Introduction

Each year, a large number of military members across the Western World release from service (sometimes called ‘separate from service’) and begin reintegration into civilian society1,2,3. Although many military veterans generally adjust well after service4, a significant portion face challenges with the transition5. For example, among recently released Canadian veterans in the 2016 Life After Service Survey, Van Til et al.6 found that 42% reported difficulty with adjustment, compared to 29% among veterans who had been released for more than four years. After release from military service, veterans are at risk of struggling with post-traumatic stress disorder (PTSD) and major depression disorder7, and report challenges with poor physical health (e.g., high prevalence of chronic pain;8,9) and difficulties with finding employment10.

One potential reason for why some veterans struggle with adjusting to civilian life is that a strong sense of shared military identity (i.e., being a soldier) is cultivated during their time in the military and typically becomes a prominent identity for this population11. While in service, the adoption of a strong military identity is required to ensure unified efforts necessary for performing as a unit (e.g., platoon, company, battalion) under stressful and dangerous conditions such as war or other emergencies12. This development of a shared identity acts to promote strong connections among veterans, which creates a sense of community that can feel like family13. Often, military norms and ways of life are distinct from civilian life. For instance, the military environment is highly structured and relies on teamwork, in contrast to civilian societies which tend, at least within the Western World, to be more autonomous and unstructured12. Veterans have described their reintegration to civilian life as feeling alien and lacking in structure and purpose13. The loss of meaning, coupled with identity loss11 and challenges with relating to others13, are some of the ways in which veterans may struggle with adapting to civilian life more broadly.

The period of time after release from military service involves not only adopting a new civilian identity, but also involves readjusting to multiple facets of civilian life. This may include managing mental and physical health challenges14, building new civilian social networks15, and finding employment10. One framework, the Military Transition Theory16, has been used to describe the various factors that may influence veterans’ adjustment, and identify three ‘segments’ related to the transition experience. The first segment, approaching the military transition, describes factors that create the foundation of transition, including personal characteristics (e.g., health conditions, expectations of and preparedness for transition), military/cultural factors (e.g., type of discharge, history of combat), and transitional factors (e.g., positive or negative experiences related to the point of transition itself, predictability of discharge)16. The second segment, managing the transition, describes factors which impact an individual’s progression from military to civilian life16. These include individual factors (e.g., personal coping style, beliefs, and attitudes), available social support from family and friends, as well as navigating broader military or community supports16. The final segment, assessing the transition, specifies several interconnected outcomes associated with transition. Outcomes related to work (e.g., securing adequate employment), family (e.g., reintegrating into family life), health (e.g., physical and psychological health), general well-being, and community (e.g., engagement in community and developing new social networks) are considered to be important outcomes of transition. It should be noted, however, that despite the label of a ‘theory’, the framework by Castro and colleagues16 does not reflect key features of a theory per se17, and is more descriptive of factors that the authors contend might relate (somehow) to veterans’ transition experiences. Psychological theories are characterized by (among other things) the provision of testable predictions/hypotheses and explanations of underpinning mechanisms related to the phenomena of interest17, in this case military transition experiences.

As a complement to the Castro et al. framework, Elnitsky, Blevins et al.18 presented an ecological model wherein interrelated biopsychosocial-cultural factors across system levels are theorized to influence military service members and veterans’ ‘successful’ transition into civilian life. By conducting a critical review of the literature, Elnitsky and colleagues highlighted various factors that can contribute to post-deployment readjustment to civilian life. These included individual-level factors (e.g., physical and psychological health, demographics), interpersonal-level factors (e.g., family, friends), community-level factors (workplace, school, healthcare), and societal-level factors (e.g., cultural, policy, economic).

Given that the transition from military to civilian life involves adjusting to several aspects of civilian life (e.g., employment, relationships, identity), the term ‘adjustment’ in this study broadly encompasses the transition experience as a whole, rather than adjustment to any one specific domain of civilian life. Despite the provision of the above-mentioned military transition frameworks, the propositions embedded within these frameworks have yet to be empirically examined. Further, there is a noted paucity of research on examining the specific period of time following release from service. Instead, studies have primarily focused on the veteran population as a whole (i.e., several years or decades after release), with only a few studies examining the experiences of recently released veterans. Some studies have examined veterans’ health and well-being in the months or years immediately following release. For instance, Vogt and colleagues19 examined well-being (i.e., one potential outcome related to adjustment) across several domains among recently released veterans and found that depressive symptoms at the time of release were the most important predictors of veteran well-being 15 months following release. Further, both Porter et al.20 and Vogt et al.21 examined veterans’ mental health and/or well-being in the first year of release. Contrary to hypothesized findings, the results of these studies revealed relatively stable trajectories of health and social well-being21 as well as depression and PTSD20 immediately following release from service. Porter et al.13 found notable changes in depression and PTSD over time among particular sub-populations for veterans, with veterans who received other-than-honorable discharge and those who were deployed while serving in the army or marines being at greater risk of mental health challenges in the months leading up to and following release from service. Of note, however, these studies focused on outcomes related to mental health and well-being, rather than veterans’ adjustment to civilian life after transition. In addition, these studies focused solely on American military veterans and there remains an absence of longitudinal studies examining transition experiences of Canadian veterans.

