Improving Behavioral Health Care for U.S. Army Personnel

To achieve the goal of psychological readiness for every soldier, the U.S. Army aims to provide the highest-quality behavioral health (BH) care possible. In 2009, the Army recognized a need to centralize BH care to improve coordination and standardization across the service. A key change was integrating BH providers into primary care settings, where they provide support to soldiers with BH conditions through expert consultations, clinical assessments, triage, and brief cognitive behavioral interventions (Hoge et al., 2015). This change has improved access and continuity of care, as well as enhanced communication among primary care providers, BH providers, and unit leaders (Hoge et al., 2015).

To support these efforts, the Office of the Surgeon General, U.S. Army, asked RAND Arroyo Center to identify factors that are associated with changes in outcomes for soldiers who receive Army BH specialty care and to develop recommendations to improve BH care and soldier outcomes.

We identified soldiers diagnosed with posttraumatic stress disorder (PTSD), depression, or anxiety who received BH specialty care and whose outcomes—PTSD, depression, or anxiety symptoms—were assessed during their care. Using treatment data on both outpatient and inpatient care delivered by the Military Health System (MHS) and symptom data collected through an online system that allows for collection of multiple patient and clinician-reported measures, called the Behavioral Health Data Portal (BHDP), we conducted analyses to identify predictors of changes in symptoms (including patient and treatment characteristics) and patterns in symptom trajectories.

Study Methods

Sample Selection

We identified three samples of active-component soldiers who received specialty BH care for PTSD, depression, or anxiety from the MHS between January and September 2016. We allowed a nine-month time frame for sample entry for a new treatment episode—defined as no specialty BH care associated with the diagnosis in the prior six months. We limited our samples to soldiers with a minimal level of symptom severity as measured by the PTSD Checklist for DSM-5 (PCL-5), the nine-item Patient Health Questionnaire (PHQ-9), or the seven-item Generalized Anxiety Disorder scale (GAD-7) (which measure PTSD, depression, and anxiety symptoms, respectively) and those with a second symptom score one to six months after the initial score. Soldiers who may have later separated or deployed but were otherwise eligible by our study criteria were included in the samples. This sample selection followed the approach that the Army uses for its symptom-based outcome monitoring.

Using the sample selection criteria, we identified 3,264 soldiers for the PTSD sample, 3,801 for the depression sample, and 4,282 for the anxiety disorder sample. The three samples were not selected to be mutually exclusive, so it was possible for a soldier to be in multiple samples. The selection criteria likely restrict generalizability, and samples may not be representative of all service members with PTSD, depression, or anxiety.

Data Sources

We focused on administrative treatment data on care provided to active-duty soldiers between December 1, 2015, and June 30, 2017. These files included records on all inpatient and outpatient health care provided by the MHS, including care provided in military treatment facilities (MTFs; direct care) and care provided by civilian providers and paid for by TRICARE (purchased care).

Our analyses relied on symptom score data from three self-report measures, collected using BHDP: the PCL-5 for PTSD, the PHQ-9 for depression, and the GAD-7 for anxiety. These tools are used to document patient symptoms, response to treatment, and remission for patients with initial scores above a defined threshold. We linked each PCL-5, PHQ-9, and GAD-7 score to the administrative data records for soldiers in our diagnostic samples.

We also consulted the literature to identify variables that could affect treatment outcomes. We identified patient and treatment characteristics that have predicted behavioral health treatment outcomes in patients with PTSD, depression, or anxiety and were available in administrative or BHDP data. We grouped these predictors into two categories: pretreatment variables (e.g., demographic characteristics, military service characteristics, comorbid diagnoses and symptoms, and use of health care before the initial elevated score) and treatment variables (e.g., psychotherapy, individual therapy, group therapy, provider characteristics, evaluation and management visits, and medications). Using this process, we defined a total of 57 pretreatment variables and 84 treatment variables, including three covariates that looked at the timing of the initial score relative to the intake visit.

