Promoting Mental Health Recovery After Hurricanes Katrina and Rita: Title and subTitle BreakWhat Can Be Done at What CostMental Health Recovery After Hurricanes FREE

Hurricanes Katrina and Rita, which struck the US Gulf coast in late summer 2005, constituted the largest natural disaster in US history. Of the prestorm population of 11 million in the directly affected area, some 1.5 million (14%) were displaced from their homes; nearly 2000 deaths were directly attributed to the storms, along with countless injuries and unprecedented economic damage.^{1 - 2}

Large-scale disasters like the Gulf storms cause substantial mental health problems across all age groups, including increased incidence of posttraumatic stress disorder (PTSD), depression and anxiety disorders, exacerbation of preexisting mental disorders, and a variety of behavior problems. Moreover, the mental health problems that commonly arise after a disaster are leading causes of disability in adults and youth; thus, such problems are both clinically important and, if unaddressed, may impede recovery efforts and compound the social and economic fallout of the disaster.^{3 - 7}

Prior studies confirmed that the Gulf storms were followed by large increases in mental health problems in the affected population that have only slowly dissipated.^{8 - 10} Beyond the storms' unprecedented scale and scope, efforts to address the mental health consequences were complicated by the fact that the affected population included a high proportion of historically underserved groups such as low-income African American persons, by limited mental health–system capacity and substantial unmet need, even prior to the storms, and by various limitations in the short- and longer-term disaster response efforts.^{11 - 12}

The Gulf storms highlight the opportunity to strengthen mental health recovery following a large-scale disaster, particularly beyond the immediate postdisaster period of humanitarian response.¹⁰ In this article, we use the example of the Gulf storms to propose a new comprehensive model for an evidence-based mental health response to major disasters, quantify the potential costs and benefits of large-scale implementation of such a model, and consider its logistical and human resource requirements. We draw on existing scientific data on evidence-based mental health responses to disasters and epidemiological findings after disasters, supplemented by expert input owing to numerous gaps in the existing published literature. We focus on medium-term response, ie, from 7 months after disaster, when new mental disorders can be diagnosed, through 24 months, when more permanent delivery strategies might be considered (we report outcomes through 30 months). We did not attempt to address immediate postdisaster emergency relief, partly because relatively more plans/protocols exist for acute humanitarian response in this period (including crisis counseling), while far fewer strategies exist for the medium-term recovery period, and partly because the immediate postdisaster response may be more specific to the details of a particular disaster.

Many government agencies and nongovernmental institutions participate in short- and longer-term disaster response. For instance, the Substance Abuse and Mental Health Services Administration (SAMHSA) led federal efforts to disseminate mental health information and training in response to the Gulf storms, state and mental health agencies were responsible for public sector health services, private health and behavioral health plans and providers handled private sector services, and nongovernmental relief organizations, particularly the American Red Cross, provided many kinds of volunteer response.¹¹ We do not attempt to analyze these institutions' actual response to the Gulf storms here. Rather, we propose a comprehensive framework for delivering evidence-based mental health treatments that would follow the initial crisis response and that take into account feasibility, capacity, and population needs following a disaster.

We developed an intervention model that could be implemented after a disaster to identify people with substantial mental health problems and provide them with evidence-based treatment, supplementing the local provider capacity via telehealth and other national resources. We used a multiperiod decision analysis model to illustrate the potential costs and benefits of implementing the proposed interventions at different levels of intensity/coverage in the population affected by the Gulf storms.

We specified a discrete-time Markov model using 6-month periods, starting with 7 to 12 months after the disaster. In the model, appropriate treatment in a given period was determined by a person's clinical status at the start of the period, and outcomes were a function of a person's prior clinical status, treatment received, and response to treatment (or spontaneous recovery, for people who are sick but untreated in a given period). We modeled service use through 24 months after disaster and outcomes through 30 months, as outcomes are affected by treatment in prior periods.

Inputs into the decision analysis model include the characteristics and geographic distribution of the affected population, incidence/prevalence of mental illness, evidence-based treatment models, the costs of such treatment, mental health system capacity in the affected areas, and options for expanding mental health treatment capacity to address surges in need and demand.

Modeling is based on the prestorm (January 7, 2005) population of 11 million residents across the 117 counties/parishes across Louisiana, Mississippi, Alabama, and Texas that the Federal Emergency Management Agency (FEMA) designated as eligible for individual relief following hurricanes Katrina and Rita. Data on the prestorm population came from the Area Resources File and the US Census.

