The Association Between Self-Rated Mental Health Status and Total Health Care Expenditure: A Cross-Sectional Analysis of a Nationally Representative Sample.

Both clinical diagnoses and self-rated measures of mental illness are associated with a variety of outcomes, including physical well-being, health utilization, and expenditure. However, much of current literature primarily utilizes clinically diagnosed data.This cross-sectional study explores the impact of mental illness and health care expenditure using 2 self-rated measures: self-rated measured of perceived mental health status (SRMH) and Kessler Screening Scale for Psychological Distress (K6).Data from the 2011 Medical Expenditure Panel Survey Household Component, a nationally representative sample of noninstitutionalized individuals (n = 18,295), were analyzed using bivariate χ tests and a 2-part model (logistics regression and generalized linear model regression for the first and second stages, respectively).Although predictive of any health expenditure, SRMH alone was not highly predictive of the dollar value of that health expenditure conditional on any spending. By comparison, the K6 measure was significantly and positively associated with the probability of any health expenditure as well as the dollar value of that spending. Taken together, both the K6 and SRMH measures suggest a positive relationship between poor mental health and the probability of any health expenditure and total expenditure conditional on any spending, even when adjusting for other confounding factors such as race/ethnicity, sex, age, educational attainment, insurance status, and some regional characteristics.Our results suggest that psychological distress and SRMH may represent potential pathways linking poor mental health to increased health care expenditure. Further research exploring the nuances of these relationships may aid researchers, practitioners, and policy makers in addressing issues of inflated health care expenditure in populations at risk for poor mental health.

INTRODUCTION

Mental illness persists as a pressing public health issue. Indiscriminate of demographic and socioeconomic status, mental illness has a widespread impact that affects individuals of all ages, ethnicities, and income levels. A 2014 report from the Substance Abuse and Mental Health Services Administration estimated that 42.5 million adults in the United States experienced mental illness in the past year[1]; moreover, the National Institute of Mental Health (NIMH) further reported that 1 in 17 Americans currently live with a serious mental illness.[2]

Although these statistics are startlingly high, the actual prevalence is likely higher. Although the majority of severe cases of mental illness will encounter professional treatment, many individuals fly under the radar[3]; nevertheless, just because an individual remains untreated, it does not imply that the individual does not need treatment. As a result, stated prevalence values likely underrepresent need, especially among those with moderate mental health illness.

Because many mental health conditions remain undiagnosed, self-rated assessment tools provide a useful and perhaps more revealing indicator of mental well-being. Existing literature demonstrates the reliability of both self-rated assessment tools and more formal screening tools.[4,5] When assessed by both clinical diagnoses and self-rated measures, mental illness was associated with a variety of outcomes, including physical health, health utilization, and expenditure.[6-8] However, the predominant method of measuring mental illness in the literature continues to be clinical diagnoses, a method that necessarily excludes individuals with “poor” mental health who remain undiagnosed.

As such, self-rated measures may capture a more comprehensive understanding of mental health in the general population. These sorts of measures may also help researchers better understand how mental health relates to other important outcomes such as health expenditure. The literature has shown that the association between mental health and patterns of health care cost may be more pronounced when using self-rated measures because individuals who have undiagnosed mental illness utilize similar medical services as those who are diagnosed.[9]

This article examined the impact of mental illness on health expenditure through self-rated mental health (SRMH) measures. Drawing on the 2011 household component of the Medical Expenditure Panel Survey (MEPS), we relied on 2 measures of SRMH: a Likert-style measure assessing whether an individual rates oneself to be in “excellent, very good, good, fair, or poor” mental health, which we broke into 5 separate dummy variables, and (2) the self-rated Kessler Screening Scale for Psychological Distress (K6). In defining expenditure, we considered total health care expense, including both indemnified and out-pocket payments. We hypothesized that individuals with relatively poorer SRMH experience elevated health care spending.

METHODS

Data Source

Data are cross-sectional, drawn from the publicly available Medical Expenditure Panel Survey Household Component (MEPS-HC). Specifically, we used the 2011 full-year consolidated data file (MEPS-HC 147), which contains a nationally representative sample of noninstitutionalized US civilians.[10] MEPS data have undergone review and have been approved by the RTI International Institutional Review Board, granted by the Office for Protection from Research Risks. The data included important information regarding demographic characteristics, health utilization, and health expenditures by Americans.