Despite the potential of multiple (and multi-domain) factors (e.g., biological, demographic, psychological, chronic conditions) associated with post-release veteran experiences, these have typically been studied in isolation, as individual factors18,22. Indeed, research has yet to ascertain which factors might be most salient in contributing towards, or impairing, an easier adjustment following re-integration into civilian life. With this in mind, the purpose of this study was to simultaneously compare multiple potential predictors of adjustment over time to identify the most important factors that might contribute to or thwart an easier military veteran adjustment. In light of an absence of strong theoretical frameworks, this study used a data-driven approach (via machine learning)23 to address our research question: What are the most important predictors of adjustment to civilian life following release from the Canadian Armed Forces? In this study we used longitudinal data to examine predictors of adjustment (broadly defined) over time among recently released veterans. Machine learning represents a powerful statistical approach, that has been used to better understand the implementation of various clinical tasks (e.g., diagnosis, prognosis, treatment planning) with various patient groups such as those experiencing cancer, cardiovascular disease, and mental health disorders24 and has been utilized to (among other things) identify important predictors of health outcomes (e.g., predictors of mortality from several economic, behavioral, social, and psychological factors)25. Here we use machine learning to identify salient predictors of adjustment following transition out of the military and examined differences between classes with respect to top predictor variables using multinomial logistic regression.

Methods

Data source

Ethical approval was obtained from The University of British Columbia’s Behavioral Research Ethics Board (H22-02856), and all study procedures were performed in accordance with the Declaration of Helsinki. All participants provided voluntary informed consent. Prior to accessing the datasets, the research question, methods, and analysis were pre-registered at the Open Science Framework (https://osf.io/d863f). Data from Statistics Canada’s Life After Service Survey (LASS) cycles in 2013, 2016, and 2019 were analyzed for this study. The LASS is a national longitudinal study (https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=5172) that collects information on the transition from military to civilian life along with health and well-being indicators of veterans. The respondents of the LASS include veterans who released from the Canadian Armed Forces between 1998 and 2012 and who had not re-enlisted at the time of data collection, and excludes regular and reserve force members who released at an entry rank (e.g., private, corporal) and reserve force members with less than 3.5 years of full-time reserve service over their career.

Participants

For the purpose of this study only military members who released within the last 5 years of the LASS recruitment window (i.e., between 2008 and 2012) were included, in order to examine the critical, and under-studied, period of adjustment shortly after release from the military. That is, study participants left the military between 2008 and 2012, and thereafter completed measures in 2013, 2016, and 2019 (see Fig. S1 for details). The LASS utilized stratified sampling and simple random sampling to recruit participants26, whereby Canadian Armed Forces veterans were stratified by rank at the point of release and clustered with regard to three groups (i.e., officer, senior non-commissioned member, or junior non-commissioned member). Longitudinal sampling weightsFootnote 1 (stratified by rank) created by Statistics Canada were applied throughout the analysis. Our sub-sample included 455 recently released veterans (released between 2008 and 2012), who completed data at all three time points (2013, 2016, 2019), and represented (via sampling weights) a total population of 11,100 Canadian veterans who released between 2008 and 2012.

