Health Care Utilization Patterns Among High-Cost VA Patients With Mental Health Conditions

Many health care organizations are exploring opportunities to improve outcomes and reduce costs for high-cost patients with the most complex needs. Programs for high-cost patients aim to enhance quality of care while optimizing service use to reduce unnecessary utilization (1,2). Comanaging mental health and general medical problems is an emerging priority given high rates of mental health conditions among high-cost patients and evidence that adults with mental health conditions struggle to manage their chronic medical comorbidities (3,4).

Although several studies have shown that mental health conditions are associated with more severe physical symptoms, greater health care utilization, and higher costs, little research has focused on characteristics and needs of high-cost patients with specific mental health conditions (3–9). Understanding these patterns could inform the design of targeted interventions.

The Department of Veterans Affairs (VA) health care system provides an opportunity to study mental health conditions and utilization patterns among high-cost patients in a large integrated care setting. Mirroring national trends, the costliest 5% of VA patients account for approximately half of total health care expenditures (10,11). The objective of this study was to investigate how patterns of health care utilization and costs differ among the costliest 5% of VA patients with and without diagnosed mental health conditions, as well as among different diagnostic groups.

Methods

Population and Data Sources

We used VA administrative records to identify a cohort of 5,233,994 individuals who were eligible for and received care in the VA health care system in fiscal year 2010 (October 1, 2009, through September 30, 2010). We calculated total VA health care costs for each patient by aggregating costs of inpatient, outpatient, pharmacy, and non-VA contract care. Because there are no billing records in the VA, we obtained estimates of inpatient and outpatient costs from 2010 Average Cost data files, which provide cost estimates based on Medicare payments for comparable care (12,13). We used 2010 Decision Support System pharmacy files to obtain costs of prescription drugs filled in the VA, and we used 2010 Fee Basis files to obtain service-specific payments made to non-VA providers.

We used ICD-9 diagnostic codes and chronic condition indicators (CCI) established by the Agency for Healthcare Research and Quality (AHRQ) (14) to identify the following mental health conditions: schizophrenia, bipolar depression, other psychosis, alcohol dependence and abuse, drug dependence and abuse, posttraumatic stress disorder (PTSD), and depression. [A table in an online supplement to this article lists the ICD-9 codes used.] After identifying the costliest 5% of patients (N=261,699), we excluded individuals with mental health conditions who did not have one of the seven diagnoses listed above (N=184), leaving a final study cohort of 261,515 patients with any of the seven psychiatric diagnoses or without any mental health condition.

Grouping Patients by Mental Health Conditions

Each high-cost patient was assigned to one of five mutually exclusive groups: no mental health condition, serious mental illness (schizophrenia, bipolar depression, or other psychosis), substance use disorder (alcohol or drug dependence or abuse), PTSD, or depression. Groups were constructed as follows: serious mental illness group included patients with serious mental illness with or without a comorbid substance use disorder, PTSD, or depression; the substance use disorder group included patients with a substance use disorder with or without PTSD or depression; the PTSD group included patients with PTSD with or without depression; and the depression group included patients with depression but none of the other mental health conditions of focus. We established the hierarchical groups by using a multivariate linear regression model in which we explored the association between mutually exclusive groups of mental health conditions and the proportion of costs generated by mental health care for each group of conditions. Groups that included serious mental illness had the strongest association with proportion of costs generated by mental health care, followed by substance use disorder, PTSD, and then depression. Individuals with more than one mental health condition were assigned to a group according to the ranked order above. This approach is similar to methods used in previous assessments of psychiatric conditions among VA patients (3). [A table presenting results of this analysis is available in the online supplement.]

Patient Characteristics

For all individuals in our cohort, we obtained sociodemographic information including age, sex, race-ethnicity, marital status, and documented homelessness during the year of investigation. In addition, we categorized patients as having no outside insurance, private or public insurance other than Medicare or Medicaid, and Medicare or Medicaid supplemental coverage. We also determined the presence of 27 chronic general medical conditions (using AHRQ CCIs) (15) that have been the focus of previous quality improvement efforts and research within the VA because of their prevalence, management challenges, or cost (15–18). We identified conditions by using ICD-9 codes from inpatient and outpatient primary and secondary diagnoses, restricting our determination to conditions that were documented at least twice in order to avoid counting conditions that were recorded to rule out diagnoses.

Utilization and Cost Measures

For inpatient utilization, we determined the number of hospitalizations and cumulative length of stay (LOS), defined as total number of days hospitalized in the study year, for behavioral, residential-domiciliary, medical-surgical, and long-term care. For outpatient utilization, we determined the number of visits for mental health, emergency, primary, and specialty care on the basis of clinic identifiers for location of care. For inpatient costs, we examined costs for behavioral, residential-domiciliary, medical-surgical, long-term care, and other hospitalizations on the basis of inpatient bed section. For outpatient costs, we examined costs for behavioral, medical-surgical, diagnostic, pharmacy, and other care. Finally, we calculated the proportion of costs generated by mental health care by dividing the mean mental health costs by the mean total costs for each diagnostic cohort.