Analyses

Our study consisted of three primary analyses. In the first, we sought to identify patterns in measure completions in BHDP, including how and when measures specific to PTSD, depression, and anxiety symptoms were completed, as well as the amount of time between a soldier's initial score and final follow-up score, the frequency with which soldiers completed outcome measures, and the relationship between the number of BH specialty visits that soldiers received and the number of scores completed. This analysis allowed us to describe the relationship between the number of completed symptom measures and the number of BH specialty visits in the six months after the initial score. We also examined the Army's use of BHDP data to compute and monitor rates of response and remission for the three target conditions.

The second analysis explored the representativeness of our selected samples. These analyses compared characteristics and health care utilization of soldiers in our multivariate analysis samples with a group of soldiers who received at least one direct care BH specialty visit for a target diagnosis but who were excluded for not meeting other eligibility criteria (e.g., soldiers who were not starting a new episode of treatment, whose symptoms were not severe enough at the initial score, or who did not have a second score in one to six months). We evaluated how each study sample differed from the broader one-visit group in terms of demographic characteristics (gender, age, and race/ethnicity), service characteristics (pay grade and deployment history), and health care utilization characteristics (outpatient visits and the rate of co-occurring conditions). We performed chi-square tests (for categorical variables) and t-tests (for continuous variables) to examine whether these variables differed significantly between each sample and our control group.

Our third analysis consisted of two parts. We first sought to identify predictors of outcome for soldiers with PTSD, depression, and anxiety—defined as change in outcome score from the initial score to the last observed outcome score (i.e., last score minus initial score). Developing these models involved a nine-step analytical process that allowed us to identify predictors of PTSD, depression, and anxiety symptom score change. Then, we conducted exploratory analyses to characterize different trajectories of improvement in scores for soldiers who received BH specialty care.

The analyses presented in this study have several strengths, drawing on a range of data sources and identifying variables that had the greatest impact on soldiers' treatment outcomes within and across treatment episodes for the three conditions in our sample. However, the analyses also have several limitations.

The sample studied was limited to soldiers receiving BH care with a new treatment episode, a minimal level of symptom severity, and at least one follow-up symptom score in one to six months. Our results do not include those who received Army BH care but did not have any symptom scores, had less than two scores, or had a lower level of symptom severity at the first visit. Therefore, our results may not generalize to these other populations. Although we examined a lengthy list of variables, we could not analyze all possible variables because they were not included in the available data (e.g., lifetime history of trauma). Furthermore, some variables that we examined may not adequately capture the treatment delivered. For example, we examined a provider-entered variable indicating treatment with evidence-based psychotherapy, but it was not associated with outcomes—a surprising result, given the research support for these therapies in reducing symptoms. The observational nature of the data limited our ability to draw causal links between the predictors and outcomes. In addition, even with the careful variable selection, there is always the possibility of remaining collinearity—highly related variables that could produce biased estimates of the association between outcome and some predictors. With the time difference between the initial and last score varying across soldiers, there is also the possibility of such a time variable moderating (or interacting with) the association between predictors and outcomes.

Findings

BHDP Is Widely Used to Track PTSD, Depression, and Anxiety Symptoms, but There Are Opportunities to Expand Symptom Tracking

Soldiers in all three samples received more scores the longer they were in treatment, suggesting that BHDP has been widely implemented and that soldiers routinely received scores during BH visits. However, we identified areas for improvement. For example, some soldiers had BH visits after their last symptom measure score. Thus, a soldier's last score was not necessarily a measure of symptoms at their last visit. In addition, soldiers in the anxiety sample were less likely than soldiers in the PTSD and depression samples to receive multiple scores; this pattern was more apparent for soldiers with 16 or more BH visits. The Army's monitoring of anxiety symptoms was implemented after monitoring for PTSD and depression, so these results could reflect differing stages of implementation. Further, we identified several differences between the Army methodology for computing depression response and remission measures and similar measures endorsed by the National Quality Forum; these differences make comparisons between Army and civilian care difficult.

