A Reduction in Health Care Expenditures Linked to Mental Health Service Use Among Adults With Chronic Physical Conditions

Cardiovascular disease (CVD), cancer, diabetes, chronic lower-respiratory diseases, and arthritis are among the leading causes of mortality and disability among American adults (1, 2). These conditions have posed a significant financial burden of hundreds of billions of U.S. dollars each year to both patients and the health care system (3–7). Among individuals with these chronic physical conditions, mental disorders, such as affective disorders and anxiety, are also prevalent as a result of coping with physical symptoms (8–15). The co-occurrence of mental health problems among patients with these chronic physical conditions could reduce adherence to treatment, decrease quality of life (16–19), and increase health care costs (20, 21).

The vast literature on how mental health service use influences health care expenditures of individuals with physical conditions focuses on how depression treatment is associated with medical costs of people with concurrent depression. Some found that total health care expenditure was reduced by treatment with antidepressants and psychotherapy (22, 23), yet others showed that receiving depression treatment was associated with increased costs (24) or no differences in costs (25). Furthermore, mental health conditions are often underdiagnosed and overlooked among individuals with complex chronic physical conditions (26, 27). A recent study of U.S. adults ages 18 and older found that having adverse physical health events altered demand for mental health services (28), which could have implications on overall health care utilization and cost over time.

To our knowledge, little research has quantified the impact of receipt of mental health services on health care expenditures in a large population sample of people with chronic physical conditions that are often comorbid with various mental health conditions. Our primary objective was to examine how the utilization of mental health services in the preceding year was associated with the change in overall health care expenditure in the subsequent year among individuals with at least one of six chronic physical conditions: CVD, cancer, diabetes, emphysema, asthma, and arthritis. We also analyzed the relationship between receipt of each type of mental health service and overall health care expenditure. Moreover, we examined the impact of mental health service use on service-level costs, including inpatient services, outpatient services, emergency department (ED) visits, office-based physician visits, and prescribed medication.

Methods

Data and Sample

We used data from the Household Component of the Medical Expenditure Panel Survey (MEPS) for years 2004 to 2014 (29). MEPS is a large-scale survey of a nationally representative sample of the noninstitutionalized population in the United States (29). The panel design of the MEPS allowed us to follow each individual in each of the 10 panels (2004–2014) for 2 calendar years and thus to examine how receipt of mental health services in the preceding year was associated with health care expenditures in the subsequent year.

The study population consisted of adults ages 18 and older who reported diagnoses of any of six physical conditions: CVD (i.e., heart disease and stroke), cancer, diabetes, emphysema, asthma, and arthritis. We derived our sample on the basis of the MEPS questions about whether an individual had ever been diagnosed as having each of the six physical conditions either before or during the first year of follow-up (30). Among 36,480 individuals who met the criteria, we excluded 3,061 persons (8.4%) who reported zero overall health care spending during a year, which is uncommon among adults with these physical conditions. Having zero health care spending may be due to self-report bias, or these persons may have had different characteristics compared with others in the sample. The analytic sample included 33,419 adults with any of the six physical conditions, of whom 7,929 had a diagnosis of co-occurring mental disorders, such as anxiety, mood disorders, adjustment disorders, schizophrenia, and other psychotic disorders.

Outcome Variable

Our primary outcome was overall health care expenditure, which captured the overall payment from all sources for health care services used within a year. We also analyzed service-level costs, separately for inpatient services, outpatient services, ED visits, office-based physician visits, and prescribed medication. Health care expenditures were adjusted for inflation by the Personal Healthcare (overall and component) Price Indices and Consumer Price Index (for medication) (31) and expressed in 2014 constant dollars.

