Parental military deployment as risk factor for children’s mental health: a meta-analytical review

A body of relevant publications was compiled through a systematic search of the electronic database system of the University of Ulm, which includes 5083 databases such as PubMed, EBSCOhost, Web of Science, and PsycARTICLES. The keywords used were (milit* families OR soldier OR army OR veteran OR deployment) AND (child* OR adolescen* OR family) AND (mental health OR mental illness OR mental disorder OR psychiatric illness OR psychiatric disorder). Moreover, eight websites referring to military projects were included [22‐29] to identify studies outside the academic publishing. If applicable, relevant publications that were not captured by the keywords but were cited in a retrieved article were manually searched as well.

Three researchers took part in the search. One, designated the independent reviewer, checked the abstracts of all the identified articles and discarded the vast majority as clearly irrelevant, including non-empirical studies, dissertations, and studies that did not involve children or did not include at least one parent in military service. The other two researchers then reviewed the full texts of the articles that remained for relevance. In cases of disagreement, the independent reviewer acted as a mediator. Discrepancies were resolved through discussion until consensus was reached by at least two of the three reviewers. The articles deemed to be relevant were then further assessed according to the criteria below.

Articles included in the meta-analysis were restricted to those that reported on families of military service members in the United States, had been published between 2001 and 2017, and involved quantitative measures that were concerned with the relationship between deployment of military parents and the presence of mental health problems in their children. The focus was on instruments that assessed symptoms of anxiety/depression, aggressive behavior, and hyperactivity/inattention. Studies that were concerned with child maltreatment, somatic outcomes (e.g., headache), school/academic variables, coping strategies, attachment, family cohesion, parenting, or familial communication were excluded.

The studies selected for inclusion in the meta-analysis were chosen to compare children of deployed military parents to one of two control conditions: children of civilian parents and children of non-deployed military parents. In the first comparison, deployed military parents included personnel of any branch of the armed forces, both active (full-time occupation in military service) and post-combat (recently returned war veterans), but excluded reserve component personnel. If available, data obtained during pre-deployment (in case of multiple deployments), current deployment, and post-deployment periods were pooled. For the civilian sample, data were obtained from the studies if included (N = 9). Information about the characteristics of the civilian samples were quite rare. Information was either not given or minimized to information that data of the civilian samples were collected as part of statewide surveys (e.g. Healthy Kids/Youth Survey). Only one study described the recruiting process of civilian families from health clinics, obstetrical practices, pediatrics office, or parenting classes and that the civilian sample not differed in level of education, age, or child gender. In other cases, studies compared their military samples with normative data (N = 5). For the remaining studies (N = 13) the authors of this meta-analyses did the comparisons of military connected children with normative data as control. In the second comparison, the deployed sample was defined as children with a parent on active duty in a combat zone (if applicable, data from single and multiple deployments were pooled). While the non-deployed sample consisted of children whose parents were reserve component personnel, military personnel who had been deployed but not sent to a combat zone, or personnel who had returned from a deployment more than 12 months ago.

The articles included in the meta-analysis were coded for basic descriptive information (authors, year of publication, study title, sample size, age of the children studied, and type of measurement instruments used) and for whether the deployed military families were being compared to civilian families or to non-deployed military families. The outcome measures were a total score for mental health along with separate scores for the subgroups of anxiety/depression, aggressive behavior, and hyperactivity/inattention. Study characteristics of the included articles are shown in Table 1. The types of informants who provided the mental health data were captured. As the data for the same individual cannot be included in a meta-analysis more than once, in studies where there was more than one informant available for the same sample, such as self-reports and reports by either parent, the report of the parent-at-home was preferred. In case of more than one independent report within a study—for instance, parent-reports for younger children and self-reports for adolescents—all independent reports were analyzed. For the determination of age as a moderator variable, children were categorized into three age groups: early childhood (EC; < 6 years), middle childhood (MC; 6 to < 11 years), and adolescence (AD; 11 to < 18 years). When studies reported results separately by age, effect sizes for each age group were recorded and treated as independent outcomes in the moderator analyses (meta-regression).

