Probable post-traumatic stress disorder and harmful alcohol use among male members of the British Police Forces and the British Armed Forces: a comparative study

ABSTRACT Background: British Armed Forces’ and Police Forces’ personnel are trained to operate in potentially traumatic conditions. Consequently, they may experience post-traumatic stress disorder (PTSD), which is often comorbid with harmful alcohol use. Objective: We aimed to assess the proportions, and associations, of probable PTSD and harmful alcohol use among a covariate-balanced sample of male military personnel and police employees. Methods: Proportions of probable PTSD, harmful alcohol use, and daily binge drinking, were explored using data from the police Airwave Health Monitoring Study (2007–2015) (N = 23,826) and the military Health and Wellbeing Cohort Study (phase 2: 2007–2009, phase 3: 2014–2016) (N = 7,399). Entropy balancing weights were applied to the larger police sample to make them comparable to the military sample on a range of pre-specified variables (i.e. year of data collection, age and education attainment). Multinomial and logistic regression analyses determined sample differences in outcome variables, and associated factors (stratified by sample). Results: Proportions of probable PTSD were similar in military personnel and police employees (3.67% vs 3.95%), although the large sample size made these borderline significant (Adjusted Odds Ratio (AOR): 0.84; 95% Confidence Intervals (CI): 0.72 to 0.99). Clear differences were found in harmful alcohol use among military personnel, compared to police employees (9.59% vs 2.87%; AOR: 2.79; 95% CI: 2.42 to 3.21). Current smoking, which was more prevalent in military personnel, was associated with harmful drinking and binge drinking in both samples but was associated with PTSD in military personnel only. Conclusions: It is generally assumed that both groups have high rates of PTSD from traumatic exposures, however, low proportions of PTSD were observed in both samples, possibly reflecting protective effects of unit cohesion or resilience. The higher level of harmful drinking in military personnel may relate to more prominent drinking cultures or unique operational experiences.


Introduction
The military and police respond rapidly during national, and international, disasters and conflicts, and often operate under high pressure, potentially being exposed to traumatic situations. Nevertheless, there may be some differences between these occupations in the types of exposure or type of traumatic experience. For example, military personnel may face intense stressors during set periods of time (e.g. deployment to a conflict situation), whereas for police employees the exposures may occur more regularly and in some cases be part of their daily routines. In terms of traumatic experiences, military personnel are more likely to have killed others, whereas police employees have to respond to and investigate civilian deaths and severe abuse of vulnerable others such as children (Hartley, Sarkisian, Violanti, Andrew, & Burchfiel, 2013;Osório et al., 2018;Stevelink et al., 2020). In addition, both groups may experience other causes of occupational stressors, such as competing demands with family life, demandcontrol imbalances or poor organizational support (Harvey et al., 2017(Harvey et al., , 2018. It is possible that varied nature of occupational and trauma stressors experienced by military personnel and police employees may have a differential impact on their mental health and patterns of alcohol consumption. Prevalence estimates of adverse mental health outcomes among members of the UK Armed Forces are well documented as a result of a representative, longitudinal study set up to explore the impact of deployment to Iraq, and subsequently Afghanistan, on the health and wellbeing of military personnel Hotopf et al., 2006;Stevelink et al., 2018). The most recent estimates from this cohort study indicate a prevalence of 6% for probable PTSD and 10% for alcohol misuse among military personnel (Stevelink et al., 2018). Most relevant mental health research concerning the UK police forces, has often been conducted in the aftermath of an emergency, and/or included only a select few police forces (Lawson, Rodwell, & Noblet, 2012;Maia et al., 2007;van der Velden et al., 2013). Though recently, a large UK survey suggested that of those police employees exposed to trauma, about one in five would develop symptoms of PTSD (Brewin, Miller, Soffia, Peart, & Burchell, 2020). Additionally, a recent meta-analysis identified a pooled global prevalence of 5% for harmful alcohol use among police employees and 14% for PTSD (Syed et al., 2020). However, an international review of hazardous and harmful drinking in trauma-exposed occupations, identified no UK studies of alcohol use in police officers (Irizar, Puddephatt, Fallon, Gage, & Goodwin, Under Review).
In this paper, we explore the proportions, and prespecified associated factors, of probable PTSD and harmful alcohol use among covariate-balanced samples of male members of the British Armed Forces and the British Police Forces. In addition, we explore whether there is a difference in the comorbidity of probable PTSD and harmful alcohol consumption between the two samples. This study is preregistered on Open Science Framework (DOI 10.17605/OSF.IO/7PTWX), where the research questions and data analyses plan are outlined in more detail.