It seemed likely that intervention strategies would differ for urban vs rural areas because of health system differences and within vs outside storm-affected counties because of differences in the population prevalence of disaster-related morbidity. In practice, we allocated the prestorm population across 3 geographic areas for the poststorm period, based on data from the US Census as of January 1, 2006: urban affected areas (8.8 million), rural affected areas (2.0 million), and prehurricane residents of the affected areas that subsequently moved outside of the affected areas (0.24 million). Because treatment recommendations vary by age, we considered 3 age strata: 5 to 14 years (17% of the total population), 15 to 19 years (9%), and 20 years and older (75%).

We used 3 categories of morbidity: none, mild/moderate (ie, meeting criteria for a mental disorder, plus serious role impairment), and severe mental health problems (ie, disorder plus severe and/or multiple role impairment), in terms of needed treatment content and intensity, but not necessarily corresponding to specific disorders.

Our assumptions about incidence of mental health problems were based on available epidemiological data from prior disasters and on the storm-affected population,^{8 - 9 ,13 - 15} synthesized based on consensus within the research team. Specifically, more than 6 months after disaster we estimated that rates of persons with mild/moderate storm-attributable problems to be 25% of adults and 30% of children and rates of severe problems to be 5% of adults and 10% of children.^{16 - 18} Morbidity in subsequent periods was determined by our assumptions about treatment and recovery. Additionally, we assumed some continued new incidence in previously healthy people over time, specifically 5% of persons having mild/moderate and 1% having severe mental health problems after 12 months and 0.5% having mild/moderate mental health problems after 18 months across age groups.^{8 ,17}

We specified transition probabilities between each possible health state as a function of whether the person received the recommended treatment during the period. We assumed that treatment-outcome relationships were consistent with those reported in the literature, ie, that individuals in this situation were not more treatment-resistant, as we are not aware of evidence to this effect. We assumed lower, but not zero, probability of improvement or recovery in the absence of treatment. For simplicity, we assumed that recovery from illness was an absorbing state, ie, once recovered, people remained free of problems for the remainder of our study period. Parameter inputs are listed in the eTable.

We developed estimates of the unit costs of screening, assessment, and treatment services using information from public sources, particularly the Medicaid fee schedule in the storm-affected states, via personal communication with 2 nationally managed behavioral health organizations (MBHOs), and with several local hospitals within and outside the storm-affected areas. Overhead associated with health care services is typically incorporated into health care prices and thus is included implicitly in our unit costs.

Two MBHOs reviewed our proposed model to consider feasibility; one determined that implementation would be within current capacity, while the other would have to add or train telephone therapists.

We specified the intensity of implementation, ie, the probability that people would receive the recommended screening, assessment, and treatment. We made simplifying assumptions for model tractability. In particular, while we allowed for false positives in screening, we assumed zero false negatives in screening and assessment, and treatment was conditional on positive screening/assessment (ie, all who received treatment were sick). We assumed that screening for adults and children out of school would administered individually, but that there were economies of scale in screening schoolchildren, and that people were screened no more than once in the initial period, except for new incident illness. Finally, we assumed that each person received either the recommended treatment or no treatment in a given 6-month period.

We developed intervention strategies to identify people with substantial mental health consequences and provide them with evidence-based treatment. In terms of specific conditions, the disaster literature identified the most relevant as anxiety disorders, PTSD, and depressive disorders, but we also considered exacerbations of severe mental illness such as psychotic disorders. However, in keeping with our population focus—and for model tractability—we developed general treatment models rather than attempting to develop treatment recommendations that would necessarily suit every individual patient; thus, treatment profiles and associated costs should be viewed as population averages, which vary across individuals.

