Epidemiological model estimates

Prevalence estimates of major depression and PTSD in conflict-affected populations, used as inputs in this study, have been previously developed using systematic review and Bayesian meta-regression methods. Methods have been described in full elsewhere (Charlson et al. Reference Charlson, Flaxman, Ferrari, Vos, Steel and Whiteford2015). In brief, a systematic review of the literature was conducted using PRISMA guidelines (Liberati et al. Reference Liberati, Altman, Tetzlaff, Mulrow, Gotzsche, Loannidis, Clarke, Devereaux, Kleijnen and Moher2009) to collate epidemiological estimates of major depression and PTSD cases which met criteria for diagnosis as per the Diagnostic and Statistical Manual of Mental disorders (DSM) or the International Classification of Diseases (ICD) (World Health Organization, 1992; American Psychiatric Association, 2000). Electronic database searches were conducted using Medline (PubMed), PsychINFO, Embase, Published International Literature on Traumatic Stress (PILOTS), African Journal Online (AJOL), SciELO Public Health and ProQuest digital dissertations.

‘Conflict-affected’ countries were defined as meeting conditions set by the Uppsala Conflict Data Program (UCDP) database (Uppsala Conflict Data Program, 2012) and/or a level of 4 or 5 on the Political Terror Scale (PTS) (Political Terror Scale, 2011). To limit our populations to those either in conflict or within the 10-year post-conflict period, conflict status had to be met within the 10-year period prior to data collection. Studies were required to capture conflict-affected populations who remain within their country of origin or have been relocated to another neighbouring and/or non-Western country. Study samples not representative of the general population were excluded. Accepted diagnostic instruments were required to follow either ICD or DSM diagnostic criteria for major depression and PTSD.

DisMod-MR, a Bayesian meta-regression tool designed specifically for GBD 2010, was used to develop predictive estimates of age-specific prevalence in conflict-affected populations, whilst simultaneously adjusting estimates for heterogeneity between studies (Ferrari et al. Reference Ferrari, Charlson, Norman, Flaxman, Patten, Vos and Whiteford2013a , Reference Ferrari, Charlson, Norman, Patten, Freedman, Murray, Vos and Whiteford b ). DisMod-MR generates estimates through a negative-binomial model that is fitted using a randomised Markov-Chain Monte Carlo algorithm (Vos et al. Reference Vos, Flaxman, Naghavi, Lozano, Michaud, Ezzati, Shibuya, Salomon, Aboyans and Aggarwal2012a ; Flaxman et al. Reference Flaxman, Vos and Murray2013). These modelled estimates were age-standardised against the population of Syria for computations in this paper (United Nations, 2015).

Prevalent cases

Given the widespread and protracted conflict taking place in Syria we assume that the entire population of Syria has been exposed to traumatic war-related events at some stage. Therefore, to derive estimated prevalent cases of major depression and PTSD, prevalence estimates were multiplied by the total population of Syria as documented by the UN World Population Data (United Nations, 2015). Prevalence estimates are drawn from a previously published meta-regression (Charlson et al. Reference Charlson, Flaxman, Ferrari, Vos, Steel and Whiteford2015) and age-standardised for the Syrian population distribution.

The proportion of PTSD cases comorbid with depression derived from general population and post conflict settings was set at 50% based on documented rates including a sample of Bosnian refugees living in Croatia (Mollica et al. Reference Mollica, Mcinnes, Sarajlić, Lavelle, Sarajlić and Massagli1999; Chapman et al. Reference Chapman, Mills, Slade, Mcfarlane, Bryant, Creamer, Silove and Teesson2012).