Measures

Military release date was assessed by Statistics Canada by linking military files and the year of release. All data were deidentified prior to being released by Statistics Canada for our use. Adjustment to civilian life (dependent variable) was measured with a single-item question, “In general, how has the adjustment to civilian life been since you were released from the Canadian Forces?”. A Likert-type scale was used to assess responses, which included ‘very difficult’ (0), ‘moderately difficult’ (1), ‘neither difficult nor easy’ (2), ‘moderately easy’ (3), and ‘very easy’ (4). Consistent with a data-driven machine-learning approach25 to identifying meaningful predictors, we sought to include as many possible variables as were available within the LASS dataset that might operate as salient predictors of adjustment to civilian life. These included variables related to demographic factors, military characteristics and experiences, access to care variables, health behaviors, social connection measures, as well as psychological and physical health indices. Specifically, 36 potential predictors of adjustment were assessed, including seven demographic characteristics (e.g., employment, age, education), three military characteristics and experiences (e.g., deployment experience, service branch), four access to care variables (e.g., medical and dental coverage), four health behaviors (e.g., drinking and smoking habits, health care utilization), two social connection variables (i.e., social support, sense of belonging), 11 psychological variables (e.g., mental health, life satisfaction, stress), and five physical health variables (e.g., general health, chronic conditions, pain). All predictor variables were derived from the LASS 2013 dataset (i.e., the first timepoint following release from the military; denoted as the baseline assessment). These 36 variables were included in the analysis to represent a range of factors which may be important for adjustment after release. All binary categorical variables were managed using dummy coding. All subscale measures were scored by Statistics Canada, with the exception of the chronic conditions variable. Details of the putative predictor variables are described in the supplementary file (Methods S1).

Analysis

This study applied a novel approach to conducting mixture growth modelling (MGM; i.e., an examination of latent classes based on response patterns over time)27,28 that involved a 3-step procedure, along with the integration of machine learning. The first step involved use of MGM to identify latent classes. The second step applied machine learning (i.e., random forest analysis) to select important predictor variables. The final step applied multinomial logistic regression to examine class differences in responses to top predictor variables. Syntax for the analyses is presented in the supplementary information in Files S1 and S2.

Step 1: identification of classes

MGM was used to identify latent classes based on similar trajectories of adjustment between individuals over time, based on responses to a single-item measure of adjustment to civilian life. This approach allows for within-class variation of the intercept and growth factors29. Mplus30 statistical software (Version 7.11) was used to conduct the analyses (see Fig. 1 for MGM). To account for the fact that veterans included in this study had released from service in the preceding 1 to 5 years (prior to the 2013 assessment window), time since release, in years (2008–2012), was included as a covariate in the model. Commonly used model fit indices were examined to ascertain model fit, wherein a smaller value indicates a better model fit31. These included the Akaike Information Criterion (AIC), Bayesian Information Criteria (BIC), and Sample-Size-Adjusted BIC (SABIC). Entropy values were also examined, where values of 0.80 or higher are considered good model classification32. Further, differences between models were tested with a Lo-Mendell-Rubin adjusted likelihood ratio test (LMR-A-LRT), with p < .05 indicating a significant improvement between models31.

Fig. 1
figure 1

Mixture growth model to identify latent classes of adjustment over three timepoints (2013, 2016, 2019). Note: T1, T2, T3 = timepoints of adjustment, I = intercept, S = slope, e = error.

Full size image

Step 2: identification of predictors using random forest analysis

We conducted a random forest analysis33 to select important predictors, with latent classes of adjustment (identified in Step 1) as the outcome, to assess a number of potential predictor variables (assessed at the first timepoint post-release with the 2013 dataset). It is similar to other types of tree-based learning where models are built using recursive binary splitting (i.e., splitting each variable to measure the amount of variance) until a single tree is created34. However, random forest analysis extends this approach by making predictions using many trees simultaneously. In addition, random forest analysis incorporates bootstrapping methods (i.e., drawing repeated subsamples, with replacement, from the original dataset) to test a randomly selected set of predictors across several decision trees35. A training set (approximately two thirds of the original sample) is used to generate the decision trees and then the remaining one-third of the sample is used to test the model fit33,34. R statistical software was used to conduct the analysis with the R package ranger36. The out-of-bag (OOB) prediction error was examined to ascertain model performance. Node impurity was measured by the Gini index37 embedded within the R package ranger36 to determine variable importance. For this study, an important benefit of applying random forest analysis is its ability to better account for multicollinearity, commonly seen in health-related survey variables, by reducing the correlation between trees (by only testing a random subset of variables when creating each tree)33,34. To identify the most important variables from the random forest analysis, we identified the most ‘important’ predictors of adjustment by their relative importance (RI)33. In light of the absence of clear recommendations, of ‘cut-off’ values, for determining top predictors33, we examined how classes of veterans differed in scores on the top five (> 0.50 RI) and top ten predictors (> 0.30 RI).