Postdisaster mental health treatment programs have focused on crisis interventions such as psychoeducation and psychological first aid during the immediate recovery period.¹⁹ Cognitive behavioral therapy (CBT) has been found to be effective in treating PTSD in adults and youth.^{18 ,20} A recent Institute of Medicine report²¹ concludes that there is adequate evidence among adults for efficacy of exposure therapy, often based on CBT principles, for treatment of PTSD. There is a large amount of literature on effective treatments for depressive disorders, but little of it addresses the postdisaster context. A growing number of studies suggest that treatment programs for depression relying on distant delivery strategies such as telephone counseling are effective.²² Such findings support the feasibility of our proposing innovative postdisaster delivery strategies that combine distant and local services delivery. Likewise, PTSD and depression have been effectively treated with CBT in youth with school-based interventions, which could be key resources for communities recovering from a disaster.^{18 - 20 ,23} Comprehensive programs to improve mental health care can improve outcomes for the poor and racial/ethnic minorities,^{24 - 25} the main underserved groups affected by hurricanes Katrina and Rita. To our knowledge, there has been no services delivery model that builds on this literature to support comprehensive mental health recovery after a disaster.

We specified 1-time screening for common disorders after a disaster using, as a prototype, 1 to 3 brief self-report screeners such as the Mental Health Inventory (MHI-5), K-6, or Patient Health Questionnaire (PHQ-9). We adjusted the mix of telephone and in-person screenings and the provider level according to age and geographic group. We assumed existing FEMA staff who, in practice, visited individuals receiving assistance following the Gulf storms could apply screeners in affected areas in months 7 through 12, when we assumed most screening would occur; however, our estimated costs of screening were largely independent of modality. We assumed that persons screening as positive for a mental health problem would require a clinical assessment of need by telephone or in person. We adjusted the mix of delivery mode (telephone or in-person) and provider type (masters, PhD, or MD) by age and geographic group.

We specified the provider type (masters, PhD, MD) and mix of services (inpatient, outpatient, psychotherapy, and medication) based on level of need, history of response to treatment in prior periods, and age and geographic group. For adults and adolescents, we thought that psychotherapy could be provided in person or by telephone, and adjusted the mix of modality by geographic area and need level for that period. For medication management, we assumed that providers were local and adjusted the mix of primary care and psychiatry by age and geographic group; we specified that telephone or local consultation by psychiatrists would be provided based on need, history of response in prior periods, and age and geographic group. We assumed that persons with severe mental illness faced a risk for hospitalization and required case management, which could be provided in person or by telephone, adjusting this modality mix by geographic group.

We assigned the amount (such as duration, number of prescriptions) of services required, assuming that treatment for mild/moderate illness was equivalent to about two-thirds of the requirements for a course of cognitive behavioral therapy (8-10 sessions) and/or several months of medication management, and that severe illness faced a known risk of a short-term hospitalization and required both psychotherapy and medication, more visits, consultation and/or supervision, and case management. In general, we specified higher volume of services and higher-level providers with more persistent need.

We focused our quantitative model on services that are provided to individuals. We address other types of resources that are applied at the program or population level in the discussion, particularly outreach/education campaigns, training programs for local and national providers, and management and coordination infrastructure.

Table 1 summarizes our recommended intervention model for adults in the urban/affected area. Information on other age groups and areas is available by request from the authors.

We focused on the following outcomes of interest:

We implemented Markov models by age and geographic group (9 analytic cells in all). We pooled results across the age and geographic groups. We estimated services and costs per 1000 persons (or per capita), the total services and costs required, and episodes averted for the complete population affected by the Gulf storms. We developed upper-limit estimates of pre-Katrina provider capacity in the affected counties from the Area Resources File.

We implemented our decision analysis model several times for different levels of intensity/coverage. Here we report results for the extremes of zero implementation, representing what would happen in the absence of any mental health intervention, and universal implementation, representing the maximum achievable outcomes implied by our recommended intervention.

We recognize that there is considerable uncertainty in many of the inputs to our model. While we implemented the model deterministically for tractability, we systematically varied the values of key parameters to gauge the sensitivity of the findings to changing assumptions.

Figure 1 illustrates the population distribution of mental health problems, given our assumptions. Figure 1A provides estimates of the time course of mild/moderate and severe mental health problems in the overall affected population under the assumption of no treatment; this can be thought of as the natural course of illness, or the upper bound of morbidity. Overall morbidity peaks around 33% (26% mild/moderate and 6% severe) and declines to about 15% (9% mild/moderate and 6% severe); prevalence of severe problems increases over the first year after disaster, in part owing to new incidence, but also because rates of spontaneous recovery (ie, without treatment) are assumed to be low and some untreated cases worsen.