In this study, meta-analyses aggregated the effect sizes in random-effect models. Meta-analyses were calculated separately for the comparison of deployed vs. civilian (or normative) data and the comparison of deployed vs. non-deployed data. For each comparison, eight different meta-analyses were implemented to calculate the effect sizes of the comparisons involving the total problem score as well as for the subgroups of anxiety/depression, aggressive behavior, and hyperactivity/inattention. For those studies that provided means and standard deviations, the standard mean difference (SMD; Cohen’s d) was calculated, while for studies that provided the number of specific events in a sample (e.g., prevalence of diagnoses, number of children having specific symptoms with clinical relevance), data were summarized using the Log-Transformed Odds Ratio (log OR). To improve the interpretability of SMD and log OR and to increase the comparability with the earlier meta-analysis by Card and colleagues [18], the effect sizes were converted to the correlation coefficient r [30]. A positive value would indicate that children of deployed parents had more problems than controls, while a negative value would indicate the opposite. As per convention, correlation values around 0.10 were considered to be small effect sizes, values around 0.30 were considered medium, and values around 0.50 were considered large [31]. Statistical heterogeneity of the effects was assessed using I² and tested with a Chi²-Test (Q statistics). I² values of around 25% (I² = 25), 50% (I²= 50), and 75% (I² = 75) were considered to represent low, medium, and high heterogeneity, respectively [32].

For the total problem score, data were obtained from all 27 studies. The effect size was significant for the 21 studies in which it was calculated using SMD (0.51*, 95%-CI 0.31–0.70), but was not significant for the six studies calculating the log OR (1.02, 95%-CI − 0.63–2.66). Meta–regression found no significant difference between the three age groups in total problem score (Q₂ = 2.61, p = 0.27). For the symptom subgroups of anxiety/depression, aggression, and hyperactivity/inattention, information was available from 17, 11, and 13 studies, respectively (for the comparisons on subgroups, see Table 2). Heterogeneity was high for all comparisons, ranging from 94% to 100%.

The results were less consistent for these data than for those involving the comparison with civilian families. Here, information was available from 18 studies for the total problem score and from 13, 8, and 8 studies, respectively, for the three subgroup scores. For the total problem score, the effect size calculated using SMD was significant (0.30*, 95%-CI 0.15–0.45) but was smaller than that seen for the comparison with civilian data. The effect size that was calculated using log OR was not significant (1.37, 95%-CI − 0.82–3.56). Meta-regression showed again that age was not a significant moderator, assessed in 18 studies (SMD) for the total problem score (Q2 = 0.40, p = 0.82). For the symptom subgroups, see Table 2 for comparisons. There was again a wide range of heterogeneity, from 0% to 100%.

The most significant limitation of this meta-analysis was the heterogeneity of the studies analyzed, i.e., the between-studies variability. According to Higgins and colleagues [24], I² values of 25%, 50%, and 75% can be tentatively classified as low, medium, and high, and several of the values seen here were more than 90%. There were several reasons for the high heterogeneity: the studies used different questionnaires and instruments to evaluate psychopathological symptoms and diagnoses; military members belonged to different branches and ranks within the armed forces; and the status of both deployment and non-deployment was defined in multiple ways. Additionally, some of the comparisons were done using civilian samples and others using normative data, and the civilian samples that were recruited for the analyzed studies might not have been as representative in terms of geographical and educational characteristics as the samples that had been recruited for the normative studies.

Another limitation was that apart from age of the children (which, as described above, was distributed very unequally), no mediators or moderators that might have influenced the findings could be explored, because reporting of the data was too fragmentary to allow for meaningful analyses. Moreover, most of the analyzed studies had cross-sectional designs, so it was not possible to draw conclusions on time-dependent courses or causality. Finally, as the number of studies that met the criteria for being included in the meta-analysis was small (21 studies in the SMD analyses and 6 in the log OR analyses), the statistical power for detecting group differences was limited.

A key component of a well-conducted systematic review is an objective and sensitive literature search of multiple sources. An additional research strategy including the term “parent” in our search criteria did not reveal relevant studies. Moreover, we have undertaken an additional review of appropriate projects of the Department of Defense or of RAND Corporation in the United States to examine potential studies that were partly outside the academic publishing. Most of the projects, such as “Military Family Life Project” [80] and “Blue Star Families, Military Family Lifestyle Survey” [81] did not reach scientific inclusion criteria due to the use of standardized and comparable instruments. Only one study was included, “The Deployment Life Study” [57]. A further promising project is “The Millennium Cohort Family Study”, recently published in December 2018 [82], may include in future reviews.