package for Social Science software version 22 (SPSS Inc., Chicago, Illinois, USA). The differences between groups were considered significant if p- values were smaller than 0.05.
| Results|| |
[Table 1] lists characteristics of our psychiatric hospitalization active-service military personnel from 2012 to 2015. [Table 2] represents their military characteristics, namely ranks, being deployed or not, types of military service, and psychiatry diagnoses. [Table 3] shows the result of the Poisson regression models to predict E-A periods of our samples.
|Table 1: Characteristics of psychiatry admission active-service military personnel from 2012 to 2015|
Click here to view
|Table 2: The military characteristics and psychiatry diagnoses of psychiatry admission active-service personnel†|
Click here to view
|Table 3: The result of Poisson linear regression model with outcome measurement as months of military service before samples' first psychiatric admission|
Click here to view
Using the exhaustive CHAID algorithm, [Figure 1] shows the result based on military ranks and [Figure 2] is based on samples' voluntary or compulsory services [Table 1], [Table 2], [Table 3] and [Figure 1] and [Figure 2].
|Figure 1: The result of the unsupervised classification (exhaustive CHAID decision tree) based on military ranks for samples' mean period of time (month ± standard error) from enlistment to the first psychiatry admission. ms, months; yrs, years.|
Click here to view
|Figure 2: The result of the unsupervised classification (exhaustive CHAID decision tree) based on voluntary/compulsory service for samples' mean period of time (month ± standard error) from enlistment to the first psychiatric admission. ms, months.|
Click here to view
| Discussion|| |
Our study demonstrated the results of applying unsupervised decision tree algorithm, exhaustive CHAID, to categorize military psychiatry patients in predicting how long in average their E-A periods will be. In our Taiwanese service men/women, the DMs who entered the army due to obligation had shortest average E-A period (about three months, [Table 1]). Otherwise, the EFs who entered army voluntarily showed the eldest mean age at enlistment (about 22 years old) and the longest E-A period [Table 1].
The majority (> 90%) of psychiatry diagnoses in service persons were major depression and nonpsychotic disorders [Table 2]. According to the current military mental health screen assessment regulation, the diagnosis, or category, of neurosis consists of anxiety disorders and some other nonpsychotic mental illnesses, for example, to include obsessive-compulsive disorder, phobia, and dysthymia. This is partially in line with a previous survey of British military psychiatry hospitalization in 2007 . In that survey, the most common psychiatry problem for admission is depression, followed by alcohol-related problems . In our study, about one percent of service persons were diagnosed as schizophrenia or bipolar disorder. The diagnoses of intellectual disability and organic mental disorders were still represented in male group, especially among DMs. We once assumed that this results from the discrepancy in the mental health between the service persons of voluntary and compulsory military service. But, the results of our Poisson regression model [Table 3] revealed that either psychiatry diagnosis or type of military service (voluntary/compulsory) did not significantly associated with E-A period after excluding interactions of other variables. Instead, the military ranks and the age of enlistment (or draft) were significantly associated (p < 0.001).
The main finding of this article is that although [Table 3] showed the statistical correlation relationships between some risk factors and the outcome variable in the Poisson linear regression models, what we are interested is the data-based nonsupervised insights instead of the hypotheses-based results. Thus, we took some of significant risk factors in [Table 3] into the decision tree analysis (CHAID) and produced [Figure 1] and [Figure 2]. Our study shed light on how practically classify (or cluster) military active service persons by their military service types and genders in terms of the average time of period to be psychiatry hospitalization (E-A period). The results of unsupervised clustering analyses revealed that the deployed recruits showed the shortest E-A period (0.8 ± 0.19 months). The longest E-A period (114.3 ± 6.72 months) was observed among nondeployed officer who enlisted in army before the age of 19.5 years. The E-A period seemed generally lower among privates, and recruits can be due to their shorter service periods. Besides, an inverse proportional trend was found between enlistment age and E-A period among the groups other than privates or recruits [Figure 1]. If we performed the unsupervised classification based on the voluntary/compulsory military services of samples [Figure 2], compulsory-service soldiers and officers showed the shorter E-A periods than their counterparts. As the rule of thumb, deployed personnel showed the shorter E-A period than the nondeployed counterparts. Counterintuitively, the inverse proportional trend rule did not perfectively apply the group of voluntary-service privates. Especially, the voluntary-service privates who enlisted over 22.5 years old showed the E-A period as 5.5 ± 0.64 months after their enlistments that was apparently the shortest among all age groups.