Airwave health monitoring study
Cross-sectional data on police employees was obtained from the Airwave Health Monitoring Study, which was established to determine possible health risks associated with the use of Terrestrial Trunked Radio (TETRA), a digital communication system used by emergency services since 2001 (Elliott et al., 2014). A total of 41,038 police employees completed measures relating to mental health and alcohol consumption, between June 2006 and March 2015. Out of the 54 existing police forces, 28 agreed to participate, with the response rates averaging 50% across participating forces. Participants completed an enrolment questionnaire including demographic, health and lifestyle items, and a health screen conducted by trained nurses. The Airwave Health Monitoring Study design and protocol have been described in detail in a previous publication (Elliott et al., 2014). We will refer to this sample as the 'police sample' throughout the rest of this paper.
Participants were eligible for inclusion in the study if they identified themselves as an active member of the police force (e.g. inspector, police constable/sergeant or police staff). Those whose data were collected in 2006 were excluded as this version of the protocol did not include the measure of PTSD. This procedure is outlined in detail in Figure 1.

Health and wellbeing cohort study
Data on UK military personnel was obtained from the health and wellbeing cohort study established by the King's Centre for Military Health Research. Participants completed a self-administered questionnaire that was available in both hard copy and electronically (latest phase only), and included questions relating to demographics, service information, experiences during and returning from deployment, mental health, physical health and lifestyle. This cohort study was initially set up to investigate the impact of deployment to the conflict in Iraq on the health and wellbeing of military personnel (data collected between 2004-2006) (Hotopf et al., 2006) including a random sample of regular and reserve personnel of the UK Armed Forces, stratified by deployment status (phase 1; n = 10,272, response rate 59%). As the conflict in Iraq continued and personnel were also deployed to Afghanistan, all personnel included in phase 1 were asked to take part in phase 2 (data collected between 2007-2009). In addition, a random sample of personnel deployed to Afghanistan between April 2006 and April 2007 (termed the HERRICK sample), and a sample of newly trained personnel who joined the Armed Forces since April 2003 (termed the replenishment sample), were included (phase 2; n = 9990, response rate 56%) ). An additional 300 responders were included who only filled in a short version of the phase 2 questionnaire after the main data collection for this phase had finished. Again, a replenishment sample of newly trained personnel who joined after June 2009 was included in phase 3 (data collected between 2014-2016), in addition to those who took part in phase 2 and agreed to future contact (phase 3; n = 8,093, response rate 57.8%) (Stevelink et al., 2018). The procedures for each phase are described in detail in previous publications Hotopf et al., 2006;Stevelink et al., 2018). We will refer to this sample as the 'military sample' throughout the rest of the paper.
To ensure comparability with the police sample, only regular serving personnel were included from the military sample for which data was collected during phase 2 (2007-2009) and phase 3 (2014-2016) of the military health and wellbeing cohort study. Further, military personnel who filled in the short questionnaire at phase 2 were also dropped. Not all measures needed for the current comparative analysis were available. Females were also excluded for the purpose of this analysis (analysis will be repeated for females in a separate study); it is important to study males and females separately, as female members of both samples are relatively unique with regards to the type of role they can hold (particularly within the military) and have higher rates of mental health problems, than males (McManus, Bebbington, Jenkins, & Brugha, 2016). Further, the proportion of females is usually substantially higher in the police force (approx. 30%) (Allen & Audickas, 2020) than in the Armed Forces (approx. 10%) (Dempsey, 2019).

Demographic, occupational and health variables
Comparable demographic, occupational, and health variables were obtained from both samples, including age, marital status, educational attainment, and smoking status. Income and police role (police staff, police constable/sergeant, inspector or above) were also obtained from the police sample. Type of Service at baseline (Naval Services, Army, Royal Air Force), rank (commissioned officer, noncommissioned officer or other) and deployment (yes/no), were obtained from the military sample.