Burden of disease

As part of the Global Burden of Disease Study 2010 (GBD2010), burden of disease estimates are produced for 20 mental and substance use disorders (Whiteford et al. Reference Whiteford, Degenhardt, Rehm, Baxter, Ferrari, Erskine, Charlson, Norman, Flaxman and Johns2013). The core metric of health loss is the DALY which is made up of both mortality (yYears of life lost; YLL) and morbidity (YLD) components (Murray et al. Reference Murray, Vos, Lozano, Naghavi, Flaxman, Michaud, Ezzati, Shibuya, Salomon and Abdalla2012b ). YLDs from a sequela are equal to the prevalent cases of the sequela multiplied by the disability weight for the health state associated with that sequela (Vos et al. Reference Vos, Flaxman, Naghavi, Lozano, Michaud, Ezzati, Shibuya, Salomon, Abdalla, Aboyans, Abraham, Ackerman, Aggarwal, Ahn, Ali, Alvarado, Anderson, Anderson, Andrews, Atkinson, Baddour, Bahalim, Barker-Collo, Barrero, Bartels, Basáñez, Baxter, Bell, Benjamin, Bennett, Bernabé, Bhalla, Bhandari, Bikbov, Abdulhak, Birbeck, Black, Blencowe, Blore, Blyth, Bolliger, Bonaventure, Boufous, Bourne, Boussinesq, Braithwaite, Brayne, Bridgett, Brooker, Brooks, Brugha, Bryan-Hancock, Bucello, Buchbinder, Buckle, Budke, Burch, Burney, Burstein, Calabria, Campbell, Canter, Carabin, Carapetis, Carmona, Cella, Charlson, Chen, Cheng, Chou, Chugh, Coffeng, Colan, Colquhoun, Colson, Condon, Connor, Cooper, Corriere, Cortinovis, De, CCARO, Couser, Cowie, Criqui, Cross, Dabhadkar, Dahiya, Dahodwala, Damsere-Derry, Danaei, Davis, De Leo, Degenhardt, Dellavalle, Delossantos, Denenberg, Derrett, Des, RLAIS, Dharmaratne and Dherani2012b ).

Within a particular sequela, GBD considers there are several health states (e.g. mild, moderate and severe major depression). Each health state has its own disability-weight – a quantification of the severity of health loss associated with a health state on a scale from 0 to 1, when 0 is commensurate with perfect health and 1 is commensurate with death. In the GBD 2010, disability weights for health states are determined based on survey respondents representing the general public (Salomon et al. Reference Salomon, Tandon and Murray2004). Comorbidity adjustments are applied to sequela-specific disability weights in the calculation of YLDs. YLDs for a disease or injury are the sum of the YLDs for each health state (e.g. mild, moderate and severe depression) associated with the disease or injury. More detailed information on GBD methodology has been found elsewhere (Murray et al. Reference Murray, Vos, Lozano, Naghavi, Flaxman, Michaud, Ezzati, Shibuya, Salomon, Abdalla, Aboyans, Abraham, Ackerman, Aggarwal, Ahn, Ali, Alvarado, Anderson, Anderson, Andrews, Atkinson, Baddour, Bahalim, Barker-Collo, Barrero, Bartels, Basáñez, Baxter, Bell, Benjamin, Bennett, Bernabé, Bhalla, Bhandari, Bikbov, Abdulhak, Birbeck, Black, Blencowe, Blore, Blyth, Bolliger, Bonaventure, Boufous, Bourne, Boussinesq, Braithwaite, Brayne, Bridgett, Brooker, Brooks, Brugha, Bryan-Hancock, Bucello, Buchbinder, Buckle, Budke, Burch, Burney, Burstein, Calabria, Campbell, Canter, Carabin, Carapetis, Carmona, Cella, Charlson, Chen, Cheng, Chou, Chugh, Coffeng, Colan, Colquhoun, Colson, Condon, Connor, Cooper, Corriere, Cortinovis, De, CCARO, Couser, Cowie, Criqui, Cross, Dabhadkar, Dahiya, Dahodwala, Damsere-Derry, Danaei, Davis, Leo, Degenhardt, Dellavalle, Delossantos, Denenberg, Derrett, Des, RLAIS, Dharmaratne, Dherani, Diaz-Torne, Dolk, Dorsey, Driscoll, Duber, Ebel, Edmond, Elbaz, Ali, Erskine, Erwin, Espindola, Ewoigbokhan, Farzadfar, Feigin, Felson, Ferrari, Ferri, Fèvre, Finucane, Flaxman, Flood, Foreman and Fo2012c ; Vos et al. Reference Vos, Flaxman, Naghavi, Lozano, Michaud, Ezzati, Shibuya, Salomon, Abdalla, Aboyans, Abraham, Ackerman, Aggarwal, Ahn, Ali, Alvarado, Anderson, Anderson, Andrews, Atkinson, Baddour, Bahalim, Barker-Collo, Barrero, Bartels, Basáñez, Baxter, Bell, Benjamin, Bennett, Bernabé, Bhalla, Bhandari, Bikbov, Abdulhak, Birbeck, Black, Blencowe, Blore, Blyth, Bolliger, Bonaventure, Boufous, Bourne, Boussinesq, Braithwaite, Brayne, Bridgett, Brooker, Brooks, Brugha, Bryan-Hancock, Bucello, Buchbinder, Buckle, Budke, Burch, Burney, Burstein, Calabria, Campbell, Canter, Carabin, Carapetis, Carmona, Cella, Charlson, Chen, Cheng, Chou, Chugh, Coffeng, Colan, Colquhoun, Colson, Condon, Connor, Cooper, Corriere, Cortinovis, De, CCARO, Couser, Cowie, Criqui, Cross, Dabhadkar, Dahiya, Dahodwala, Damsere-Derry, Danaei, Davis, De Leo, Degenhardt, Dellavalle, Delossantos, Denenberg, Derrett, Des, RLAIS, Dharmaratne and Dherani2012b ; Whiteford et al. Reference Whiteford, Degenhardt, Rehm, Baxter, Ferrari, Erskine, Charlson, Norman, Flaxman and Johns2013).