Figure 1B provides comparable estimates under the assumption of 100% implementation of our recommended model. The area between the morbidity trajectories can be thought of as the maximum recovery achievable from active interventions; this is shown graphically in Figure 2 for mental health problems overall. Per 1000 persons, the shaded green area in Figure 2 suggests a reduction of 327 6-month episodes of mental illness (221 mild/moderate and 106 severe) during our 2-year analysis period. Put another way, full implementation would eliminate 34.9% of the episodes of illness that would arise with no treatment.

Table 2 lists the costs of implementing our model under 100% implementation (vs 0% treatment), overall and by type of service and time period. Full (vs no) coverage of our intervention model would cost $1133 per capita over 7 to 24 months after disaster, which corresponds to $12.5 billion across the overall storm-affected population of 11 million people. This suggests an average cost of $3460 per averted 6-month episode of mental health problems. (Because a given episode of treatment changes transition probabilities across multiple outcomes, we cannot separately estimate costs of averting mild/moderate or severe episodes.)

Table 3 shows the number of different types of services that would be required 7 to 12 months after disaster with 100% model implementation for the full storm-affected population. Each of the 11 million people in the affected population would be screened, and 4.8 million would have a clinical assessment. Full treatment implementation would involve provision of 15 million Master's degree–level and 8.5 million PhD-level therapy sessions, 6.6 million primary care visits, 5.4 million psychiatrist contacts, and 0.45 million nights of mental health–related hospital care during the 7- to 12-month period after disaster. Results for 13 to 24 months are available from the authors.

Measured in FTE positions, we estimate that full implementation would require approximately 7500 FTEs of Master's degree–level and 5000 of PhD-level therapists for in-person outpatient psychotherapy; 14 500 Master's degree–level and 8200 PhD-level FTEs overall (in-person and telephone) for outpatient care; 1700 psychiatrist FTEs for outpatient medication management; and 2500 hospital beds for inpatient psychiatric care (which would require additional psychiatrists, psychiatric nurses, and associated staff, beyond the estimated FTEs for outpatient care). By comparison, based on prestorm data from the 2004 Area Resources File, there were 14 245 social workers and 4647 psychologists (proxies for Master's degree– and PhD-level therapists, respectively), 1443 psychiatrists, and 1729 inpatient psychiatry beds across the storm-affected counties. The Area Resources File estimates are an upper bound, as not all providers are in full-time practice; some time would be committed to established patients and the storms' displaced providers and damaged hospitals.^{11 ,26}

In addition, successful population-level implementation would also require outreach efforts, provider training, particularly for care of children, and development of a management, communication, and accountability infrastructure. Based on general information on national campaigns such as the National Institute of Mental Health's Real Men, Real Depression program and local campaigns in New York after the terrorist attacks of September 11, 2001, we estimate that such overhead costs would increase total program costs by perhaps 2% to 3%, with 5% as an upper limit (which would correspond to $625 million). While these elements are clearly essential for any large-scale intervention to be effective, detailed estimation of their costs is beyond the scope of this study.

Based on the decision analysis framework described here, we estimated that universal provision of an evidence-based mental health intervention model to the population affected by a disaster such as the 2005 Gulf storms would cost approximately $1133 per capita for the affected population in months 7 to 24 after the disaster, with nearly half of this spending in months 7 to 12 owing to an initial surge in need. In turn, we estimated that the services purchased by this money would reduce the number of 6-month episodes of storm-attributable mental health problems by 35%, corresponding to a per capita average of 2 extra months spent free of mental illness for each person in the disaster-affected population.

Given the structure of our model, varying the overall level of implementation of the proposed interventions would reduce both costs and benefits in approximately equal proportion. For instance, 25% coverage (vs none) would cost $347 per capita over months 7 to 24 (30.6% of the cost of 100% coverage vs no coverage), and avert 102 six-month episodes of illness (31.0% of 100% coverage vs none). This corresponds to a cost per averted episode of $3416 vs $3460 for 100% coverage vs none; this difference is because under less than universal coverage, a smaller proportion of treatment episodes are provided to refractory patients. Similarly, if the base rate of treatment were 25%, moving to 100% yields a cost per averted episode of $3549, or $3535 to move to 70% coverage. Other permutations would likely shift this balance unfavorably; for instance, if some people who initiate treatment discontinued early, they would consume services and incur costs when they start treatment, but discontinue treatment before it is (fully) therapeutic.