Outpatient Mental Health Services

Mental health services included psychotherapy and medication treatment. Psychotherapy was defined as seeing a psychologist or as receipt of psychotherapy or mental health counseling during an office-based physician visit, outpatient visit, or ED visit. Medication treatment for mental health was identified by purchases of prescription drugs associated with psychiatric conditions. Information on type of provider (i.e., psychologist) and visit category (i.e., psychotherapy or mental health counseling) in MEPS medical event files was used to determine psychotherapy use. We used clinical classification codes 650 (adjustment disorders), 651 (anxiety disorders), 657 (mood disorders), and 659 (schizophrenia and other psychotic disorders) to identify prescription drug purchases associated with mental disorders (a table with a complete list of ICD-9-CM codes associated with the clinical classification codes is included in an online supplement to this article). Most common medications associated with these clinical classification codes included selective serotonin reuptake inhibitors, selective serotonin and norepinephrine reuptake inhibitors, antidepressants, antipsychotics, and anxiolytic or sedative drugs.

Covariates

Individual demographic and socioeconomic characteristics included age, sex, race-ethnicity, marital status, census region, metropolitan residence, income, education, and insurance status. Because the utilization of mental health services could depend on the severity of physical illness as well as the need for mental health care, we controlled for perceived physical and mental health status as proxy for the severity of physical and mental health conditions, respectively. Our measures included physical component summary and mental component summary scores that were computed with information from the 12-item Short-Form Health Survey (32). Co-occurrence of mental disorders and alcohol or drug use disorders was identified with clinical classification codes from the MEPS household condition files.

Analytic Model

The identification of the effect of receiving mental health services on subsequent health care expenditures hinged on a difference-in-differences (DID) framework. We compared an average change in health care expenditures from the preceding year to the subsequent year among individuals with the chronic conditions who utilized mental health services in the preceding year to an average change in health care expenditures among those who did not.

Although overall health care expenditure did not have zero value, it was distributed highly right-skewed with individuals who had extremely high cost. To obtain more consistent parameter estimates, we employed a generalized linear model (GLM) with the log link function and the gamma distribution (33). We included interaction terms for the indicators of receipt of mental health services in the preceding year and the indicator of the subsequent year in our models. The coefficients on the interaction terms measured the percentage point changes in health care expenditure in the subsequent year attributed to receipt of mental health services in the preceding year.

We then examined the effects on health care expenditures of receiving mental health services by type of service using two-part DID models (33–35), because service-level costs may be right-skewed and could also have many zeros (e.g., patients who did not have any hospital admissions or ED visits). In this approach, the first part estimated the impact of receiving mental health services on the probability of incurring any cost by using logistic regression and the second part modeled the change in expenditures among those who had nonzero costs by using GLM.

All models controlled for annual individual demographic and socioeconomic characteristics, perceived health status, co-occurrence of mental disorders and alcohol or drug use disorders, and time trend. We computed the average marginal effects of the interaction terms in terms of dollar amounts to estimate the change in health care expenditures associated with use of mental health services. Bootstrapped standard errors of the marginal interaction effects were obtained based on 1,000 repetitions.

Although our models included different sociodemographic characteristics and time trend, selection bias might still be present because of potential unobserved person heterogeneity. To mitigate this concern, we adjusted our estimates with inverse-probability weights to better balance baseline characteristics of the two comparison groups (i.e., those who used mental health services and those who did not). A properly reweighted inverse-probability–weighted estimator is largely used in treatment effect estimation as a tool for causal inference to adjust for measured confounders (36, 37). Further, an inverse-probability–weighted DID estimator has a doubly robust property because it adjusts estimates for both observed and unobserved selection (38, 39). We obtained inverse-probability weights by fitting a logistic regression of the indicator of receipt of any mental health services on all the covariates above, as well as smoking status and year dummies. Our analyses used Stata software, version 13.1 (40). Because we used publicly available MEPS data sets, review by an institutional review board was waived.

For a sensitivity analysis, we estimated linear regression models with and without a log-transformation of expenditures to check the robustness of our results. We also applied survey weights in MEPS and excluded outlier observations.