Step 3: examination of class differences on responses to top predictors with multinomial logistic regression

Multinomial logistic regression (MLR) was used to determine how latent classes of veterans differed in scores of the top predictors of adjustment to civilian life. First, we examined the variance in class trajectories accounted for by the top five predictors in our model (partial model). We then examined the extent to which the inclusion of the top ten predictors (full model) significantly accounted for any additional variance, and relied on indicators of effect sizes (via odds ratio; OR), to interpret our results. Effect sizes are interpreted as small (OR = 1.5), medium (OR = 3.5), and large (OR = 9.0)38. Effect sizes of negatively weighted regression coefficients can be interpreted by dividing 1 by the associated OR (1/OR).

Results

In line with data use and release requirements from the data custodian (Statistics Canada) (see footnote 1)Footnote 2, and to ensure the data were sufficiently deidentified and participant anonymity protected, all reported findings were weighted and rounded (see footnote 1), with all descriptive outputs rounded to the nearest base 100. Descriptive statistics for all predictor variables are presented in supplementary information Tables S1 and S2. Mean scores and standard deviations for adjustment at each time point among classes are presented in Table S3. Our weighted sample represented 11,100 Canadian military veterans who released between 2008 and 2012 (weighted Mage = 44.6, weighted SD = 11.0). A majority of respondents were male (88%) and had experienced deployment during their time in service (81%). Participants had most recently served in either the army (weighted, rounded N = 5900, 53%), navy (weighted, rounded N = 3400, 31%), or air force (weighted, rounded N = 1800, 16%) branches of the Canadian Armed Forces.

MGM was used to identify trajectories of adjustment after release and to classify veterans into distinct groups or ‘latent classes’ based on a similar pattern of responses. Weighted responses to adjustment, using longitudinal weights (see footnote 1), were analyzed across all timepoints (2013, 2016, 2019). Five mixture growth models were estimated and model fit indices were compared to determine the best model fit for the data (see Table 1).

Table 1 Summary of model fit indices for models 1–5.
Full size table

The MGM revealed that three classes were determined to be the optimal model based on model fit indices and the number of participants in each class (> 10) (see footnote 2). The results of the MGM indicated the presence of three distinct patterns of adjustment to civilian life over time (see Fig. 2), after adjusting for time since release. The first class (Nweighted = 2900, 26.4% of sample), low adjustment (intercept β1 = − 3.457; slope β1 = − 0.277, SE = 0.274, p = 0.313), represented veterans who reported difficulty with adjustment to civilian life immediately following release and continued to report difficulty with the adjustment over time. The second class (N weighted = 4600; 41.8% of sample) reported a moderate degree of adjustment (i.e., neither difficult nor easy) and remained relatively stable in their adjustment to civilian life over time; as such, they were labelled as moderate adjustment (intercept β1 = 0.000; slope β1 = − 0.289, SE = 0.252, p = 0.251). Veterans in the third group (Nweighted = 3600; 32.7% of sample), labeled high adjustment (intercept β1 = 2.975; slope β1 = − 0.055, SE = 0.307, p =  0.858), consisted of veterans who reported adjusting relatively easily to civilian life at the first timepoint following release, and maintained fairly high and consistent responses to adjustment across all three timepoints. Weighted means and standard deviations of years since release (included as a covariate) were similar across veterans reporting a low adjustment (Myears = 3.11, SD = 1.47), moderate adjustment (Myears = 3.18, SD = 1.48), and high adjustment (Myears = 3.37, SD = 1.48).

Fig. 2
figure 2

Weighted intercept and slope growth factors of adjustment to civilian life from 2013 to 2019 among classes. Note: The reported number of veterans in each class reflect the rounded weighted sample population (as per Statistics Canada requirements). Adj = Adjustment. Individual trajectories of adjustment are not presented due to Statistics Canada data restrictions requiring deidentification of individuals in the dataset.

Full size image

The classes estimated by the MGM served as the outcome variable in the subsequent random forest analysis. All predictor variables were derived from the 2013 data (denoted as baseline data) collection period to predict the classes or grouped responses of veterans. The model parameters included 36 predictors, with six variables tested per split and 500 trees33. Overall, our random forest analysis had an out-of-bag prediction error of 37.4%, indicating that our trained model accurately classified data in the testing set for 62.6% of cases (i.e., a measure of model accuracy). Predictors were assigned absolute values of variable importance indicating the mean decrease in impurity (i.e., extent to which the model is improved by the inclusion of each predictor)33. All predictors were assigned a value of relative importance (calculated by dividing the variable importance of each variable by the maximum variable importance of the variables). A full list of importance and relative importance values for each variable included in the analysis is reported in supplementary information Table S5. For the purpose of this study, the top ten variables with the highest relative importance (i.e., relative importance [RI] of 0.30 or greater), as presented in Table 2, are discussed (for full results of the 36 predictor variables, refer to Table S5).