Analyses

Analyses of variance and chi square tests were used to compare individuals in the five diagnostic cohorts on the basis of sociodemographic characteristics and the prevalence of 27 chronic medical conditions. An a priori significance level of p<.001 was used to account for multiple comparisons. We used multivariable linear regression to examine the relationship between the diagnostic cohorts and number of inpatient stays, number of outpatient visits, inpatient costs, and outpatient costs. All models used “no mental health condition” as the reference group and adjusted for patient age, sex, race-ethnicity, marital status, documented homelessness during the year of investigation, presence of 27 comorbid chronic general medical conditions (included as individual covariates), and correlation within facilities. A postestimation analysis was performed to estimate the mean inpatient and outpatient utilization and cost values for each diagnostic group when all covariates were held constant.

We used Stata, version 12.0, to perform all analyses. Analyses used deidentified data and were approved by the Stanford University Institutional Review Board.

Results

Characteristics of High-Cost VA Patients

Almost half (48%) of the most costly 5% of VA patients had at least one psychiatric diagnosis, and of these individuals, 49% had two or more mental health conditions. Depression was the most prevalent mental health condition (29% of all high-cost patients), followed by substance use disorder, PTSD, and serious mental illness (20%, 17%, and 13%, respectively). When we assigned patients to mutually exclusive groups by using our hierarchical ranking system, 13% of patients were assigned to the serious mental illness group, 15% to the substance use disorder group, 9% to the PTSD group, and 11% to the depression group. Table 1 summarizes data on sociodemographic characteristics of high-cost patients in the five diagnostic groups. Among patients with at least one psychiatric diagnosis, those in the serious mental illness and substance use disorder groups tended to be younger and more likely to be non-Hispanic black, not married, without supplemental insurance, and homeless compared with those in the PTSD and depression groups (p<.001).

The PTSD and depression groups had a higher mean number of comorbid chronic medical conditions (4.3 and 4.6, respectively) than the serious mental illness and substance use disorder groups (3.2 and 3.1, respectively), and higher rates of most medical conditions. Patients without a psychiatric diagnosis had a mean of 3.9 comorbid chronic medical conditions and were more likely than those in any of the mental health groups to have high-cost medical conditions such as cancer and chronic renal failure (Table 2). All differences among groups were statistically significant (p<.001).

Utilization by High-Cost Patients

Inpatient and outpatient utilization measures for each of the groups were estimated from regression models adjusted for sociodemographic characteristics and are shown in Table 3. Among individuals with mental health conditions, those with serious mental illness had the longest cumulative behavioral LOS (13.8 days), compared with individuals with a substance use disorder (6.9 days), PTSD (3.6 days), and depression (2.5 days). Cumulative residential-domiciliary LOS, a measure of residential care for the purposes of mental health and psychosocial rehabilitation (for example, support for housing and employment needs), was much greater for the substance use disorder group (19.1 days) than for the other groups (4.7–5.6 days). Patients in the serious mental illness and substance use disorder groups had greater outpatient mental health utilization (37.6 and 33.2 mental health visits, respectively), compared with individuals in the PTSD and depression groups (18.6 and 9.5, respectively).

The total number of hospital stays for any reason over the study period was greater in the serious mental illness and substance use disorder groups (2.2 and 2.3 stays, respectively) than in the PTSD, depression, and “no mental health condition” groups (1.5, 1.8, and 1.7, respectively) (Table 3). The relatively low number of hospital stays among patients without a diagnosed mental health condition was partially offset by longer cumulative LOS for medical-surgical care among these patients compared with all other groups except the depression group. Regarding outpatient care, the serious mental illness and substance use disorder groups had the highest number of emergency department visits, and the PTSD and depression groups had the highest number of specialty care visits. Individuals in the groups with a mental health condition had more primary care visits, on average, than individuals without a diagnosed mental health condition.

Variation in Costs Among High-Cost Patients

Table 4 presents inpatient and outpatient cost data for the diagnostic cohorts, including the proportion of costs generated by mental health care. Among individuals with mental health conditions, inpatient and outpatient behavioral health costs were greatest for the serious mental illness group ($16,562 and $11,006, respectively) and lowest for the depression group ($2,978 and $2,685, respectively). Residential-domiciliary costs, another measure of mental health costs, were greater for the substance use disorder group ($7,135) than for the serious mental illness group ($2,105), PTSD group ($1,921), and depression group ($1,744). For non–mental health costs, patients without a diagnosed mental health condition had higher inpatient and outpatient medical-surgical costs ($28,036 and $12,170, respectively) than any of the groups with mental health conditions. Among individuals with mental health conditions, inpatient and outpatient medical-surgical costs were highest for the depression group ($26,655 and $11,405, respectively) and lowest for the serious mental illness group ($18,033 and $7,998, respectively).