Results

Characteristics of the Study Population

Table 1 summarizes characteristics of the study sample in the preceding year, for the entire sample and by use of mental health services. The mean overall health care expenditure in the preceding year among individuals with at least one of the six physical conditions was $9,067. Among the 33,419 persons in the overall sample, 20.1% received mental health services. The second and third columns present characteristics of two groups without inverse-probability weighting. Among those who used mental health services, 8.5% received psychotherapy only, 71.5% received medication treatment only, and 20% received a combination of both. The mean overall health care expenditure among individuals who received any mental health services was $13,141, which was higher than the average cost of $8,043 among those who did not use mental health services.

In the sample, 35% had CVD and 18% had cancer. The proportions of individuals who had diabetes or who had emphysema or asthma were 27% and 25%, respectively; 59% had arthritis. A total of 7,929 individuals (23.7%) reported that they had been diagnosed as having a mental disorder. As measured by the physical component summary score, those who received mental health services had poorer self-perceived health status, compared with those who did not receive mental health services. Among those who received mental health services, 95% reported that they had mental disorders. Six percent of those who did not receive mental health services reported that they had mental disorders. Compared with those who did not receive mental health services, those who received services had a higher proportion of co-occurring alcohol or drug use disorders (1.9% versus 0.5%). Compared with those who did not receive mental health services, those who did were more likely to be female, White, and poor and less likely to be married and employed.

The fourth and fifth columns of Table 1 present characteristics of the two groups with use of inverse-probability weights. After weighting, the two groups became more balanced in observed characteristics, especially perceived mental health status (mental component summary score), co-occurrence of mental and alcohol or drug use disorders, marital status, and family income.

Table 2 reports mean health care expenditures, total and by type of service. The increase in overall expenditure was greater among individuals who did not use mental health services (an increase from $8,043 to $8,624), compared with the increase in expenditure of those receiving mental health services (an increase from $13,141 to $13,295). Similar patterns were observed for inpatient expenditure and prescribed medication cost. Average ED expenditure slightly decreased in both groups. Office-based provider cost and outpatient cost slightly decreased among those using mental health services, but these costs increased among individuals who did not use mental health services.

Table 3 presents coefficient estimates from our GLM models for overall health care expenditure. Model 1 estimated change in overall cost associated with receiving any mental health services, and model 2 estimated change in overall cost associated with types of mental health services. In model 1, the increase in expenditure in the subsequent year among individuals who received any mental health services was smaller than the increase in expenditure among those who did not receive these services—by 12.6 percentage points (p<0.05). Model 2 suggests that receiving a combination of psychotherapy and medication treatment reduced the increase in subsequent overall cost by 21.7 percentage points (p<0.01).

Table 4 presents results from two-part DID models used to estimate the impact of mental health service use on service-level costs. We did not find a statistically significant relationship of mental health service use and subsequent inpatient cost, ED cost, or outpatient cost. Although use of mental health services did not reduce the probabilities of incurring costs for office-based visits and prescribed medication, it was associated with a reduction in costs. Utilization of any mental health services lowered the subsequent cost increase for office-based visits by 15.7 percentage points (p<0.05) and for prescribed medication by 17.7 percentage points (p<0.05). Receiving a combination of both psychotherapy and medication treatment lowered the cost increase for office-based visits and prescribed medication by 33.3 percentage points (p<0.001) and 22.7 percentage points (p<0.05), respectively.

Table 5 presents changes in health care expenditures associated with use of mental health services in terms of dollar amounts (2014 constant dollars). On average, adults who received any mental health services had a cost saving in the subsequent year of $1,146, compared with those who did not use mental health services. Receipt of both psychotherapy and medication treatment was associated with savings in overall cost, office-based visit cost, and prescribed medication cost of $2,690, $1,056, and $577, respectively.

Sensitivity analyses using linear regression models with and without log-transformed expenditures, with exclusion of outliers, or adjusting the estimates with survey weights yielded consistent results.