Table 2 Importance and relative importance of top 10 predictors of adjustment in the random forest analysis.
Full size table

Multinomial logistic regression was used to examine the extent to which classes of veterans significantly differed on each of the predictors of adjustment. Our partial model (predictors with > 0.50 RI) first included the top five predictors of adjustment (namely, life satisfaction, mastery, mental health, BMI, and age) and accounted for 30.3% of the variance in class trajectories (McFaddens R2 = 0.303). Our full model (predictors with > 0.30 RI) accounted for significant additional variance (ΔMcFaddens R2 = 0.032, p < .001) when the next five top predictors of adjustment were included in the model (i.e., the addition of social support, satisfaction with main activity, general health, income, and financial satisfaction). Results for the full model (top ten predictors) are presented in Table S6 (for results from the partial model [top five predictors] refer to Table S7).

Veterans who reported an easier adjustment to civilian life (high adjustment) had greater odds (with a small-sized effect) of displaying better life satisfaction (OR = 2.010, p < .001, 95% CI [1.900, 2.120]) compared to those with a moderate degree of adjustment. Relative to all other variables in the multinomial logistic regression, life satisfaction had the largest effect. Veterans who reported an easier adjustment also had greater odds (with small effects) of reporting better mastery (OR = 1.120, p < .001, 95% CI [1.100, 1.140]), mental health (OR = 1/0.537, p < .001, 95% CI [0.496, 0.582]), satisfaction with main activity (OR = 1/0.643, p < .001, 95% CI [0.593, 0.697]), and financial satisfaction (OR = 1/0.859, p < .001, 95% CI [0.792, 0.932]) compared to veterans with a moderate degree of adjustment.Footnote 3 Veterans who adjusted easily also had greater odds of displaying better social support (OR = 1/0.982, p < .05, 95% CI [0.967, 0.997]), higher income (OR = 1.09, p < .001, 95% CI [1.070, 1.120]), lower BMI (OR = 1/0.983, p < .05, 95% CI [0.969, 0.997]), and were younger in age (OR = 1/0.991, p = .001, 95% CI [0.985, 0.996]) compared to respondents in the moderate adjustment class (see footnote 3). General health did not significantly differ between veterans in classes of high and moderate adjustment to civilian life (OR = 1.050, p = .189, 95% CI [0.975, 1.130]).

Veterans who reported difficulty with adjustment to life after service (low adjustment class) had greater odds (with small-sized effects) of reporting worse mental health (OR = 1.700, p < .001, 95% CI [1.570, 1.830]), general health (OR = 1.700, p < .001, 95% CI [1.570, 1.840]), and financial satisfaction (OR = 1.480, p < .001, 95% CI [1.390, 1.570]) compared to veterans with a moderate degree of adjustment (see footnote 3). Veterans who had difficulty with the adjustment also had greater odds (with small sized effects) of reporting worse life satisfaction (OR = 1/0.892, p < .001, 95% CI [0.849, 0.937]), mastery (OR = 1/0.976, p < .05, 95% CI [0.960, 0.992]), and satisfaction with main activity (OR = 1.070, p < .05, 95% CI [1.020, 1.140]) compared to veterans who adjusted moderately well. Veterans who struggled with their adjustment had greater odds of displaying lower levels of social support (OR = 1.040, p < .001, 95% CI [1.030, 1.060]), higher BMI (OR = 1.020, p < .001, 95% CI [1.010, 1.030]), and were younger in age (OR = 1/0.984, p < .001, 95% CI [0.978, 0.989]) compared to veterans in the moderate adjustment class (see footnote 3). Veterans in classes of low and moderate adjustment to civilian life did not significantly differ in income (OR = 1.000, p = .874, 95% CI [0.980, 1.020]). Weighted mean scores and standard deviations of each top predictor of adjustment to civilian life within each class are presented in Table S8.

Discussion

This study is the first study, to our knowledge, to offer evidence of distinct groupings of veterans who adjust differently over time to life after military service. The mixture growth models provided evidence for three empirically distinct classes of veterans. These included (1) veterans who had an easier adjustment to civilian life (32.7% of the weighted sample), with adjustment remaining relatively stable over time (high adjustment), (2) veterans who had a moderate degree of adjustment (41.8% of weighted sample) over time (moderate adjustment), and (3) veterans who had difficulty with the adjustment (26.4% of weighted sample) following release and continued to have difficulty adjusting over time (low adjustment). The limited research that has examined the period of transition has observed veterans’ experiences of adjustment at a single time point4,39. The results of the current study extend findings from previous cross-sectional work4,39 and provide evidence for trajectories of adjustment over time among recently released veterans (particularly, during the initial 5 + years after release). To our knowledge, this is the first study to examine longitudinal data among Canadian veterans and to report associated effect sizes for predictors of adjustment.