We used GBD methodology and disability weights to estimate YLDs for major depression and PTSD in conflict-affected populations (Table 1). As GBD does not assess PTSD we have used disability-weights associated with overarching group of all anxiety disorders as proxy values. We used Ersatz (version 1.31, Brisbane, Australia; available at: http://www.epigear.com), an add-in that allows bootstrapping in Excel (Microsoft Corporation, Redmond, WA), to propagate uncertainty from the model input parameters through to the final YLD estimates (EpiGear International, 2012).

Table 1.

GBD 2010 severity splits and disability-weights

Mental health service requirements

Resource requirements to provide recommended packages of care for PTSD and depression were modelled using the WHO mhGAP costing tool (Chisholm et al. Reference Chisholm, Burman-Roy, Fekadu, Kathree, Kizza, Luitel, Petersen, Shidhaye, De Silva and Lund2015). This tool utilises an Excel spreadsheet to calculate numbers of full time equivalent (FTE) staff, inpatient beds, bed days, and outpatient visits required to deliver basic packages of care for priority mental, neurological and substance abuse disorders. The mhGAP costing tool recommends several packages of care for each disorder, developed based on a systematic review of evidence and international consultation (World Health Organization, 2008). For depression, these include basic psychosocial treatment, advice and follow-up for moderate-to-severe cases; anti-depressant medication for moderate-to-severe cases; intensive psychosocial intervention for a subset of moderate-to-severe cases; and psychosocial care for perinatal depression. The package recommends that treatments for depression be delivered through a mix of primary health care, ancillary, outpatient and inpatient services. mhGAP does not include anxiety therefore we chose to apply the depression care packages and exclude inpatient treatment. The rationale for this is discussed later in the methods.

The calculation of resource requirements required data on the number of people eligible to receive a mhGAP intervention package. In our model, we assumed that these intervention packages were delivered to people older than 15 years who had a moderate–severe case of depression and/or PTSD. Epidemiological inputs for these calculations included: (a) the number of prevalent cases of PTSD, depression, and comorbid PTSD and depression in Syria, calculated as described earlier; and (b) the proportion of moderate and severe cases of depression (28%) and PTSD (30%), sourced from the GBD 2010 study (Table 1). Additional assumptions were made by the mhGAP tool to calculate the population eligible to receive the perinatal package of care (refer to online Table S1 in the appendix). The eligible population in this case was the proportion of all women in the population aged 15–44 years who gave birth during the year and had moderate–severe perinatal depression. We estimated a fertility rate of 11% based on United Nations Populations data (UNdata, 2012), that 15% had perinatal depression (O'hara & Swain, Reference O'hara and Swain1996; Fisher et al. Reference Fisher, Mello, Patel, Rahman, Tran, Holton and Holmes2012), and that the proportion of moderate–severe cases corresponded with the percentage for depression quoted above.

Each package of care includes a treatment coverage target (i.e. the proportion of prevalent cases to receive that package of care). Target coverage for 2030 was drawn from the default settings in the mhGAP costing tool (Table 2). A conservative baseline coverage rate of 1% was estimated from limited data available on current mental health service delivery in Syria (World Health Organisation, 2nd Quarter, 2014a , b ). Although there are no clear signs of the conflict in Syria coming to an end we elected, for demonstration purposes, to use a hypothetical baseline year of services scale-up to be 2015.  To scale up the packages of care over 15 years from baseline to target coverage rates, two versions of the resourcing model were calculated. The first assumed an exponential scale up (i.e. small rate of scale-up initially with a more rapid rate of scale-up in the final years), while the second assumed a linear scale-up, where resources were increased at a constant rate over time.