The data were collected using an overlapping panel design.[10,11] Each year, a new panel of sample households is selected to capture 2 calendar years of data. The 2 years of data for each panel are collected in 5 rounds of interviews. The panel design of the survey reflected the changes in respondents’ health status, income, employment, eligibility for public and private insurance coverage, use of services, and payment for care over the survey period.

Sample Population

We limit our sample to civilian, working-age adults. As such, we excluded individuals under the age of 18 and over the age of 65. We made this choice for several reasons. First, individuals over the age of 64 experience a higher proportion of physical health conditions[12] and per person personal health care spending;[13] seniors also pay for health care in a fundamentally different way (ie, they qualify for Medicare). Second, similar to seniors, returning veterans may also experience unique health issues. Importantly, the literature emphasizes the increased risk of mental health problems that many returning veterans face.[6] Taken together, this suggested that the relationship between mental health and expenditure maybe distinct for these groups. As a result, we limited our sample to working-age, adult civilian population.

Excluding individuals outside the 18 to 64 age range reduces the sample population from 35,313 to 21,510 (13,803 observations dropped). Next, we excluded individuals who indicated that they served as active duty military personnel during any of the 3 rounds (removing an additional 82 observations). Lastly, we excluded those observations with incomplete data for the relevant variables (a further 3133 observations deleted), resulting in a final sample size of 18,295.

Dependent Variable: Health Expenditure

MEPS contained a question gauging self-rated total 2011 health expenditure between January 1, 2011 and December 31, 2011.[13] Total expenditure included both out-of-pocket and indemnified spending on variety of health care costs, including inpatient visits, outpatient visits, emergency room, prescribed medicine, and other.[8] Based on this question, we created 2 expenditure measures. First, we created a binary measure, indicating whether an individual had any health expenditure over this period. Second, we created a continuous measure, which reflected the dollar value of spending.

Key Explanatory Variable: Perceived Mental Health and Psychological Stress

We included 2 measures of SRMH as our explanatory variable: the K6 and a series of dummies gauging perceived mental health. The K6 is a widely used 6-item self-administered screening scale developed by Kessler et al. The K6 measures nonspecific psychological distress by asking respondents how often they felt symptoms of mental illnesses (eg, hopelessness and depression) within the past 30 days.[14-16] Responses were categorized on a 5-point scale (none of the time, a little of the time, some of the time, most of the time, or all of the time. The scale ranges from 0 to 24, with higher K6 scores indicating the greater likelihood of psychological distress.

The K6 is endorsed as an important tool in both clinical and research practice.[17] Kessler et al[14] reported there is “excellent internal and consistent reliability” associated with the K6 (Cronbach's alpha = 0.89) – a result that has also been replicated in other studies.[18,19] As such, the K6 provides a valid, standardized estimation of the prevalence of psychological disorders among nonclinical populations.[20] Moreover, many argue that the K6 is a superior predictor of health-related quality of life compared with diagnoses.[14,20]

In addition to the K6 assessment, MEPS also asked respondents to rate their overall perceived mental health status. Specifically, the survey asked, “In general, would you say your mental health is excellent, very good, good, fair, or poor?” Based on this question, we created a series of 5 dummy variables, which gauged perceived mental health status. Research suggests that SRMH measures are a good proxy for clinical criteria. Paralleling the literature on clinical mental health diagnoses,[21] the literature highlights a positive association between “poor” self-perceived health status, depressive symptoms, and health care expenditure.[9,22,23] Given that the K6 is administered during Round 4 of Panel 15 and Round 2 of Panel 16, we drew on the mental health measure from the same periods (Round 4/2).

Other Control Variables

Additional controls included sex, age, race/ethnicity, geographic region, metropolitan statistical area (MSA), employment status, education, family income (percentage above or below the federal poverty level), marital status, insurance coverage, and perceived physical health. Our measures of perceived mental health, perceived physical health, and employment status all came from the same period (Round 4/2). During the reference period, 28 individuals indicated that they did not work, but they had job to which to return. These individuals were considered “employed.” The remaining variables reflect the final 2011 measures.

Statistical Analysis

The data were analyzed using the statistical software package Stata 13 (StataCorp, College Station, TX). To account for the complex survey design, we adhered to the weight guidelines outlined by MEPS. We used balanced repeated replication (BRR) weights for our variance estimates to adhere to the weight guidelines outlined by MEPS. As a result, all summary statistics and empirical model estimates incorporated these BRR weights. In finalizing our replicate weights, we drew from the MEPS 1996 to 2011 Replicates for Variance Estimation File (MEPS HC-036BRR).

