INTRODUCTION

The coexistence of depression and diabetes is associated with increased mortality and morbidity, including higher rates of complications, increased disability and lost productivity, lower quality of life, and increased risk of death.1 Compounding this impact is the high prevalence of diabetes and depression, with a meta-analysis showing major depression occurring in 11 % of individuals with diabetes and clinically relevant depression in 31 %.2 In addition, 45 % of diabetes patients are estimated to have undiagnosed depression,3 and patients with diabetes are more likely to be diagnosed with depression than those without diabetes,4 suggesting that the influence of depression on the outcomes of patients with diabetes may be greater than prevalence numbers alone suggest.

Major depression and depressive symptoms in patients with diabetes are also consistently associated with decreased income, increased healthcare utilization, and higher healthcare costs.1,5 Studies have shown that healthcare expenditures for patients with diabetes and depression are 2 to 4.5 times as high as those of patients without depression.6,7 Differences between those with and without depression include higher diabetes-related costs, more time at inpatient facilities, more ambulatory care visits, and more prescriptions.8–10 Patients with diabetes and "probable depression" have also shown increased healthcare costs compared to patients without depression,11 and a study of depression symptom severity found that healthcare costs increased as the severity of depressive symptoms increased.7

These costs are even more important when considered at a population level, given the high prevalence and expense of diabetes to the US population.12 Though the recognition and treatment of depression for patients with diabetes is consistently recommended, it is less than optimal.13 Studies show that only 50 % of diabetes patients with depression are recognized in primary care settings, and diagnosis does not change treatment patterns.13 One study examined the effects of implementing a depression intervention for diabetes patients in nine primary care practices, and found a 5-year mean total cost reduction of $3907 for patients who received the intervention compared with patients receiving usual care.14 A systematic review of the cost-effectiveness of psychological interventions to treat depression in patients with diabetes also found net savings, though few such studies existed.15

While research has consistently shown increased costs of depression in patients with diabetes, we are not aware of any studies investigating national estimates of costs for both symptomatic and asymptomatic patients, or studies providing trends over time for these costs. The present study used the Medical Expenditures Panel Survey (MEPS) to investigate the impact of depression on healthcare costs in a nationally representative sample of patients with diabetes. Patients were categorized as having no depression, unrecognized depression based on a validated depression screening questionnaire, asymptomatic depression, or symptomatic depression. This is the first study of its kind, and it adds significantly to the literature by considering the healthcare cost differences between patients with these four categories of depression over a period of 8 years, using a novel cost methodology to provide national estimates.

RESEARCH DESIGN AND METHODS

The study used data from the 2004–2011 Medical Expenditure Panel Survey (MEPS) to estimate the cost of depression among adults with diabetes (aged ≥18 years).

Data

MEPS provides nationally representative estimates of healthcare use and expenditure, and is maintained and co-sponsored by the Agency for Healthcare Research and Quality (AHRQ).16,17 It collects detailed information on individual sociodemographic characteristics, health conditions, healthcare use and expenditures, sources of payment, and health insurance coverage.16 MEPS has three components: Household Component (HC), Medical Provider Component (MPC), and Insurance Component (IC).16,17 Information on the HC is collected by self-reports, and the MPC requests data on medical and financial characteristics from hospitals, physicians, home healthcare providers, and pharmacies in order to validate and supplement information received from the MEPS-HC respondents.16,18 Diagnoses coded according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) are also collected as part of the MPC.