It is interesting to note that even after several years of living in civilian society, none of the classes of veterans displayed substantive changes (i.e., increased or decreased trajectories between 2013 and 2019) across all three timepoints. These findings offer insight into how veterans’ adjustment to civilian life remains relatively stable over time, but that distinct groups of veterans adjust somewhat seamlessly, moderately well, or have difficulty after release with little change for several years that follow. This finding aligns with other studies20,21 that examined mental health and well-being outcomes during the period of transition (rather than adjustment to life after service) and found relatively stable trajectories in the first year after release from service. Given the observation in this study that three distinct classes (or groups) of adjustment were found among recently released veterans, it is possible that the period prior to release may be a particularly important point of intervention to support veterans in their transition, as time spent reintegrating into civilian life does not seem to automatically lead to adjusting more easily to civilian life. One example of supports currently offered prior to release are the career transition services offered by Veterans Affairs Canada, which focus on preparing service members for post-release civilian employment or education through workshops on career planning, resume building, interviewing skills, job searching, and more. It should be noted, however, that supporting the needs of veterans in the period prior to release has received limited research attention and thus warrants future investigation.

An important precursor to intervening and improving these trajectories of adjustment among groups of veterans is understanding the predictors of adjustment. Some prior research has sought to identify potential individual-, interpersonal-, and societal-level factors18 that may play a role in military veteran adjustment. For example, MacLean and colleagues39 examined predictors of adjustment using cross-sectional data from the LASS and found that veterans who released at a lower rank, for involuntary or medical reasons, or who served in the army had higher odds of a difficult adjustment. Our study further adds to prior research as it is the first study, to our knowledge, to examine the LASS dataset longitudinally. The current study made use of a machine learning approach33 to simultaneously examine a full range of predictor variables and ascertain their relative importance in relation to veteran adjustment. The random forest analysis revealed that the top ten most important predictors of adjustment to civilian life, in order of importance, were life satisfaction, mastery, mental health, age, BMI, social support, satisfaction with one’s main activity, general health, income, and financial satisfaction5.

It is worth noting that compared to measures of physical health, other demographics (e.g., employment, gender), military experiences (e.g., deployment), and health behaviours, several of the top ten most important predictors (measured at baseline) were psychological variables. It is entirely conceivable that some of these variables (e.g., health behaviours, physical health, military experiences) indirectly contribute to adjusting more seamlessly to civilian life, via their effects on psychological factors, but due to their more distal nature did not emerge as direct predictors of adjustment in this study. Regardless, it is reasonable to conceive that one’s perception of adjusting well after transition is shaped by psychological factors, and that important psychological ‘predictors’ could be targeted to support transition. This means that providing opportunities for veterans to develop a sense of mastery, improve their mental health, and help them understand what they might need to feel ‘satisfied’ with life, finances, and one’s main activity (e.g., employed, student, retired) after service may be important avenues to support veterans in identifying goals and actionable change that can lead to better adjustment after release.

The most important predictor of adjustment to civilian life was life satisfaction, which is closely aligned with subjective well-being40,41. Specifically, in this study, veterans who reported an easier adjustment to life after service (spanning six to 11 years post-release) also reported feeling more satisfied with their lives at the first timepoint following release. Veterans who had a somewhat seamless adjustment over time were also more likely to report feeling satisfied with specific aspects of their life after service, including their current main activity (e.g., employed, student, retired) and finances at the first timepoint following release. Previous research with (non-military) adults has found that that life satisfaction is an important and modifiable ‘health asset’ that prospectively predicts better downstream physical, behavioral, and psychological health42. Life satisfaction captures the extent to which a person generally experiences positive well-being, whereas adjustment reflects the extent to which an individual copes with the demands of the military-to-civilian transition. With that said, it is entirely conceivable that a general sense of life satisfaction, independent of the transition out of the military, may account how veterans report that they have adjusted. The bivariate correlation between adjustment and life satisfaction at time 1 was r = .54 (see Table S4), which suggests that the two constructs are correlated, but distinct. Nevertheless, caution should be exercised in interpreting a directional effect of life satisfaction on downstream adjustment, and indeed future research is warranted that unpacks the complex interplay between indicators of well-being and subsequent adjustment to any new life circumstance (following military transitions or other life transitions) and vice versa.