Formal estimation of cost-effectiveness was outside the scope of this study. Prior studies have mapped episodes of mental illness to quality-adjusted life years to provide a general framework for considering cost-effectiveness,^{27 - 28} in particular, using evidence from methodological studies suggesting that depression reduces the value of a quality-adjusted life year by 0.2 to 0.4 (of a maximum value of 1).^{29 - 30} If we apply these scaling factors to episodes of mental illness in our model, our estimated cost per averted episode corresponds to an estimated cost per quality-adjusted life year of $17 301 to $34 603. Even the upper end of this range, to be conservative, is within the range of generally accepted medical practice such as regular screening for colorectal, prostate, and breast cancers in average-risk patients and blood pressure screening in normotensive people.^{26 ,31 - 33}

Our intervention cost and outcome estimates are likely to be upper bounds, for several reasons. Our reference population was no treatment (natural illness recovery), whereas usual care included some services. Yet this may not be out of range, as many affected people may choose not to receive mental health services even if offered them owing to stigma, cultural acceptability of services, competing needs, and other factors. The Gulf storms, in particular, disproportionately affected racial/ethnic minority individuals and low-income populations, which tend to have less access to and lower use and quality of mental health treatment.^{12 ,34}

This raises the issue of feasible implementation. Assuring the delivery of evidence-based mental health interventions can be challenging in the absence of a disaster. Moreover, there is evidence that demand—or at least clinical need—for mental health services exceeds the available supply in many parts of the country, even without surges in need following a disaster.³⁵ In the US public sector, mental health providers are primarily oriented to persons with severe and persistent mental illness such as schizophrenia, while private sector systems may be more familiar with disorders like depression that are common after disasters, but have limited experience caring for disadvantaged or displaced populations and often little mandate to do so. In addition, there is generally weak infrastructure for reliably delivering the types of psychotherapy such as CBT that are known to be effective in postdisaster situations.

Our findings suggest that population-level implementation would almost certainly exceed local provider capacity. While disaster preparedness may help motivate some expansion of local capacity, it is unlikely to be feasible or efficient for each geographic area to have adequate local reserve resources to meet postdisaster needs. Instead, response could draw on national reserve resources. Indeed, outside human and other resources played an important role in the response to the Gulf storms; yet, those resources were largely assembled after the storms had occurred and needs identified. Response to future disasters may be substantially enhanced if it could draw on a preestablished, national, ready reserve of providers trained in evidence-based treatments, along with a logistical infrastructure to deploy them effectively in person and via telehealth and to coordinate their work.

We emphasize that national need not mean public. In practice, along with the Veterans Administration and the Department of Defense, the largest existing networks of mental health providers in terms of both number of providers and geographic coverage are those of private MBHOs, which may also have relevant managerial and logistical capabilities and experience. Disaster preparedness is now considered a public good, and national mental health preparedness will likely require some federal sponsorship for financing as well as to establish the parameters for a national mental health response, develop the rules under which it would operate, activate/deploy the response infrastructure, and monitor outcomes.

Telehealth through available large managed care companies seems like a natural fit with the goal of achieving a nationally distributed network of reserve providers, and a growing body of research supports telephone psychotherapy as a viable delivery option.^{22 ,36 - 37} However, further research is needed to assess the effectiveness of telephonic mental health response in postdisaster settings and to identify operational requirements such as referral mechanisms, clinical supervision, outcome monitoring, and billing/reimbursement for population-based care.

One particular local supply issue we identified was a likely shortage of inpatient beds and associated clinical staff; addressing this shortage would require new structural resources such as psychiatric wards in field hospitals and/or evacuating patients to facilities outside the affected area that have extra capacity. Further research is also required to establish best practices for meeting these service delivery needs.

Policy changes are likely to be required to facilitate a 2-year response by providers from outside a disaster-affected area. In general, providers must be licensed in the state in which they are providing services or be federally certified (eg, employed by the Veterans Administration or the US Public Health Service), neither of which covers most potential responders from outside a disaster-affected area; the services can also be classified as nonprofessional or educational, neither of which is generally germane here. In the case of the Gulf storms, some of the affected states waived licensure requirements for some professions. In Louisiana, some waivers were maintained after the storms for provider groups that do not charge for services in the affected areas. However, these responses occurred somewhat ad hoc and mostly after the fact. Based on the experience following the Gulf storms, it may be appropriate to consider developing a standard national strategy to streamline licensing and malpractice issues, eg, by allowing providers licensed anywhere in the US to practice in any FEMA-designated disaster area for a specified period of time, along with a viable mechanism to provide malpractice coverage to providers who participate in disaster response. Similarly, modification or extension of policies affecting medical licensing and malpractice are likely to be needed to facilitate telecare, both to cover cross-state provision of care and to enable reimbursement.