PTSD
In the police sample, probable PTSD was measured using the 10-item Trauma Screen.
Questionnaire (TSQ) (Brewin et al., 2002). Response options were on a five-point scale ranging from 'not at all' to 'extremely'. Any responses other than 'not at all' were scored as 1 (score range 0-10).
A score of 6 or more was defined as indicative of probable PTSD. The TSQ was only asked if participants responded positive to the following screening question: 'Have you been bothered by a disturbing incident which has occurred over the past 6 months?'.
In the military sample, probable PTSD was measured using the 17-item National Centre for PTSD Checklist, civilian version (PCL-C) (Blanchard, Jones-Alexander, Buckley, & Forneris, 1996). Response options on the PCL-C are based on a five-point scale ranging from 'not at all' to 'extremely' (score range 17-85). A score of 50 or more was defined as indicative of probable PTSD. Research indicates that the PCL-C and TSQ, using the same defined cut-off scores as above, show similar prevalence estimates in the UK general population (McManus et al., 2016). a Participants with data for both phases were included in phase 3 to create a more equal distribution, which reflects the distribution of the Airwave data. b Participants who completed the Airwave Health Monitoring Study survey in 2006 were not asked the PTSD items, and so, were dropped. The phase 1 KCMHR data were also dropped as this was the same timeframe.

Alcohol consumption
In the police sample, alcohol consumption was measured using a past week's drinks diary, which asked participants to state the number of drinks they had consumed, for the following: white wine, red wine, fortified wine, spirits and beer (converted to units). In the military sample, the Alcohol Use Disorder Identification Test (AUDIT) (Saunders, Aasland, Babor, De la Fuente, & Grant, 1993) was used. To harmonize the measure of alcohol consumption with the police sample, two items were used ('how often do you have a drink containing alcohol?' and 'how many units do you have on a typical day of drinking?'), to estimate total weekly units. AUDIT responses for the latter item are usually on a five-point scale, ranging from 1-2 drinks (scored as 0) to 10 or more drinks (scored as 4), but participants were given additional options, up to 30 or more, and provided responses in units. The frequency of consumption was multiplied with the midpoint for typical units, e.g. a participant who drinks two to three times a week and has 7 to 9 units on a typical day of drinking, would score 16 on total weekly units (2 × 8).
The UK's Chief Medical Officer's recommendation for weekly alcohol consumption was used to code both samples as 'low-risk' (≤14 units) and the National Institute for Health and Care Excellence (NICE) guidance for males was used to code anyone drinking above this as 'hazardous' (>14-50 units) and 'harmful' (>50 units) drinkers (Department of Health and Social Care, 2016;NICE, 2014). In the police sample, participants were asked if they currently drink alcohol; those who responded 'no' were categorized as 'non-drinkers'. In the military sample, participants who responded with 'never' to the first item of the AUDIT, were categorized as 'nondrinkers'.
Both datasets included a measure of binge drinking, i.e. 'how often do you have six or more drinks on one occasion?'. Responses were given on a five-point scale; however, the comparability of this variable was reduced as the outcomes were worded slightly differently in each sample, except for 'daily or almost daily'. Therefore, a binary variable, 'binge drinks daily or almost daily' vs 'does not binge drink daily or almost daily', was created to reflect more harmful drinking behaviours.

Entropy balancing
Entropy balancing is a multivariate reweighting method, building on propensity score matching but allowing the full use of the larger sample (rather than selecting a matched sub-sample), used to achieve covariate balance on a range of pre-specified variables (year of data collection, age and educational attainment), to increase comparability (Hainmueller, 2012). Entropy balancing was used to create a weight value for all police employees, to be more comparable to the military dataset (which is smaller and more occupationally distinct), which is then used as a weight when estimating proportions in police employees.
A binary variable was created for year of data collection (January 2007-June 2011, reflecting phase 2 of the military sample, vs July 2011-December 2016, reflecting phase 3 of the military sample), using broad ranges rather than precise year to account for differences between the samples. Age was grouped into 10-year age bands, from <30 (starting at 18 and 20 years old for police employees and military personnel, respectively) up to ≥50 years old (up to 70 years old across both samples). Educational attainment was split into low (O levels/GCSEs or below) and high (A levels and equivalent or higher). There was some variation in the wording of the education question, as police employees were asked to report their qualifications at the time of the survey, and military personnel at the time of joining service. However, during research advisory groups, it was suggested that it is unlikely that police officers would gain education as part of their service, whereas military personnel may gain qualifications as part of their training. Entropy balancing was conducted in STATA using the ebalance command (Hainmueller & Xu, 2013).