The readers are cautioned not to overinterpret the study finding because this study has three limitations:
- The study did not have control group, or nonadmission personnel. The absence of this comparative group hinders us to conduct the survival analysis to clarify the causality.
- The previous British survey in 2007  also observed that more women are admitted with depression and more men with alcohol-related disorders. That observation was not seen to our samples due to limited sample size in the female group. But, insights provided in this study still can be generalized to some other good points which are warranting further investigations.
- An essential drawback of decision tree technique is only limited number of variables can be used in the final tree. If one arbitrarily compiles many variables on the final decision tree, it will be too complex to be understood.
According to the registry of one military psychiatry teaching hospital in Taiwan, those who entered army due to obligation showed the shortest time duration from enlistment to first psychiatric hospitalization, especially who were deployed. For the voluntary service men/women, the elder enlistment age and the status of being deployed is associated with shorter service time periods before psychiatric hospitalization than other groups.
We caution that the findings should be treated as preliminary results due to the nature of this exhaustive CHAID technique. We need some sort of validation for the study results before we recommend the use of the exhaustive CHAID technique.
| Acknowledgements|| |
The opinions expressed are authors' personal opinions. They are unnecessarily reflecting on those of their hospitals or institutions.
| Financial Support and Sponsorship|| |
| Conflicts of Interest|| |
There are no conflicts of interest.
| References|| |
Harris JI, Farchmin L, Stull L, et al.: Prediction of changes in self-stigma among veterans participating in partial psychiatric hospitalization: the role of disability status and military cohort. Psychiatr Rehabil J
2015; 38: 179-85.
Hoge CW, Lesikar SE, Guevara R, et al.: Mental disorders among U.S. military personnel in the 1990s: association with high levels of health care utilization and early military attrition. Am J Psychiatry
2002; 159: 1576-83.
Fertout M, Jones N, Keeling M, et al.: Mental health stigmatisation in deployed UK armed forces: a principal components analysis. J R Army Med Corps
2015; 161 (suppl 1): i69-76.
Tai YM, Yang SN: The military psychiatry in Taiwan. Taiwan J Psychiatry
2018; 32: 87-8.
Frades I, Matthiesen R: Overview on techniques in cluster analysis. Methods Mol Biol
2010; 593: 81-107.
Biggs D, De Ville B, Suen E: A method of choosing multiway partitions for classification and decision trees. J Appl Stat
1991; 18: 49-62.
Carmona-Bayonas A, Jiménez-Fonseca P, Font C, et al.: Predicting serious complications in patients with cancer and pulmonary embolism using decision tree modelling: the EPIPHANY index. Br J Cancer
2017; 116: 994-1001.
Venkatesan E, Velmurugan T: Performance analysis of decision tree algorithms for breast cancer classification. Indian J Sci Technol
2015; 8: 123-30.
Bae JM: The clinical decision analysis using decision tree. Epidemiol Health
2014; 36: e2014025.
Handley TE, Hiles SA, Inder KJ, et al.: Predictors of suicidal ideation in older people: a decision tree analysis. Am J Geriatr Psychiatry
2014; 22: 1325-35.
Finnegan A, Finnegan S, Gamble D: A review of one year of British armed forces mental health hospital admissions. J R Army Med Corps
2007; 153: 26-31.
[Figure 1], [Figure 2]
[Table 1], [Table 2], [Table 3]