In carrying out our empirical analysis, we first conducted a series of bivariate χ2 tests, assessing whether differences existed between those with any positive expenditure versus those with no expenditure for each categorical outcome.

Given the large number of individuals with zero health expenditure (24% of the sample population), we ran a 2-part model. In the first part of the model, we estimated the probability of any health expenditure using a logistic regression model. In detailing the results, we presented the adjusted odds ratios. In the second part of the model, we estimated health expenditure conditional on any positive spending (ie, we excluded individuals with zero health expenditure from the analysis). We analyzed this relationship using a generalized linear model (GLM) with a gamma distribution and a log link. The log-linked GLM with a gamma distribution facilitates the modeling of response variables that are positively censored at zero.[24] In addition, many have argued that GLM models may help address skewedness in health expenditure variables specifically.[25,26] Buntin and Zaslavsky[25] provide a detailed analysis of the relative benefits of using this class of models when analyzing health expenditure.

RESULTS

Descriptive Statistics and Bivariate Analyses

Table 1  provides descriptive statistics for the sample population. The typical observation is age 41, female, Caucasian (non-Hispanic), employed, married, in a middle-income bracket, has earned a high school degree, and holds private insurance. Close to 40% of those sampled rate themselves to be in “excellent” mental health. Close to 90% of observations rated themselves to be in at least “good” physical health. Almost a quarter of those sampled (24%) spent zero dollars on health expenditure in 2011. Conditional on a positive level of health expenditure, the average respondent spent $5253 total on health care (including out-of-pocket and indemnified expenses).

TABLE 1

Descriptive Statistics and Bivariate χ2 Test: Any Positive Expenditure (Yes or No) vs Categorical Outcomes

The summary statistics highlight several important distinctions between spending groups. Relative to those with positive health expenditure, a larger fraction of individuals with zero expenditure ranked themselves to be in “excellent” mental health (44% compared with 37%) and at least “good” physical health (96% compared with 86%). A larger proportion of men had zero health expenditure. As expected, those with zero health expenditure tended to be younger. The high-income bracket makes up 42% of those with positive expenditure and, by comparison, makes up only 24% of those with zero expenditure. Not surprisingly, a significant fraction of those with zero health expenditure (43%) report being uninsured.

These differences are further emphasized by a series of bivariate χ2 tests, which assessed whether significant differences exist between those with any positive expenditure relative to those with no expenditure for each categorical outcome (Table 1 ). The bivariate χ2 tests underline significant differences for all categories except MSA and employment status.

Self-Rated Mental Health and Health Expenditure, Results From the 2-Part Model

The results from the 2-part model (Table 2) suggest a nuanced relationship between health expenditure and SRMH. Compared with those who rate themselves to be in “excellent” mental health, individuals who rated themselves to be in “good” or “fair” mental health experienced significantly higher odds of any health expenditure. On the contrary, relative to those with “excellent” SRMH, individuals who rated themselves to be in “very good” or “poor” mental health did not experience significantly higher odds of any spending. Looking at the adjusted odds ratios associated SRMH dummies as a whole, the results signaled that the relationship between SRMH and the probability of any spending might be nonlinear.

TABLE 1 (Continued)

Descriptive Statistics and Bivariate χ2 Test: Any Positive Expenditure (Yes or No) vs Categorical Outcomes

Conditional on positive spending, SRMH dummy variables did not appear to be highly associated with health expenditure. Recall, the higher the K6 score, the higher the likelihood of psychological distress. Unlike the SRMH indicators, K6 was significantly and positively associated with both the probability of any health expenditure and the dollar value of expenditure conditional on a positive level of spending.

Figure 1 presents the predictive margins of total health expenditure conditional on positive spending across the quartile values of the K6 within our study sample: Q0 (minimum) = 0, Q1 (25th percentile) = 2, Q2 (50th percentile) = 2, Q3 (75th percentile) = 5, and (maximum) = 24. Confidence intervals associated with these measures are also presented. One observes a dramatic increase in expenditure as K6 score increases. Even when focusing on the interquartile range, Figure 1 highlights a marked differential between health expenditures associated with the 25th and 75th percentile K6 scores.