Estimating sample differences
Frequencies and percentages, with 95% confidence intervals (CI), were used to describe demographic, occupational and health variables for each sample. Descriptive statistics were also reported for the outcome variables, i.e. probable PTSD, alcohol consumption (non-drinkers, low risk, hazardous use and harmful use), binge drinking and comorbid probable PTSD and harmful alcohol consumption. The percentages were reported with entropy balance weights applied.
Stratifying by sample, logistic regressions (when PTSD and binge drinking were outcomes) and multinomial logistic regressions (when harmful alcohol use was the outcome, using low risk drinking as the reference group) were used to determine any associations between the demographic, occupational and health variables with PTSD and harmful alcohol use.
Logistic regressions were used to determine sample differences in probable PTSD and binge drinking. Multinomial logistic regressions determined sample differences in harmful alcohol use (low risk drinking as reference) and comorbidity of probable PTSD and harmful alcohol use (presence of neither as reference). Analyses were adjusted for variables hypothesized a priori to be associated with PTSD and harmful alcohol use: marital status and smoking status. Previous evidence shows that being married or in a relationship is protective against PTSD (Jakupcak et al., 2010) and harmful drinking (Prescott & Kendler, 2001), compared to those who are not in a relationship, whereas smoking has been found to be positively associated with PTSD (Fu et al., 2007) and harmful drinking (Room, 2004). We did not adjust for age and education as these variables were used to create the entropy balancing weight, though higher education is thought to be protective against PTSD and harmful drinking.
We conducted sensitivity analyses to explore whether the previously outlined associations differed if we restricted the police sample to inspectors, constables and sergeants, excluding police staff. Police staff may be less comparable to serving regular military personnel, as they are considered to have more desk-based duties compared to inspectors and police constables/sergeants.
Unadjusted and adjusted odds ratios or multinomial odds ratios, with 95% confidence intervals are reported. All statistical analyses were conducted in STATA SE 15.

Ethics
The Airwave Health Monitoring Study received ethical approval from the National Health Service multisite research ethics committee (MREC/13/NW/0588). Written informed consent was obtained from all participants. Ethical approval was obtained for each of the phases of the Health and Wellbeing Cohort Study from both the UK Ministry of Defence Research Ethics Committee and the local Ethics Committee at King's College London.

Sample characteristics
The total sample size was 31,255, including 23,826 police employees and 7,399 military Personnel (Table 1). The entropy balancing resulted in balanced estimates of age and education whereby approximately 75% of both samples were under the age of 40 years and approximately 57% of both samples had a higher educational attainment. About 80% of police employees were constables and sergeants. Over 55% of military personnel were non-commissioned officers (NCOs) and had been on deployment. Almost 25% of military personnel reported smoking, compared to just 10% of police employees.
Supplementary Table 1 compares the unweighted frequencies and percentages with the entropy balanced frequencies and percentages for the police sample. The weighted estimates for the outcomes of interest (i.e. PTSD, categories of alcohol consumption, binge drinking) were similar to the unweighted estimates. As expected, due to the covariatebalancing, the weighting increased the proportion of police employees under the age of 40 years and with lower educational attainment, similar to the military sample. The weighting lowered the proportion of inspectors and the proportion with a salary over £60,000 (See Supplementary Table 1).

Sample differences in probable PTSD and harmful alcohol use
For probable PTSD, 3.95% of police employees and 3.67% of military personnel met criteria. For harmful alcohol use, 2.87% of police employees and 9.59% of military personnel met criteria, with 1.50% of police employees and 3.04% of military personnel reporting daily or almost daily binge drinking. Military personnel were less likely to meet the criteria for probable PTSD compared to police employees (Adjusted Odds Ratio (AOR) 0.84, 95% CI 0.72 to 0.99), however this association was borderline significant only in the adjusted regression (Table 2). In contrast, military personnel were significantly more likely to report harmful alcohol use (AOR 2.79, 95% CI 2.42 to 3.21) compared to police employees and this is also reflected in their binge drinking behaviour (AOR 1.67, 95% CI 1.37 to 2.03). Military personnel were also more likely to report comorbid PTSD and harmful alcohol use (AOR 2.84, 95% CI 1.67 to 4.84), though this is likely driven by harmful alcohol use.
The sensitivity analyses whereby the military personnel were compared to inspectors and police constables/sergeants only, and police staff were dropped, did not reveal any differences in the associations found between PTSD and alcohol consumption characteristics between military personnel and police employees (Supplementary Tables S3 and S4).

