Psychological Distress and Trends in Healthcare Expenditures and Outpatient Healthcare
Published Online: May 23, 2011
Paul A. Pirraglia, MD, MPH; John M. Hampton, MS; Allison B. Rosen, MD, ScD; and Whitney P. Witt, PhD, MPH
A critical first step in assessing the value of healthcare in the United States as it relates to mental health is determining whether trends in mental health symptoms have had an impact on changes in healthcare expenditures over time. It is well recognized that healthcare costs have been increasing,1 with an average growth rate of 5.4% in healthcare expenditures between 1993 and 2000.2 Mental health spending has increased as part of total spending, though at a lower rate.3 Spending on prescription drugs to treat mental health and substance abuse conditions increased markedly between 1996 and 2000.4 Mental health conditions—including depression,5-7 anxiety,5,8 bipolar disorder,9-11 and schizophrenia12have been linked to higher healthcare costs. These higher costs are not attributable solely to higher mental health costs. While some reports suggest that the prevalence of mental health conditions in the United States is stable,13 there is also evidence that mental distress has increased over time.14
Given the increased use of prescription medications to treat depression, anxiety, psychotic illnesses, and other mental health conditions, one might expect that while the prevalence of mental health conditions might not change, the overall burden of mental health symptoms might decline over time. Therefore, since healthcare costs are known to be increasing, a trend toward improving value of healthcare as it relates to mental health would be reflected by a relationship between mental health symptoms and expenditures; that is, either the proportion of those with elevated mental health symptoms would decrease over time or the spending for those with a greater burden of mental health symptoms would drop over time relative to those individuals with fewer symptoms.
Quantifying how mental health plays a role in changing healthcare costs is not easily done. Consideration of the impact of diagnosed mental health conditions over time may be influenced by changes in screening and in recognition, as well as shifts in nosology. Examining only diagnosed or self-reported conditions also misses the impact of mental health conditions that are undiagnosed or unreported. Furthermore, how mental health may be related to healthcare expenditures is not solely determined on the basis of having a mental health condition, but on the burden of mental health-related symptoms; that is, severity of symptoms may bea key determinant for related expenditures, which has been shown, for example, in depression15 and bipolar disorder.16 Because healthcare expenditures and utilization in those with known mental health conditions tend to be higher, overall costs must be evaluated.
We sought to examine trends in overall mental health and subsequent healthcare expenditures in the United States by using linked data from 2 nationally representative surveys of noninstitutionalized individuals, the National Health Interview Survey (NHIS) and the Medical Expenditure Panel Survey (MEPS). These linked surveys are well suited to address trends in mental health and expenditures. Starting in 1997, a 6-item scale of the Kessler Psychological Distress Scale (K6), a shortened form of a previously developed 10-question psychological distress scale, was introduced into NHIS, which provides a measure of psychological distress that has been validated with respect to mental health diagnoses.17 Starting in 1996, MEPS used the NHIS sampling frame to track national health expenditures. Using these linked data sets, we sought to examine whether there has been a change over time in psychological distress in the United States and whether adjusting for factors known to be associated with mental disorders would impact a trend in psychological distress if it were observed. We also sought to examine whether there were trends in healthcare expenditures and outpatient visits associated with the level of psychological distress, with the expectation that good value in healthcare related to mental health symptoms would be reflected by either a decrease over time in the population burden of psychological distress or a reduction over time in subsequent healthcare costs and expenditures
among those with greater psychological distress.
Data Source and Study Sample
We performed a sequential cross-sectional study wherein we examined 8 consecutive years of data. Data are from NHIS and MEPS, both of which provide nationally representative data. NHIS data are collected on a new sample each year, and MEPS uses a subset of this sample in the 2 subsequent years. While the MEPS data are a subset of the NHIS sample, MEPS still uses a survey sample design such that the weighted values are nationally representative, which minimizes selection bias that might occur with selection of the MEPS subsample. 18 We obtained NHIS data from 1997 through 2004, which provided information on psychological distress (described below), and the 2 subsequent years of MEPS data for each of these samples to obtain subsequent utilization and expenditure data. We limited our study to those persons 18 years and older, resulting in weighted data representing 25,618,369 adults. The study was exempted from institutional review board approval as the data were publicly available and participants could not be identified from the data.
Psychological distress was measured using the K6, a 6-item scale of the Kessler Psychological Distress Scale. It was developed for use in general-purpose health surveys because it is short, has strong psychometric properties, and can discriminate Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) cases from noncases with consistency across sociodemographic subsamples.19 The K6 was demonstrated to detect more than 90% of DSM-IV diagnoses as determined by the Structured Clinical Interview for DSM-IV criteria.17 The K6 asks how often during the past 30 days the respondent felt nervous, hopeless, restless, or fidgety; so depressed that nothing could cheer them up; or that everything was an effort; or worthless. Responses are on a 4-point scale (none of the time, a little of the time, most of the time, or all of the time). The sum of the response codes for the 6 items gives a score with a range of 0 to 24, with higher scores indicating greater psychological distress. We used cut-points previously reported in the literature to stratify K6 scores into no/low psychological distress (0-6), mild-moderate psychological distress (7-12), and severe distress (13-24).17,20 Of note, the threshold K6 score of less than 12 versus 13 or greater was found to be optimal for discerning a serious mental illness diagnosis, with a total classification accuracy of 0.9220; this definition has been used in other reports as well.