Other methods to consider for increasing supply rapidly may include rapid retooling of providers for other kinds of health or social services, training and deployment of nonprofessional/lay providers, and/or developing a deeper local reserve of community leaders with relevant skills who could participate in responses to local community emergencies. Some such innovations might additionally strengthen community mental health resources in nondisaster circumstances, a dual use that has been emphasized in other areas of preparedness.

Until the recent primary care authorization for New Orleans that potentially includes support for behavioral health services, there was no specific federal allocation for mental health services in response to the Gulf storms. This raises the question of the priority that should be given to mental health response, given many competing recovery needs. The Gulf storms have dramatically illustrated the scope and persistence of mental distress, which cuts across age and cultural groups and is likely to impede many aspects of both individual and community recovery, even as research has shown economic benefits from mental health interventions.³⁸

We have focused on mental health recovery over 7 to 24 months after disaster, and additional strategies may be required to promote longer-term improvements subsequently. For example, while chronic disease management programs for depression improve outcomes, disseminating and sustaining these interventions outside of a disaster has been difficult, as key components are often not covered by insurance policies. Similarly, many schools were found to have adequate crisis response plans after Katrina, but few had the resources to sustain provision of mental health services.³⁹ Yet even investing in short-term interventions may have long-term health benefits over many years for disadvantaged population groups.⁴⁰ Thus, we currently face the option of proactively developing a plan to intervene quickly following disasters beyond current efforts, most notably crisis counseling to promote mental health recovery for survivors; we wonder whether this would facilitate recovery in other life domains and suggest that it is important to determine how to promote long-term recovery for individuals and communities.

More generally, we have taken a largely medical perspective on ameliorating the mental health consequences of disasters. Comprehensive disaster response requires intervention across multiple domains, including short- and long-term efforts to ensure survivors' mental, physical, social, and economic needs. While there are likely to be powerful synergies across these domains, we recognize that the optimal mix of interventions across them remains an open question. We necessarily leave this to future research.

A key related issue is when and for whom to activate this type of intervention program. In principle, the framework described here could be applied for disasters of varying scope and scale. In practice, the intervention details will vary by the scope, scale, nature, and consequences of the disaster, eg, depending on the extent of population displacement and on how the needs in an area exceed its available and/or remaining delivery capacity.

Our analyses have important limitations. There are many knowledge gaps regarding evidence-based response to disasters, and a recent report by the Institute of Medicine highlighted concerns about insufficient evidence of efficacy for most PTSD treatments among adults.²¹ Our models could not include data on individual differences in response to traumatic events, which are key predictors of outcome, so we present a population rather than individual perspective on services that promote recovery. There is some degree of uncertainty around each of the model's many inputs, and thus our overall findings. We focused on the context of the Gulf storms, and estimates for other disasters could require different assumptions and yield different results, but we designed the treatment and per capita services model to be applicable across a wide range of contexts.

Given the recent evidence of the high burden of disease, significant unmet need among the survivors of the Gulf storms, and widespread discussion of disaster preparedness more generally, we hope that the response model proposed here may be a useful starting point for policy discussions to improve services for people with persistent illness following the Gulf storms and to plan a coordinated response strategy for future disasters.

Correspondence: Michael Schoenbaum, PhD, Division of Services and Intervention Research, National Institute of Mental Health, 6001 Executive Blvd, Room 7142, MSC 9629, Bethesda, MD 20892-9669 (schoenbaumm@mail.nih.gov).

Submitted for Publication: July 3, 2008; final revision received December 10, 2008; accepted December 29, 2008.

Financial Disclosure: None reported.

Funding/Support: This study was supported by National Institutes of Mental Health grant P50MH54623; the RAND Corporation; the Robert Wood Johnson Foundation; and the South Central Veterans Affairs Mental Illness Research and Education Center.

Disclaimer: The views expressed in this article do not necessarily represent the views of the National Institute of Mental Health, the National Institutes of Health, the Department of Health and Human Services, or the United States Government.