Discussion

Our findings suggest that receipt of mental health services may subsequently lead to savings in overall health care cost among adults with any of six prevalent chronic physical conditions, i.e., CVD, cancer, diabetes, emphysema, asthma, or arthritis. Use of mental health services in the preceding year was significantly associated with a reduction in subsequent overall health care expenditure by 12.6 percentage points. The average saving was estimated to be $1,146 dollars (2014 constant U.S. dollars) per individual. The estimated saving was equivalent to 8.7% of overall health care expenditure in the preceding year of individuals receiving any mental health services ($13,141). Compared with the average cost of mental health services in the preceding year, which was $5,098 (i.e., the difference in average overall expenditures between individuals who used mental health services and those who did not), the estimated saving of $1,146 dollars was about 22.5% of the cost incurred for mental health care. Because mental health problems are common among adults with chronic physical conditions but often go unrecognized and thus left untreated, our study suggests a potential impact on health care expenditures of integrating mental health services for these individuals.

Adding to the current literature, our results also imply a greater financial benefit of providing a combination of psychotherapy and medication treatment for individuals with chronic physical conditions. The combination of psychotherapy and medication treatment subsequently lowered the increase of overall health care cost by 21.7 percentage points or $2,690 dollars for these adults. Previous studies have also found that the combination of psychotherapy and antidepressants was associated with a decrease in total health care expenditure among diabetes patients (22, 23, 41, 42) or was cost-effective for CVD patients (25).

We did not find a statistically significant impact of mental health service use on costs for inpatient services, outpatient services, and ED visits, which might be attributable to a short follow-up time. Some studies have suggested that cost reduction related to mental health care could result in a longer follow-up time (23, 25). Nevertheless, we found subsequent cost savings for office-based visits and prescribed medication among adults who used mental health services. Some studies suggested that cost containment associated with mental health services could be attributable to either reduction of future cost related to mental health conditions (43) or a decrease in future health care use resulting from improved chronic physical conditions (44).

Our findings may also speak to potential cost savings from health care delivery transformations rapidly expanding nationwide that integrate general medical and mental health care. Some health care delivery models with such integration have been shown to improve management of chronic illnesses, depression screening, and follow-up (45) while controlling cost growth (46, 47). Although our results suggest that providing mental health services, especially psychotherapy, may reduce overall costs for people with costly chronic physical conditions, they do not exclude the possibility that covering mental health services for the entire insured population will also in the long run reduce overall costs and increase quality of life. Future research is needed to investigate how state and federal policies that encourage behavioral health integration through delivery system reforms affect health care costs and health outcomes.

Our study had some limitations. First, given the nature of the MEPS, we were not able to investigate the impact on health care expenditures of receipt of mental health care in a longer follow-up time. Second, the data did not allow for separation of expenditures for physical conditions and mental health conditions. Also, because of the high prevalence of comorbidities in the study population, it was not possible to examine the impact of mental health services utilization separately for each chronic physical condition. Further studies may benefit from examining how mental health care influences costs for patients with chronic physical illnesses in the long-term, as well as examining separate costs for each physical condition and mental health conditions. Third, this was an observational study with 2 years of follow-up, and our estimates might be biased because of different trends in health care costs between the two groups (those who used mental health services and those who did not) during the pre-period. Although matching two groups by using inverse-probability weights could mitigate this concern by minimizing observed differences between the groups, and the application of the quasi-experimental DID approach possibly captured unobservable group differences, potential selection bias might not be completely ruled out with observational data. Finally, demand for mental health services could depend on prices and availability of services. Because of data limitations, we could not control for some state-level characteristics, such as state policy regarding mental health treatment or prices of mental health services.

Conclusions

Utilization of mental health services was associated with reduced overall health care expenditure among adults with major chronic physical conditions, including CVD, cancer, diabetes, emphysema, asthma, and arthritis. Psychotherapy alone had a larger effect on cost containment than did medication treatment alone—indeed the latter did not appear to have a meaningful effect on overall costs. The combination of the two, however, yielded a modestly larger effect than did psychotherapy alone. Future research should evaluate whether emerging health care delivery models that integrate general medical and mental health care are cost-effective.