An additional limitation of this study was that we were not able to examine how factors prior to release predicted adjustment to civilian life, and thus future research using robust longitudinal designs is warranted in order to examine the replicability of this finding and to ascertain directionality. In the general population, a number of salient factors, that were not measured as part of the LASS, have been found to be closely tied to life satisfaction. These include, but are not limited to, having a sense of purpose43 and meaning in life44. Among a military population, previous work has found that veterans often report a lack of purpose after release13,45. Examining whether life satisfaction drives a more seamless adjustment or vice versa represents an important question to be addressed in future research, as well as other correlates of life satisfaction (e.g., meaning, purpose) that might enable military veterans to better adjust to civilian life.

In this study, veterans who felt that they had a greater sense of mastery (i.e., feelings of control over one’s own life) after leaving the military reported an easier adjustment over time (i.e., in the six to 11 years post-release). These findings align with cross-sectional studies conducted with Canadian veterans39,46 that found military veterans with higher levels of mastery had greater odds of adjusting well after service. The importance of mastery for recently released veterans may be related to aspects of civilian life that allow for (and often require) autonomous decision making, which is in stark contrast to a highly structured environment in the military47. Providing opportunities for military personnel who are about to release from service to develop a sense of personal control, through active participation in decision making processes for example, may be an important area for veterans’ support in addition to the typical supports offered by agencies such as Veterans Affairs (e.g., employment, health care, benefits).

It is also worth noting that a more general measure of mental health was an important predictor of adjustment to civilian life, but other measures of mental health were not, such as PTSD symptoms or diagnosed mood and anxiety disorders. Specifically, veterans who had difficulty with the adjustment were more likely to report worse mental health compared to those with moderate adjustment over time. It may be that the general measure of mental health used in this study captured sub-clinical levels of mental health, rather than clinical threshold levels of PTSD, mood disorders (e.g., depression), or anxiety, and that these more generalized symptoms of mental health play an important role in adjustment. Similarly, veterans who had difficulty with adjustment were more likely to report poorer general health compared to veterans with moderate trajectories of adjustment. When considering the differences between veterans who had an easier adjustment compared to those with a moderate degree of adjustment (i.e., neither difficulty nor easy), veterans were more likely to report better mental health but there were no reported differences in general health. Taken together, mental health and general health appeared to have the greatest effect on predicting difficulties with adjustment over time (when compared to moderate trajectories) relative to other top predictor variables. In addition to providing support to those who receive a clinical psychiatric diagnosis or have a chronic health condition, it is possible that veterans who report poor general psychological health/functioning may benefit from broader health supports during transition. This could ensure that veterans who are potentially at-risk of struggling with the transition could be identified prior to release and could be connected with more comprehensive transition supports during the administrative release process (e.g., supports for employment transition, but also general counselling and supports for engaging in healthy diet and physical activity behaviours). This aligns with a previous cross-sectional study based on the LASS dataset48, which identified general mental health as a meaningful indicator of veterans being at risk of experiencing a difficult adjustment.

Another important predictor of adjustment to civilian life (albeit with a small-sized effect), and one which has been shown to be important for veterans’ psychological health49,50, was social support. Veterans who had a more seamless adjustment were more likely to report higher levels of social support after release and veterans who had difficulty with the adjustment were more likely to report lower levels of social support compared to veterans who reported moderate adjustment post-release. Researchers have found that community-based programs can encourage social connections among veterans, such as through socializing over a coffee51 or sport and exercise52. Continuing to find ways for veterans to develop meaningful social connections within their civilian communities following release may be an important means of supporting veterans in their adjustment over time.

In addition to psychological and social factors, other physical health and demographic variables were top predictors of adjustment but their effects were very small in size when predicting differences between classes. These included age, income, and BMI. Specifically, veterans who had difficulties with the adjustment were more likely to have higher BMI, and veterans who had an easier adjustment were more likely to report lower BMI, compared to veterans with moderate adjustment trajectories. Given the differences reported by veterans in broad measures of physical health (as opposed to other tested measures such as chronic physical health conditions), future work could consider the extent to which weight status may change over time and contribute to adjustment in the years following release from military service. Compared to veterans with a moderate degree of adjustment, veterans who adjusted well to civilian life were more likely to have higher income levels (although with very small effect sizes). No differences in income were reported between veterans with low (i.e., difficult) and moderate trajectories of adjustment. Although age was an important predictor in the random forest analysis, the differences in age between trajectories of adjustment were trivial. This is consistent with Borowski et al.’s study53 which used latent class analysis to identify distinct age-stratified groups of veterans in their responses to measures of well-being following transition, with similar patterns of well-being reported between younger and middle-aged veterans.