Table 2.

Recommended packages of care and coverage targets

^a Anti-depressant medication = Fluoxetine, 20 mg tab; Amitriptyline, 50 mg tab.

Within each mhGAP package of care multiple treatment types may be included, such as primary care visits, ancillary care visits, and hospital inpatient and/or outpatient care. For each treatment type, the tool delineates the proportion of people in the care package who receive that treatment type, the number of visits or inpatient days required per person, and the number of minutes per visit. Further care package parameters are detailed in the online supplementary table.

Anxiety disorders are not included in the mhGAP costing tool. As evidence-based treatments for PTSD and major depression are similar (National Institute for Health and Care Excellence, 2009), the care packages for moderate to severe depression (excluding perinatal depression) were assigned for the treatment of moderate to severe PTSD. However, it was assumed that no inpatient treatment would be required for PTSD (National Institute for Health and Care Excellence, 2005). Prevalent cases of comorbid depression and PTSD were assigned treatment packages for depression, which include some inpatient treatment, as these cases are likely to require more intensive intervention.

The number of visits required for primary care, ancillary care and outpatient care in each care package were calculated as the product of the eligible population, the treatment coverage target, and the average number of visits per person. The number of bed days required for inpatient care in each care package was calculated as the product of the eligible population, the treatment coverage target, the percentage use and the number of bed days per person.

The numbers of visits or bed days were summed to provide a total for each treatment type and disorder. These were used to calculate the number of FTE staff required to deliver the recommended packages of care at the target coverage rate for each year. The mhGAP costing tool provides HR parameters for the following mental health provider types: psychiatrist, other physician/doctor, nurse, psychologist, other psychosocial worker and other provider/worker. For each provider type, the tool details the number of days worked per year, the average number of consultations per day for each community/outpatient treatment type, the average duration of a visit for each community/outpatient treatment type, and for inpatient care the number of staff per 25 bed ward. For each treatment type, the proportion of care delivered by each provider type is also detailed (see online supplementary table for full staffing parameters).

For each provider type, the number of FTE community/outpatient staff was calculated as the total number of visits in each setting multiplied by the proportion of care for that setting delivered by the relevant provider, divided by the number of consultations per year (number of consultations per day multiplied by the number of days worked per year). The number of inpatient beds was calculated as the number of bed days divided by 365 days per year, multiplied by 1.15 to assume 86% occupancy. For each provider type, the number of FTE inpatient staff was calculated as the number of beds divided by 25 beds per inpatient ward, multiplied by the number of that provider type per inpatient ward.

Avertable burden with increasing treatment coverage

The following section outlines the steps used to calculate the burden of depression and PTSD that could be averted by scaling up mhGAP intervention packages in Syria between the years 2015–2030. A population model was constructed to calculate the change in DALY burden that would occur in a post-conflict country (Syria) under four intervention scenarios: (1) no treatment; (2) current coverage; (3) 30% target coverage; and (4) 100% target coverage. The ‘no treatment’ scenario calculated the DALYs arising under a hypothetical ‘partial null’ scenario where all existing mental health services are removed. The current coverage scenario calculated the DALYs that occur with the existing coverage rate (i.e. 1%). The third scenario analysed the DALYs arising from a gradual scale-up of intervention coverage from a baseline of 1% to a final target coverage of 30%. Finally, the fourth scenario analysed the DALYs that would occur if intervention packages were gradually scaled up to an ideal target coverage of 100% in the year 2030. The DALY burden in our model is reflective of the YLD burden given the low levels of case fatality attributable to depression and anxiety. As in the mhGAP analysis, we evaluated the effect of assuming a linear or exponential scale-up for each intervention scenario.

A multiple cohort Markov model was constructed in Excel to model the disease dynamics of several multi-age cohorts in the 2015 Syrian population (Habbema et al. Reference Habbema, Boer, Barendregt, Killewo, Heggenhougen and Quah2010) (refer to online figure S1 in appendix). Our model was based on a previous disease model, DisMod 2 (Barendregt et al. Reference Barendregt, Van Oortmarssen, Vos and Murray2003), which simulates how a population cohort moves between three health states (healthy, diseased and dead) on a year-to-year basis. Transitions between each health state correspond with defined epidemiological parameters (i.e. incidence, remission, case fatality and other cause mortality). Our model calculated the year-to-year transition of a multi-age cohort between each health state over a 15-year time horizon, which is in keeping with the time frame used in the mhGAP tool and is similar to the 10-year time period used in previous health economic studies (Edejer et al. Reference Edejer, Baltussen, Adam, Hutubessy, Acharya, Evans and Murray2012). The reference year of the analysis was 2015.

