Military and demographic predictors of mental ill-health and socioeconomic hardship among UK veterans

Data collection for the first phase of the KCMHR cohort study began with the commencement of the Iraq War in 2003 [11]. Participants were identified by the UK Ministry of Defence’s Defence Statistics unit. A list of all personnel, excluding special forces and high security personnel, who had deployed on to Iraq between Jan 18 and April 28, 2003, was generated. A similar list of UK service personnel serving in the armed forces on March 31, 2003, but not deployed to Iraq was generated as the comparison group. The total sampling frame comprised 17,499 individuals. Several strategies were used to contact potential participants. Initially study participants were allocated either to receive a questionnaire by post, or, where military postcodes indicated a large number of personnel at the same location, serving personnel were assigned a visit from the research team. Visits and mail-outs were done simultaneously. Those who did not wish to participate during a base visit could leave at any time. Data collection was performed June 2004–March 2006, with 10,272 responding, a response rate of 59% (62% among the deployed group and 56% in the non-deployed group). Full details of sample identification and data gathering procedures can be found elsewhere [11].

Data collection for the second phase was performed from Nov 2007 to Sep 2009, and included those respondents at the first phase who were still eligible for contact (n = 9395), as well as a new sample of those who had deployed to the conflict in Afghanistan (n = 1789) as well as a new sample of non-deployed controls (n = 6628). 9990 (56%) responded at phase 2; response rates were 68% for the follow-up sample, 50% for the new respondents who had deployed to Afghanistan and 40% for the new non-deployed controls [12].

Phase 3 data was collected between Oct 2014-Dec 2016. Those who consented to further contact were sent questionnaires (n = 12,280), along with an additional sample of non-deployed comparators (n = 8581). 8093 (44%) responded on this third phase; of these 58% responded in the follow-up sample from which most of the participants in this study are drawn (response among the replenishment sample was 24%) [13].

The samples were representative of the sampling frames available from the Defence Statistics, excluding those in Special Forces. The veterans within this study may have left at any time over the course of these data collection phases. To maximise the available data, we used the most recently-provided data for each individual; thus relationships between outcome and exposure factors are effectively cross-sectional, except where data is missing and is being imputed from last available data. This study focuses on ex-regular veterans, and hence those still serving at last response were excluded; furthermore both those who were serving as a reserve, or were no longer serving but had only served as a reserve, were excluded as their civilian career history differs from regulars. In total this study includes 3095 participants who responded in the most recent phase as well as 896 who last responded at phase 2 and 301 who last responded at phase 1, for a total of 4292 ex-Service personnel in the study.

Mental health measures were acquired via self-report questionnaire for all phases. Probable PTSD was determined using the 17-item National Centre for PTSD Checklist (PCL-C). PCL-C comprises 17 items each rated on a 5-point scale score 1–5 (range 17–85). We used the recommended cutoff of 50 or more to determine caseness. At this cutoff the PCL has an overall diagnostic efficiency of 0.825 as verified against the gold standard Clinician-Administered PTSD Scale. Cronbach’s alpha for the scale is 0.939 [14].

Common mental disorders (CMD) were measured using the 12-item General Health Questionnaire (GHQ-12). Each of the 12 items provides four options for response of increasing severity; these were scored 0–0–1-1, using the commonly-used cutoff of 4 to qualify as a case (range 0–12). Cronbach’s alpha for GHQ-12 has been estimated as being between 0.82 and 0.86 [15].

Alcohol misuse was determined using the World Health Organization Alcohol Use Disorders Identification Test [16], which includes three items on alcohol intake, three on alcohol dependence and four on alcohol-related problems, for a total score range of 0–40. This study used a relatively high cutoff of 16 (defined as hazardous use that is likely to be harmful to health), as is customary in studies of this cohort due to the high level of drinking in the UK military [17].

Socio-economic outcomes were also taken from self-report questions. Questionnaires asked whether respondents were currently working in a civilian job, as a private security contractor, self-employed, not working and looking for employment, not working due to ill health, or retired. Those not working (whether due to ill health or looking for employment) were classified as unemployed (including a small number who were over 65 but did not state they were retired); those who indicated some form of employment (including self-employed) were classified as employed; those retired, in education, or in some other non-paying role were included in a third category. For brevity this last category is not discussed in this article, but data is available on request.

Age was determined at time of most recent questionnaire response and subdivided into 5- or 10-year brackets. We also include variables for last reported education (divided into four categories: no qualifications, O-level or equivalent, A-level or equivalent, and degree level or higher), and childhood adversity in two domains: childhood externalising behaviour and family adversity.

Two separate variables for childhood externalising behaviour and family adversity were derived from a 16-item checklist using the root “When I was growing up …” [18]. This was an original checklist based on a combination of three items adapted from the Adverse Childhood Exposure scale [19] and other items based on general population evidence of childhood exposures for later adverse health outcomes; the checklist has not been validated as a measure, nor have either of the derived variables. Family adversity score was determine by summing endorsements to not coming from a close family, getting shouted at a lot at home, not feeling valued by family, regularly witnessing fighting or verbal abuse between parents, not having a family member they could talk to, being regularly hurt by a parent/caregiver, having a parent with substance abuse problems, and/or not doing things as a family [20]. Those who endorsed getting into fights at school plus at least one of playing truant from school, being suspended or expelled from school, or doing things that got them into trouble with the police were classified as demonstrating childhood externalising behaviour [21].