To better characterize the population and to account for potential confounders in the relationship between expenditures and psychological distress over time, we obtained a number of variables from the baseline (ie, NHIS) data. These characteristics included age, sex, marital status, race/ethnicity, yearly income, insurance status, medical comorbidity, and psychiatric diagnoses. Our choice of potential confounders was primarily based on prior literature regarding affectivedisorders. Age, race, and income were related to affective disorders in the National Comorbidity Survey,21 and sex and marital status were related to affective disorders in the National Comorbidity Survey Replication.22 The relationship between greater medical comorbidity and depression is well recognized.23
Age was divided into decades (18-29, 30-39, 40-49, 50- 59, 60-69, 70-79, 80 ). Race/ethnicity was coded as white, black, nonblack Hispanic, and other. Marital status was married, single, separated, divorced, widowed, or unknown. We categorized yearly income relative to poverty status: less than 200% Federal Poverty Level (FPL), 200%-400% FPL, more than 400% FPL, and unknown. Health insurance status was divided into private, public only (ie, not covered by a commercial insurer and covered by a federal, state, or local program), or none. We determined medical comorbidity using the Charlson Comorbidity Index applied to the International Classification of Diseases, Ninth Revision (ICD-9) codes for reported medical conditions.24 We wanted to account for the presence of psychiatric conditions, which were broadly captured (either present or absent) as affective disorders; psychoses, schizophrenia, and related disorders; and anxiety, somatoform, dissociative, and personality disorders (see the eAppendix at www.ajmc.com for the ICD-9 codes for each of these categories).
Expenditures and Visits
We examined (1) total expenditures (which include inpatient, outpatient, office-based, emergency department, home health, prescriptions, dental, and other expenditures), (2) outpatient and office-based expenditures, and (3) outpatient and office-based visits. The total expenditures category is composed of all health services expenditures associated with office-based visits, hospital outpatient visits, emergency department visits, inpatient hospital stays, dental visits, home healthcare, prescription medicines, vision aids, and other medical supplies and equipment. Note that we will combine outpatient and office-based categories and herein refer to these simply as outpatient. As we had 2 years of follow-up data, we chose to handle these as the per person average of the 2 years. Expenditures were annualized and expressed in 2006 dollars.
We initially examined how demographic and clinical characteristics, total expenditures, outpatient expenditures, and outpatient visits differed across the 3 strata of psychological distress (K6 scores 0-6, 7-12, or >12). To do this, we used analysis of variance for normally distributed continuous values, the Mann-Whitney U test for non–normally distributed continuous values, and c2 tests for categorical data.
We then examined our data for trends over time using bivariate linear regression models where each characteristic was treated as the dependent variable and the year was handled as the independent variable. Similarly, to determine whether psychological distress changed over time, we examined for a linear trend in mean K6 by year in a simple regression model with K6 as the dependent variable and year as the independent variable. To assess whether other factors influenced the trend in psychological distress, we added to the model covariates (age, sex, marital status, race/ethnicity, yearly income, insurance status, medical comorbidity) that have been observed to be associated with various mental health conditions. In addition, we created a polytomous model with the 3 K6 groups as the dependent variable and time as independent variable while adjusting for these other factors.
For expenditures and utilization, we performed 3 sets of regression analyses, 1 set for each: (1) total expenditures, (2) outpatient expenditures, and (3) outpatient visits. To test trends, we first examined a simple model examining whether there was a significant relationship between expenditures and strata of psychological distress (which was expected, with greater expenditures for those with greater psychological distress) and between expenditures and year. We then tested the interaction of strata of psychological distress and year in a model that also included these as main effects. Lastly, we ran a complete model with the year by psychological distress interaction fully adjusted for potential confounders. Because we did not wish to exclude those with zero expenditures over the 2-year study period, we used fully adjusted 2-part models25 to estimate the mean value of total and outpatient expenditures by psychological distress strata. In part 1 of the 2-part model, we calculated predicted probabilities of incurring any expenditure using logistic regression. In part 2 of the 2-part models, we transformed the non-zero expenditure data using the natural log and accounted for heteroscedasticity by including Duan’s smearing estimator in the calculation of the retransformed expected values. Results from part 1 (predicted probability of incurring any expenditures vs none) and part 2 (predicted expenditure conditional on having any expenditures) were multiplied to produce each person’s expected expenditures.
PDF is available on the last page.