For patients in the serious mental illness and substance use disorder groups, a much higher percentage of costs was generated by mental health care (41% and 31%, respectively) compared with patients in the PTSD (18%), depression (11%), and “no mental health condition” (8%) groups (Table 4).

Discussion

This study adds to growing evidence underscoring the need to coordinate general medical and mental health care to improve outcomes and optimize utilization among high-cost patients (4,19). Our identification of different utilization and cost patterns among subgroups of patients with mental health conditions suggests a number of opportunities to refine health care delivery models targeting high-cost patients.

Several studies have shown that comorbid mental health problems, primarily depression, result in worse outcomes and higher costs for a range of chronic general medical conditions (19–21). Our findings build on this prior research by showing that utilization and costs of mental health and medical-surgical care differed among various groups of high-cost patients with mental health conditions. For example, patients with PTSD and depression had higher rates of costly comorbid medical illnesses (such as cancer, chronic renal failure, diabetes, heart failure, and ischemic heart disease), higher numbers of primary and specialty care visits, and longer medical-surgical hospitalizations than patients with serious mental illness and substance use disorders. Conversely, patients with serious mental illness and patients with substance use disorders incurred a greater proportion of care (and substantially greater costs) in the mental health setting than patients with PTSD or depression.

These findings support a strategy that stratifies high-cost patients such that patients with serious mental illness or substance use disorders receive intensive interventions focusing on their specific mental health conditions while patients with conditions such as PTSD and depression are steered toward multifaceted interventions that provide comprehensive care for both general medical and mental health conditions. An example of the former model (targeted mental health care) is the VA’s Mental Health Intensive Case Management program, an intensive, multidisciplinary outpatient program that serves as a cost-effective alternative to hospitalization for approximately 4,000 veterans nationwide with severe mental disorders and a history of high hospital use (22,23). There are also examples of programs that coordinate care for general medical and mental health comorbidities, some of which have resulted in reductions in psychosocial problems, including homelessness and alcohol use, as well as reductions in emergency department use, inpatient admissions, and costs (24,25).

Although the optimal structure for intensive management programs is still evolving, addressing mental health conditions among high-cost patients is likely to be a critical focus, especially for programs serving safety-net populations. For example, a study of the Washington State Rethinking Care program (a 24-month community-based intensive care management program for a subset of high-cost Medicaid clients with mental illness in the aged, blind, and disabled eligibility categories) found that individuals who actively participated in the intervention increased their use of outpatient mental health services. Although total costs did not change in an intent-to-treat analysis, inpatient admissions and emergency department visits decreased, suggesting a shift to more appropriate service use (24). Several other studies have found that programs focused on populations with a high prevalence of mental health conditions are able to reduce the costs of care or improve health outcomes for patients (1,26).

Our analysis was limited by several factors. First, we assigned patients to mutually exclusive diagnostic groups even though almost half had more than one psychiatric diagnosis. However, when we explored alternative groupings, our main findings did not substantially change [see online supplement]. Second, we used administrative data to identify patients with mental health conditions even though previous studies suggest that as many as half of patients with depression and PTSD do not have formal diagnoses (27,28). Underdiagnosis could diminish the observed differences between the diagnostic groups. However, our focus on high-cost patients probably lowered the likelihood of underdiagnosis. Furthermore, we restricted our determination of mental health conditions to those that were documented at least twice in the administrative records, an approach that has been shown to improve the positive predictive value of psychiatric diagnoses in VA administrative data (29,30). Third, these analyses did not account for care that VA patients may have received outside the VA, including contract care covered by the VA and care covered by supplemental Medicare or private insurance. Dual coverage was more common among patients with PTSD and depression than among patients with serious mental illness and patients with substance use disorders, but it is unclear how non-VA services and costs would be distributed with regard to mental health versus non–mental health care.

Fourth, we were unable to assess whether the level of service-connected disability varied across the different diagnostic groups. Although level of service connection would be unlikely to affect use of most services that we analyzed, there are certain services, such as long-term care, that are restricted to patients with a specific level of service-connected disability. This could explain some of the observed variation in long-term care use among the different groups. Finally, it is important to acknowledge that our results should be interpreted as specific to the VA. For example, residential-domiciliary care is somewhat unique to the VA and accounted for a large proportion of the mental health costs of the group with substance use disorders. However, the VA is the largest integrated health care system in the United States, and the VA’s experiences may serve as a model for emerging integrated care systems, such as accountable care organizations, that seek to address the health care needs of high-cost patients and deliver high-value care.

Conclusions

The findings suggest that a one-size-fits-all approach may not be appropriate when designing intensive management programs for high-cost patients with mental health conditions. Although patients with conditions such as depression and PTSD are likely to benefit from integrated models of care that address their comorbid general medical conditions, patients with serious mental illness and patients with substance use disorders may require intensive, disease-specific services. Dedicating resources in this manner will help ensure that patients receive services that meet their most pressing needs and address their most complicated and costly conditions.