Stronger Patient-Reported Therapeutic Alliance Was Associated with Improved PTSD, Depression, and Anxiety Outcomes

Despite using methods to identify the “best-in-class” pretreatment and treatment factors predicting clinical outcomes, we found that no pretreatment variables were consistently associated with outcomes. That is, no demographic or risk factors consistently were associated with all three targeted outcomes within a diagnosis (i.e., change scores, response to treatment, and remission) or across diagnoses. However, one treatment factor was consistently associated with outcomes both within and between diagnoses: therapeutic alliance. Even when controlling for other treatment factors, a perceived strong working relationship between soldiers and their providers was associated with decreased PTSD, depression, and anxiety symptoms; PTSD and depression response to treatment; and PTSD, depression, and anxiety remission.

Increased Supply of Benzodiazepines Was Associated with Worse PTSD, Depression, and Anxiety Outcomes

Besides therapeutic alliance, no other treatment factor was consistently associated with improved outcomes over time within or across diagnoses. However, a larger supply of benzodiazepines dispensed was associated with poorer PTSD, depression, and anxiety outcomes compared with no supply of benzodiazepines. Although our study did not assess whether soldiers who received these medications used them as prescribed, soldiers with more than a 30-day supply of the drug experienced poorer outcomes.

Many Soldiers' Trajectories of Symptom Change Did Not Demonstrate Improvement

Outcome quality measures currently tracked by the Army—response to treatment and remission based on last symptom scores—showed improvement in some soldiers but also highlighted a need to continue improving the effectiveness of Army BH care. Specifically, rates of achieving either response or remission within one to six months were 35 percent for PTSD, 45 percent for depression, and 41 percent for anxiety. It remains possible that soldiers continued to improve after their last score because our data only captured scores up to six months after their initial score. While one to six months is a relatively short time to evaluate symptom improvement and/or remission of symptoms, some patients have been shown to reach remission of depression within six months (Angstman, Rohrer, and Rasmussen, 2012). Our analyses also identified three or four different patterns, or trajectories, of symptom change for each sample. The majority of soldiers with PTSD (83 percent) were included in a trajectory that did not demonstrate improvement in their symptoms. Among patients with depression, 34 percent were included in a trajectory that showed no improvement, and 45 percent showed a small improvement. Forty-five percent of the anxiety sample were in a trajectory that showed no improvement. Exploring predictors of the trajectories yielded mixed results. Often, predictors that captured increased utilization were associated with a lack of improvement, but it is likely that soldiers who were not improving were more likely to receive additional care. Although we adjusted our models for severity at initial score, there may have been unmeasured confounders that led to these findings.

Recommendations and Policy Implications

Recommendation 1. Provide Feedback on Therapeutic Alliance and Guidance to Providers on How to Strengthen Alliance with Their Patients

A perceived strong working relationship between soldiers and their providers was associated with better outcomes in all three samples, and this finding is consistent with the literature on this topic. Providing clinicians, clinical leads, and MTFs with information about how soldiers perceive their alliance with providers may help providers address difficulties in the therapeutic relationship directly with the patient and use this information as an opportunity to repair the relationship or address concerns about treatment that the patient may have. Provider training in this area may help minimize treatment dropout and improve outcomes. Our analyses were somewhat limited by relying on the last therapeutic alliance score, which was frequently assessed at the same time as the last symptom score, potentially biasing results by increasing the association between these two variables. Therefore, we recommend that the Army ensure that therapeutic alliance is routinely assessed early in treatment. This will ensure that the results of the measure are actionable during treatment for the clinician and that these analyses can be replicated using alliance scores that are collected prior to the last symptom outcome score.