Beyond the top ten important predictors of adjustment identified in this study, previous research has identified other challenges to adjustment including physical health conditions such as chronic pain8, a limited transfer of military skills to civilian work10, and deployment experiences39,54. However, when compared simultaneously in this study, these other challenges faced by veterans did not emerge as important predictors of adjustment to civilian life in this study. Some prior research has also found that women in the military are significantly more likely to experience harassment or assault49, experience high rates of PTSD and other mental health diagnoses after release from military service55,56, and often face other forms of gender inequalities throughout their career57. In the current study, sex did not emerge as an important predictor of adjustment to civilian life when simultaneously considered alongside other predictors.

There are several strengths to this study that should be noted. First, this study utilized a longitudinal dataset, weighted to be representative of the Canadian military population (who released between 2008 and 2012), that was designed specifically to examine the health and well-being of veterans after release from the military. In particular, this study examined the understudied initial period of adjustment to civilian life by limiting our sample to veterans who had been released within the previous five years prior to the 2013 survey data collection (i.e., 2008–2012). The findings highlight several important psychological predictors of adjustment that could be considered in future theoretical frameworks of transition, and have the potential to inform targeted supports prior to release or early on in the transition period. For example, these findings could be used to inform the screening tool48 used by Veterans Affairs Canada to identify veterans who may be at risk of a difficult adjustment. This screening tool was initially informed by separate cross-sectional analyses of predictors of adjustment in the LASS dataset48. In contrast, in our study, we were able to identify predictors of adjustment trajectories, and in the process identify the relative importance of each of these predictors. From a knowledge mobilization perspective, our study identified additional risk factors not found by VanTil and colleagues48 such as social support, which have the potential to inform future iterations of the risk screening questionnaire. Another notable strength of this study is the use of machine learning to simultaneously examine a number of potential predictors of adjustment to civilian life. In particular, a strength of random forest analysis is its’ ability to better account for potential multicollinearity among predictors that are often prevalent among health-related variables33,34.

Balanced against these strengths, limitations should also be noted. First, the target dependent variable (adjustment to civilian life) was assessed by Statistics Canada using a single item measure (as were some of the predictor variables). The use of single-item measurements can be beneficial to minimize participant burden in longitudinal studies and are particularly useful in research involving vulnerable populations58. However, such measures have also been critiqued as being more susceptible to measurement error58 and having unknown reliability. With that said, previous research using single-item measures of complex constructs such as happiness59 and subjective well-being60 has found such measures to be highly correlated with multi-item measures and adequately reflect the target constructs. While our assessment of subjective adjustment can be considered a self-reported global measure of adjustment, it would be worthwhile for future research to examine adjustment in relation to specific life domains such as employment, social networks including friendships and family, as well as community integration16. In this way, researchers can determine whether adjustment within specific life domains vary across time differently and whether different subsets of predictors differentially predict adjustment trajectories within specific life domains.

Given that the survey development and data collection were not conducted by the research team, there are several variables not included in this study that have been shown to impact military veterans health and well-being. For instance, veterans have reported experiences of loss of purpose13,45 and loss of identity11 following release from service, but these constructs were not measured in the LASS. Physical activity, a modifiable behaviour, was not assessed in the LASS but might support veterans’ general mental health and well-being (and thus, adjustment). Given that engaging in physical activity has been suggested to be an effective treatment for managing mental health symptoms among veterans52,61,62, it would be worthwhile to ascertain the extent to which physical activity can support veteran’s well-being (and possibly adjustment post-release). Further, it should be noted that due to the design and recruitment procedures involved with the LASS study, no pre-release measures were obtained from study participants and, thus, we were precluded from examining the effects of pre-release indicators (e.g., mental health conditions that arise during active service) that may impact the trajectories of post-release adjustment. Similarly, other studies examining this period of transition (e.g., 19,63) have only examined veterans’ health and well-being following release from service; thus future research designs should include measures prior to and in the years following release.

Despite recognition of the challenges associated with adjusting to life after service, there has been limited research attention dedicated to understanding important predictors of adjustment and the ways in which veterans can be best supported during this key period in their lives. Findings from this study demonstrate the importance of psychological (e.g., mastery), social support, and other (e.g., financial) factors that may lead to a more seamless transition to civilian life during the immediate years following release.