FIGURE 1

Predictive margins of total health expenditure conditional on positive spending. The predictive margins of total health expenditure conditional on positive spending across the quartile values of the K6 within our study sample—Q0 (minimum) = 0, Q1 (25th percentile) = 2, Q2 (50th percentile) = 2, Q3 (75th percentile) = 5, and (maximum) = 24.

Taken together, both the K6 and the SRMH measures indicated a positive relationship between poor mental health and the probability of any health expenditure and total expenditure conditional on any spending. As expected, physical health and insurance indicators were highly significant. Relatively worse physical health was associated with a higher probability of any spending and a higher level of expenditure conditional on positive spending. Conversely, compared with privately insured individuals, being uninsured was negatively associated with the probability and level of spending. Education was highly associated with both the probability of any expenditure and the level of spending. Although the income measures are largely insignificant, income is highly correlated with education, suggesting that education may be picking up most of this effect. The results further suggested that factors like race/ethnicity, gender, age, education, and some regional characteristics also shaped spending behaviors. Non-Hispanic African American and Hispanic groups experienced a significantly lower likelihood of any health expenditure relative to non-Hispanic whites. Although these differences continued in the second stage of the analysis, the result was only significant for Hispanics.

DISCUSSION

Our findings indicate an essential link between self-rated perceived “poor” mental health, K6, and total health care expenditure. Taken together, our results suggest that “poor” mental health and psychological distress, as reported by an individual, are associated with an increased probability of any expenditure and the actual level of expenditure conditional on any spending. This relationship appears to be particularly strong when looking at the K6 measure. These associations remain significant after controlling for other important covariates such as race/ethnicity, sex, age, educational attainment, and regional characteristics. Our study reinforces the findings in the literature.[27] Our work not only exposes the association between SRMH and health expenditure, but the differential impact of varying SRMH measures on this relationship.

The results can be contextualized within the larger debate surrounding health care inflation. Indeed, US health care spending grew by 3.7% in 2012, reaching $2.8 trillion or $8915 per person.[28] NIMH reported that serious mental illness alone costs the United States an estimated $193.2 billion per year; much of this cost is attributable to medications used to treat mental disorders, hospitalization, and clinic visits.[29] Although the Affordable Care Act is estimated to save the federal government approximately $20 billion over 10 years, the “health care costs remain unevenly distributed: 10% of patients account for 64% of costs”; this elevated spending is mostly attributed to patients with chronic conditions.[30] Even when controlling for co-occurring physical health issues, mental illness still predisposed an individual toward elevated expenditure.[8] Despite spending more on health care than other economically advanced countries, Americans are not much healthier.[31]

Within this context, SRMH measures may be a helpful resource in identifying those individuals who are both in need of mental health resources and at risk of elevated health care expenditure. Like any disease that remains untreated, mental illness can worsen over time, leading to higher costs of treatment[32,33] and lower quality of life.[34] Although the majority of severe cases will seek professional advice and/or take medication, more moderate cases may slip through the cracks.[3] Catching these individuals early may enhance the relative efficacy of mental health treatment and, as our results suggest, potentially impact broader health expenditure. In a recent study, Jang et al[35] found SRMH to be significantly predictive of health service use and beneficial in raising one's awareness of a mental health problem. Such results further support our findings, which highlight the potential usefulness of SRMH measures both in identifying unmet and high needs cases and in predicting those at risk of increase health care spending.

Our study has several limitations. First, our data are cross-sectional, which limits our ability to make causal inferences. That said, our data are nationally representative, which aids the external validity of our study. Second, given that our data are cross-sectional, we cannot discuss the impact of an individual's cumulative mental health history on expenditure across the life span. Finally, mental health can be assessed using different instruments; these instruments may be more or less sensitive to varying aspects of a given condition. As a result, using different mental health measures may result in different findings. That said, we do draw on several mental health measures, which diversifies this risk. Moreover, each of these SRMH measures is positively associated with the probability and level of spending.

Despite these limitations, this study offers new information on the association between SRMH and health care expenditure, while controlling for other important factors influence this relationship. Building upon previous research, our findings provide further rationale for investments in self-rated assessments tools. Such tools will help identify the marginal patient in need of care who might otherwise go untreated, impacting not only the trajectory of treatment and patient health, but also the long-term costs. Health care providers, employers, researchers, and policy makers may want to consider the elevated risks of increased health care expenditure among persons with poorer SRMH in developing policies and interventions to contain health care expenditures.

TABLE 2

Two-Part Model, Total Health Expenditure in 2011