Recommendation 2. Expand Tracking and Feedback on Benzodiazepine Prescribing

One of the most consistent findings in our analyses was that soldiers who had a larger supply of benzodiazepines—more than 30 days—were more likely to have worse outcomes. The clinical practice guideline for PTSD cautions against using benzodiazepines as monotherapy or augmentation therapy for the treatment of PTSD (U.S. Department of Veterans Affairs and U.S. Department of Defense, 2017), and these medications have been identified as potentially harmful in this population. In 2018, the Defense Health Agency initiated a program to track benzodiazepine prescribing among providers who treat PTSD and acute stress disorder called the PTS Provider Prescribing Profile (Military Health System Communications Office, 2018). Results are monitored and shared with MTF commanders. The Army's Behavioral Health Service Line is also tracking benzodiazepines and atypical antipsychotic prescriptions for PTSD (Woolaway-Bickel, 2019). Data on benzodiazepine prescribing could also be provided to clinic leadership and individual providers as a potential approach to improve patient outcomes, including response to treatment and remission metrics. Additional work could be conducted to identify the duration of benzodiazepine use that may lead to worse outcomes, but our analyses suggest that more than a 30-day supply is associated with worse outcomes.

Recommendation 3. Increase Provider Use of Measurement-Based BH Care

The Army continues to expand and monitor the use of BHDP for BH care. Our sample was limited by our selection criteria, making it difficult to assess the extent of measurement-based care. However, a 2016 survey of BH providers suggested that 76 percent of Army BH providers screened soldiers for PTSD or depression using a validated measure, but only 59 percent reported using symptom data to inform treatment (Hepner et al., 2017). Measurement-based care involves the repeated collection of outcome data and the use of those data to inform decisions throughout the course of treatment (Fortney et al., 2016). These data provide timely feedback to providers about patient progress, allowing providers to quickly identify patients who are not improving or deteriorating (Boswell et al., 2015). As the Army continues to expand the use of BHDP and the collection of symptom measures, an important strategy for improving BH treatment outcomes may include supporting providers in frequently collecting and using these data in treatment decisions.

Directions for Future Research

The analyses presented in this study provide results that can guide the Army in improving outcomes for soldiers who receive BH care. These analyses also raised several questions that could be addressed in future research. These include

• Identifying quality of care measures that can help providers focus on aspects of treatment that have the highest likelihood of improving soldier outcomes (sometimes referred to as “driver” metrics). These analyses would target therapeutic alliance and supply of benzodiazepines to identify detailed specifications for tracking metrics to assess these variables that are associated with treatment outcomes.
• Evaluating whether refinements in the definitions of response to treatment and remission currently used by the Army could improve assessment of significant symptom improvement and increases in psychosocial functioning. Our analyses highlighted that several soldiers who receive at least some Army BH care are excluded from the outcome measures. Expanding inclusion, or developing alternative metrics, may provide more opportunities to more thoroughly monitor the effectiveness of Army BH care. Further, modifications to measure specifications could also improve the ability to compare Army performance with civilian care settings.
• Exploring the utility of alternative approaches to monitoring Army outcome measures (i.e., response to treatment and remission). This could include tracking these metrics stratified by populations of interest (e.g., broken out by demographic characteristics or those on a medical evaluation board). Stratified reporting of outcome measures has been suggested as an alternative to complex case-mix adjustment models.
• Exploring the utility of expanding Army BH outcome monitoring beyond symptom measures. The Army is a leader in monitoring outcomes of BH care. Thus far, this monitoring has focused on symptom measures. This is a logical focus because symptoms are the most proximal outcome of BH care and are likely to be improved by high-quality care. However, the Army could explore monitoring other outcomes of BH care. Potential targets could include indicators of readiness, functioning, or quality of life.
• Developing and evaluating more effective treatments for PTSD, depression, and anxiety disorders. Our analyses add to existing recent literature that has called for more effective treatments, particularly for service members with PTSD. The observed rates of response to treatment and remission highlight the continued need for more effective treatments within both Army and non-Army settings.