Multi-age cohorts were based on age groups contained in the mhGAP tool above 15 years of age – i.e. 15–29, 30–44, 45–59, 60–69, 70–79 and 80+ years. The mid-age population for each age cohort was used to model each cohort over the 15-year time horizon. For instance, the mid-age of the 15–29 year population is 22 years. So the population for the 15–29 year age cohort experienced the intervention over fifteen years using the mortality and disease parameters that applied to the ages 22–37 years. The next 30–44 year cohort had a mid-age of 37 years and was modelled between the ages 37–52 years and so on.

Epidemiological input data

Input parameters were needed for the ages between 0 and 100 years to model the epidemiological experience of each age cohort over time. Population data were taken from UN population statistics. Mortality data for the 2015 Syrian population was obtained from GBD 2010.

DisMod 2 software was used to estimate a complete set of internally consistent disease parameters for PTSD and comorbid depression/PTSD in Syria. Prevalence data were obtained from the GBD 2010 study. A remission rate of 1.40 person-years was used for depression based on an average annual duration of 37.7 weeks. A remission rate of 0.17 person-years was used for PTSD based on the proportion of remitted cases obtained from a previous systematic review (Morina et al. Reference Morina, Wicherts, Lobbrecht and Priebe2014). The relative risk (RR) of mortality among depression cases was 1.9, based on a previous systematic review by Baxter 2011. The case fatality rate for PTSD was assumed to be zero at all ages. Entering the aforementioned input parameters into DisMod 2 allowed us to calculate internally consistent parameters on the incidence, prevalence, case fatality and remission for both depression and PTSD, respectively. Parameters were derived for 1-year age groups between the ages 0–100 years. These calculated parameters were, in turn, used as inputs for the multi-cohort Markov model.

Intervention package details

We derived a set of intervention packages for the intervention scenario based on the intervention packages used in the mhGAP costing tool. Intervention packages, once again, targeted moderate to severe cases of depression and PTSD. The effect sizes for each intervention (expressed as the % change in remission and/or disability weight) were obtained from the previous work (Chisholm et al. Reference Chisholm, Sekar, Kumar, Saeed, James, Mubbashar and Murthy2000; World Health Organization, 2006; Chisholm & Saxena, Reference Chisholm and Saxena2012). The effectiveness of each intervention package was calculated by multiplying the efficacy of an intervention (e.g. % reduction in remission) by the total coverage (i.e. target population coverage multiplied by the scale-up coverage). The annual scale-up coverage was determined by the nominated intervention scenario. For example, scale-up coverage was constant over 15 years at 0% for the partial null and 1% for the current coverage scenarios. In the third and fourth intervention scenarios, the scale-up coverage (linear and exponential) increased from a baseline of 1% to a target coverage of 30 and 100%, respectively.  The combined effectiveness of each intervention package for each disorder was calculated by using the following equation (Edejer et al. Reference Edejer, Baltussen, Adam, Hutubessy, Acharya, Evans and Murray2012):

(1.1) $$\eqalign{{\rm Combined}\,{\rm effectiveness} & = (1 - c_1^{\ast} e_1) + (1 - c_2^{\ast} e_2) \cr &\quad + \cdots + (1 - c_n^{\ast} e_n)} $$

where c = coverage rate; e = effectiveness; and n = total number of interventions.

Two parallel life tables were constructed using the DisMod 2 equations to compare the effect of scaling up these intervention packages over a 15-year time horizon (i.e. intervention scenario) v. a no intervention scale-up scenario (i.e. comparator). This was done separately for PTSD and comorbid depression/PTSD to calculate the DALYs averted when interventions for these diseases are scaled up, respectively. Ersatz was used to implement the programming for the multi-cohort Markov model.

Sensitivity analysis

Our baseline analysis did not apply a discount rate to account for the time preference of health benefits; where a benefit received now is worth more than a benefit received in the future. We thus conducted a sensitivity analysis to evaluate the impact of applying a 3% annual discount rate to the model.