Associations with probable PTSD and alcohol consumption
Among both samples, greater educational attainment was significantly associated with decreased odds of PTSD (police employees: Odds Ratio (OR) 0.80, 95% Confidence Interval (CI) 0.65 to 0.98; military personnel: OR 0.69, 95% CI 0.54 to 0.89) ( Table 3). Police employees aged ≥50 years (compared to those aged 30 to 39 years) had decreased odds of PTSD (OR 0.70, 95% CI 0.55 to 0.89). Among military personnel, being divorced/separated was significantly associated with increased odds of PTSD (OR 3.05, 95% CI 2.08 to 4.44,) and smoking (OR 1.95, 95% CI 1.51 to 2.53). Deployment reduced the likelihood of PTSD (OR 0.72, 95% CI 0.53 to 0.99) as well as holding a higher rank (NCOs OR 0.58, 95% CI 0.44 to 0.75, Commissioned Officers OR 0.30, 95% CI 0.19 to 0.46). Those in the Army were more likely to meet the criteria for PTSD, compared to those in the Royal Air Force or Naval Service. For both samples, higher educational attainment reduced the odds of harmful alcohol use, whereas smoking and being over 40 years of age (compared to those aged 30 to 39 years) increased the odds of harmful alcohol use (Table 4). Harmful alcohol use was about twice as likely among police employees in the top two income brackets. Military personnel who had deployed had 2.14 times the odds (95% CI 1.64 to 2.81) of harmful alcohol use, compared to those who had not deployed. Higher ranked military personnel were less likely to report harmful alcohol use compared to other ranks.
Interestingly, younger military personnel (<29) were more likely binge drink compared to those aged 30 to 39, but younger police employees were less likely to binge drink (Table 4). For both samples, binge drinking was less likely among those with higher educational attainment and more likely among current smokers. Police employees who reported to be single were less likely to binge drink (OR 0.34, 95% CI 0.19 to 0.61) compared to their married/cohabiting peers. In contrast, single, divorced and separated military personnel had an increased odds of binge drinking compared to their married/cohabiting counterparts. Military personnel who had deployed had a higher odds of binge drinking compared to those who had not deployed (OR 1.69, 95% CI 1.07 to 2.66).

Key findings
This is the first study to directly compare the proportions, and associated factors, of probable PTSD and harmful alcohol use, among male members of the Armed Forces and Police Forces. We found several key findings. First, similar proportions of probable PTSD (approx. 4%) in both samples. Second, military personnel reported higher proportions of harmful alcohol use than police employees (approx. 10% vs 3%, respectively). Third, comorbid PTSD and harmful alcohol use were more common in military personnel compared to police employees, driven by the higher proportion of harmful alcohol use. In military personnel, holding a lower rank, being in the Army, divorced or separated or a current smoker all increased the likelihood of probable PTSD. Interestingly, the risk of binge drinking was higher in the youngest military personnel but higher in older police employees. Higher educational attainment was   protective against probable PTSD, harmful alcohol use and daily binge drinking in both samples. Likewise, military personnel with lower ranks were more likely to drink harmfully. But contrarily, this was true for police employees with higher, not lower, salaries.