Military-related factors included rank, Service arm, role, history of deployment, and method of leaving the Services. Rank was determined from the most recent response, and categorised into Commissioned Officer, Senior Non-Commissioned Officer (SNCO) (sergeant-equivalent rank and above), Junior Non-Commissioned Officer (JNCO) (lance-corporal and corporal-equivalent ranks), and other ranks (i.e. private soldiers and equivalent enlisted ranks). Service arm was separated into Army, Royal Air Force (RAF), and Naval Services (which includes Royal Marines). Deployment was separated into those who had deployed to Iraq and/or Afghanistan at least once, and those who had deployed to neither. Role in parent unit (a small proportion of individuals may have different roles on different deployments) was reduced to those with a combat role, and all other roles. Method of leaving was determined by self-report; “End of service term or run out date”, “premature voluntary release”, and “voluntary redundancy” were categorised as “planned leaving”, “administrative discharge”, “temperamental unsuitability”, “disciplinary discharge” and “compulsory redundancy” were categorised as “unplanned leaving”, and “medical discharge” was treated as a third category.

Individuals may have provided multiple responses to the same questions in different phases of data collection. Consequently, responses were taken at the latest phase in which they became a veteran (i.e. most recent post-transition responses). Where data for socio-demographic and military factors were missing, prior responses were utilised, but this was not done for outcome variables to ensure that only contemporaneous transition outcomes were analysed.

There were excellent completion rates for most variables (< 0.5% missing data for most variables, < 5% missing data for education achievement, PCL, GHQ, AUDIT, and socioeconomic outcomes), therefore analyses involving these variables was performed on a complete-case basis.

Associations between mental health/socioeconomic outcomes and demographic/military factors were determined by logistic regression, producing odds ratios (OR) as indicators of effect size. For employment status, multinomial logistic regression (logistic regression with more than two discrete outcomes) was utilised to produce multinomial odds ratios (mOR). Subsequent to the univariate analysis, an adjusted model was produced in which all factors associated with each outcome were included as potential confounders to take account of non-independence between covariates.

The underlying cohort was randomly sampled from the serving population of the UK Armed Forces and weighted for both sample and response; no weights were applied as the veterans in this article represent a non-random subsample. Analysis was performed using STATA 15 [22].

Table 1 shows the distributions of the demographic and military variables. Most of the veterans included in this sample are over 40 years old, male, served in the Army, were senior non-commissioned officers, left in a planned way, and attained A-level or higher qualifications.

Mental health and socioeconomic outcomes are shown in Table 2. CMD is the most frequent negative outcome encountered by veterans, but a sizeable percentage experience PTSD, alcohol misuse, financial difficulties and unemployment, ranging from 7.4 to 11.8%.

Significance and effect size of the associations between demographics, military-related factors, and post-service mental health of veterans are shown in Table 3. After adjustment, probable PTSD was associated with childhood adversity, lacking any formal qualifications, being ex-Army, lower rank, deployment to Afghanistan/Iraq and having a combat role. The largest impact was from leaving the Armed Forces due to medical discharge; other unplanned leaving, rank and childhood adversity also had substantial effect sizes.

Some of these associations held for CMD, which was associated with lower rank, childhood adversity, and unplanned leaving, particularly those with medical discharges. Deployment, role, and Service arm were not associated with veteran CMD. The oldest veterans were less likely to report symptoms of CMD.

There was an inverse relationship between age and alcohol misuse, and men were more likely to misuse alcohol than women. Those educated to O-level standard were more likely to misuse alcohol than those who were more educated, and those who experienced childhood adversity were more likely to misuse alcohol. Military factors were not associated with alcohol misuse except that alcohol misuse was associated with unplanned leaving, but not with medical discharges.

Associations between demographic/service-related factors and socioeconomic outcomes are shown in Tables 3 and 4. Young veterans were more likely to be unemployed, but there was no overall trend with respect to age and financial difficulty. Those with higher educational attainment were less likely to be unemployed or experience financial difficulty. Those with a history of childhood adversity were more likely to encounter financial hardship after leaving service, but not unemployment. Service arm did not affect unemployment or financial difficulty, but higher ranks were less likely to encounter financial hardship (though rank did not make a difference to likelihood of unemployment). As with mental health outcomes, the largest impact was due to method of leaving, where again medical dischargees were the most at-risk group of unemployment and financial difficulty.

Study respondents were originally selected while still serving in the UK Armed Forces; thus, this is effectively a prospective cohort. This avoids the limitations of identifying and reaching a retrospective group of veterans, which is utilised in most other studies on veterans and has a high risk of being unrepresentative. The original sample was generated to be representative of the makeup of two sampling frames from the UK Armed Forces in 2003, those deployed to Iraq and those not deployed to Iraq (at that time). The sample of veterans here, which was non-random as it was a sub-sample defined by the event of leaving and hence analysed without weighting, should nonetheless be approximately representative of veterans who left between 2003 and 2016. It is unlikely that any departure from representativeness would not have biased any of the associations reported in this study. Compared with MoD transition surveys, this study covers a much wider period of time post-service and represents a sample that includes the full spectrum of the trained strength of the Armed Forces in contrast to the limited selective sample of those who undertook resettlement activities [27]. As this study uses data from several phases of data collection, not all results are contemporaneous, but this has the advantage of presenting outcomes over a broader range of time. Unplanned leavers are underrepresented in this study, as they may drop out of contact in the time between selection into the study and sampling. It was not possible to determine time of onset of mental health problems, as in most population-based studies; the analysis is of prevalent cases of symptoms post-service, which includes both new and old cases of the disorders being explored. This issue also prevents longitudinal analyses. The information was reported by the participants and we cannot exclude the possibility of recall bias.