Proportions of probable PTSD, harmful alcohol use and binge drinking
Whilst there was a borderline significantly lower proportion of PTSD among military personnel than police employees, the proportions were similar (3.67% vs 3.95%). Both estimates are lower than observed in the UK general population (4.4%, using the PCL-C) (McManus et al., 2016) and lower than the most recent figures from the full cohort from which the military sample was derived (6.2%) (Stevelink et al., 2018). However, the current sample included only serving regular military personnel with the full military cohort showing higher proportions of probable PTSD in exserving and reserve personnel (Stevelink et al., 2018). Personnel with poor mental health are more likely to leave service and have poorer outcomes after leaving (Buckman et al., 2013;Iversen et al., 2005). We observed lower proportions of probable PTSD in police employees than other UK and international studies, as a recent study identified that 8% met criteria for PTSD (Brewin et al., 2020), and a systematic review estimated that 14% of police employees develop PTSD (Syed et al., 2020). However, these studies were based on occupational surveys, focussing on traumatic exposures, mental health and working conditions, which we have suggested causes a non-random bias from framing effects, increasing the reported prevalence of mental health disorders , whereas current data were obtained using a non-mental health focussed survey. Our finding that the proportions of probable PTSD in both the Armed Forces and Police Forces were lower than those in the general population is of interest. This may be because both use stringent employment selection procedures and medical screening (Royal Navy, 2016;Violanti et al., 2017), which may lead to a more resilient workforce, otherwise known as the 'healthy worker effect' (Li & Sung, 1999). A recent review identified that the rates of PTSD among police officers were lower than for civilians who experienced similar traumatic events (Regehr et al., 2019), indicating that police employees may be more resilient, following trauma-exposure (Andrew et al., 2008). This could be because police employees and military personnel are trained to operate in traumatic situations, therefore trauma is an expected part of their role and experienced collectively, rather than individually as with civilians, with evidence showing that team and supervisory support is protective of mental health, in military personnel (Jones et al., 2012). Further, there may be selection bias as those who experience poor mental health are more likely to leave service early (Buckman et al., 2013), or reporting bias as these occupational groups may underreport their symptoms due to stigma and barriers to care (Haugen, McCrillis, Smid, & Nijdam, 2017;Sharp et al., 2015).
We found that the proportions of harmful alcohol use, and daily binge drinking, were three times greater in military personnel compared to police employees, with similarly increased proportions of comorbid PTSD and harmful alcohol use. Approximately 10% of military personnel met criteria for harmful alcohol use whereas only 3% of police employees did so, which is similar to males in the UK general population (Digital, 2018). The higher proportions of harmful drinking in military personnel may relate to coping motivations for drinking, hence the higher proportions of comorbid PTSD and harmful alcohol use. Recent findings showed that military personnel who drink to cope are more likely to drink harmfully and binge drink (Irizar, Leightley et al., 2020). Alcohol use has historically been part of military culture, often used to create social bonds or to destress following deployment (Ames, Cunradi, Moore, & Stern, 2007;Jones & Fear, 2011). In addition, 'dry periods' during operational deployment may make it more difficult to manage alcohol use upon return (Goodwin et al., 2017;Jacobson et al., 2008;Stevelink et al., 2018), though deployment can be protective against dependence. Military organizations are referred to as 'greedy institutions', with extensive demands for commitment and dedication, leading to the development of a strong military identity and group cohesion, with traditions and rituals encouraging heavy drinking to increase social bonding within the unit (Hatch et al., 2013). Alcohol has also been seen as an integral part of police culture, through social ponding rituals or to cope with the stress of the job, as drinking with colleagues provides an opportunity for an informal debriefing of shared experiences (Abdollahi, 2002;Richmond, Kehoe, Hailstone, Wodak, & Uebel-Yan, 1999;Violanti, 2004). However, it is possible that the drinking culture has shifted in police employees, more so than military personnel, due to the 'greedy' nature of the military institution.

Factors associated with probable PTSD, harmful alcohol use and binge drinking
There were sample differences in the factors associated with harmful drinking and binge drinking. Police employees with higher salaries (indicating higher rank) were more likely to drink harmfully, and contrarily, military personnel holding a lower rank were more likely to drink harmfully. Moreover, the youngest police employees were less likely to binge drink compared to the older age groups, but the youngest military personnel were more likely to binge drink (consistent with the UK general population (Kuntsche, Kuntsche, Thrul, & Gmel, 2017)). The pattern of harmful drinking in police employees is similar to that of the general population, which shows that alcohol use is decreasing in younger people but increasing in older people (Bardsley et al., 2018;Oldham et al., 2020). Younger military personnel are more likely than those over 50 to report social pressure motivations for drinking (Irizar, Leightley et al., 2020) and may be more susceptible to military culture, which has historically facilitated risky drinking (Ames et al., 2007;Jones & Fear, 2011).
Smoking, which was more prevalent in military personnel than police employees (25% vs 10%), was only associated with probable PTSD in military personnel. Smoking was associated with harmful drinking and binge drinking in both samples, replicating findings from the full military cohort, though smoking rates have declined Thandi et al., 2015). In both samples, higher educational attainment was protective against all outcomes (PTSD, harmful drinking and binge drinking), in line with existing literature (A. C. Iversen et al., 2008;Jones et al., 2015;Tang, Deng, Glik, Dong, & Zhang, 2017). In military personnel, holding a higher rank and deployment were associated with a lower risk of PTSD, in congruence with previous findings from the first two phases of the cohort study, though deployment was associated with an increased risk of PTSD in the third phase of the cohort study (Stevelink et al., 2018), but this was more likely in reserves than regular serving personnel. The relationship between deployment and PTSD is complex and there are several interacting factors, such as whether personnel held a combat role and number of deployments Sundin, Fear, Iversen, Rona, & Wessely, 2010).

Strengths and limitations
This study utilized two large samples with good response rates (above 50%). Using entropy balancing, we increased the comparability of the samples by balancing covariate distribution of variables known to be associated with the outcome variables across all participants, unlike methods such as propensity matching. Further, we were able to harmonize several variables which had variations, including the categorization of alcohol consumption and daily binge drinking. Nevertheless, a previous study has shown that higher quantities of alcohol are reported in a weekly drinks diary, as used in our police sample, compared to quantity-frequency measures (e.g. AUDIT) as used in our military sample (Heeb & Gmel, 2005). However, our study included additional response options for our military sample to capture higher quantities (up to 30 drinks or more), preventing an upper limit on weekly drinks. Another caveat is that police employees only completed the TSQ if they experienced a traumatic stressor in the past 6 months, possibly missing participants with chronic or delayed onset PTSD (Utzon-Frank et al., 2014), and reducing the comparability with the military sample, where the timeframe of the PCL-C was not limited. In our pre-registered protocol, we stated that we would harmonize the measures of probable PTSD by selecting 10 items of the PCL-C which were most comparable to the TSQ, but due to different scoring approaches we were unable to do this, and so the full PCL-C and TSQ were used for the military sample and police sample respectively, though both scales show similar prevalence estimates in the UK general population (McManus et al., 2016). However, there are currently no formal comparisons of the TSQ and PCL-C, other than comparing prevalence estimates, so the validity of this approach is unknown. Despite these incompatibilities, the samples were carefully matched, both manually (i.e. selecting male serving regulars only) and statistically (i.e. balancing covariates), to increase comparability.

Implications
The current emphasis on ensuring support is available for both the Armed Forces and Police Forces should continue, through 'active monitoring' for those who have been exposed to trauma and further help being accessible to those who need it (Greenberg, Megnin-Viggars, & Leach, 2019). Further, a brief screen for harmful alcohol use should be integrated with mental health support, as those who use alcohol to cope may experience further mental health decline (Strid, Andersson, & Öjehagen, 2018), which is particularly important for military personnel, given the higher rates of harmful drinking and comorbid PTSD and harmful drinking.

Conclusions
We identified comparable proportions of probable PTSD in a covariate-balanced sample of male UK military personnel and police employees and much higher proportion of harmful alcohol use in military personnel, compared with police employees. The latter findings highlight a need for evidence-based interventions, such as brief alcohol screens, which are effective in civilian occupational settings (Hermansson, Helander, Brandt, Huss, & Rönnberg, 2010), in military occupational settings, more so than other occupational groups. Low proportions of PTSD were observed in both samples, which could be indicative of protective effects of unit cohesion and resilient workforces. These factors should be explored further.

Acknowledgments
This research has been conducted using the Airwave Study Tissue Bank Resource. We thank all participants in the Airwave Study for their contribution. We would also like to thank Paul Elliott, Gao He, Andy Heard and Maria Aresu from the Airwave Health Monitoring Study Research team for the provision of the data and data management support over the course of this project Contributions PI and SAMS contributed equally to the conceptualisation of the study, the design of the methodology, the statistical analysis and the writing of the original manuscript, as well as the reviewing and editing. DP contributed to the resources and data curation. SHG, NG and SW contributed to reviewing and editing the manuscript and contributed to the development of the formal analysis. NTF and LG contributed equally to the supervision of the research, the conceptualisation of the study and both provided extensive feedback on the formal analysis and written manuscript.

Data availability
The police employee data underlying the results presented in the study are available to researchers who apply for access to the Airwave Health Monitoring Study via the following URL: https://police-health.org.uk/applyingaccess-resource. The military personnel data is not publicly available.