The panel design of the MEPS survey includes five rounds of interviews covering two full calendar years.16 Each panel collects data from different individuals, so merging of panel data creates a pooled cross-sectional sample. We used 8 years of MEPS data to create a pooled sample as follows: 2004/2005, 2006/2007, 2008/2009, and 2010/2011. Adults with diabetes were included based on patient self-reports. MEPS does not distinguish between type 1 and type 2 diabetes. Based on data from the Centers for Disease Control and Prevention, 5 % of diagnosed cases are type 1 diabetes and 90–95 % of diagnosed cases are type 2 diabetes.12

For each year, we merged data for adults with diabetes from the household survey of the medical condition files and full-year consolidated files using the unique person identifier on a one-to-one match.18 The pooled sample has a common variance structure necessary to ensure compatibility and comparability of variables within the complex sample design.19 Analytic sampling weight variables were adjusted by dividing by the number of years pooled. The sum of these adjusted weights represented the average annual population size for the pooled period, and reflected an “average annual” basis rather than the entire pooled period.20 As a result, analyses account for the sampling weights, clustering, and stratification design to estimate the nationally representative aggregate and incremental healthcare expenditure for the US population.16,17

Medical Expenditures

The dependent variable was the direct cost calculated as total healthcare expenditures for the calendar year of each observation. More specifically, in MEPS, an expenditure is defined as the sum of direct payments for care provided during the year, including out-of-pocket payments and payments by private insurance, Medicaid, Medicare, and other sources.16 The total medical expenditure comprises office-based medical provider, hospital outpatient, emergency room, inpatient hospital (including zero night stays), prescription medicine, dental, home healthcare, and other medical expenditures.16 The 2004–2011 direct medical costs were adjusted to a common 2014 dollar value using the Consumer Price Index provided by the US Bureau of Labor Statistics (http://data.bls.gov/cgi-bin/cpicalc.pl).

Depression

The primary independent variable was four mutually exclusive depression categories created from ICD-9-CM codes and the Patient Health Questionnaire-2 (PHQ-2) depression screening tool. Previously validated ICD-9-CM codes 296, 300, 309, and 311 associated with at least one office-based medical provider, hospital outpatient, or emergency room visit, inpatient hospital stay, or prescription medication were used to identify individuals with clinical depression.21,22 Responses to the PHQ-2 were used to identify depression symptoms. The PHQ-2 is a two-question self-report scale included in the MEPS questionnaire. It was designed to inquire about an individual's depressed mood over the prior 2 weeks by asking questions about whether the respondent had little interest or pleasure or felt down/depressed/hopeless. Answer options for each range from 0 to 3, with total scores of 0 to 6 representing a summation of the values, and higher values indicating a greater tendency towards depression.23 The PHQ-2 using a cutpoint of ≥3 has sensitivity of 83 % and specificity of 92 % for identifying major depression, and is often used for screening of depressive symptoms.23 The scale was developed to indicate the probability of depression, and is not intended to measure severity or to represent a diagnosis of depression. As such, in this analysis it is used as an indicator of the presence or absence of symptoms as measured by the scale. Two dummy variables were created for each patient using the ICD-9 diagnosis and PHQ-2 cutpoint of ≥3, below which the number of symptoms as measured by the scale indicated a low probability of currently screening for depression.

ICD-9 and PHQ-2 measures for the same calendar year were used to create four categories:

  1. 1.

    No depression, defined as individuals with neither depressive symptoms as determined by the PHQ-2 nor ICD-9 diagnosis of depression;

  2. 2.

    Unrecognized depression, defined as individuals with PHQ-2 depressive symptoms but without ICD-9 diagnosis of depression;

  3. 3.

    Asymptomatic depression, defined as individuals with ICD-9 diagnosis of depression (diagnostic codes 296, 300, 309, 311) but without PHQ-2 depressive symptoms; and

  4. 4.

    Symptomatic depression, defined as individuals with both PHQ-2 depressive symptoms and clinical depression by ICD-9 diagnosis (codes 296, 300, 309, 311).

Categories do not represent severity of depression, but rather the presence or absence of a diagnosis (as defined by the ICD-9 codes) and/or the presence or absence of symptoms (as defined by the PHQ-2 scale).

Covariates

Adjusted analyses used covariates collected through self-reports. Covariates were selected based on the literature to include variables that account for possible variation in medical expenditures or that address disease burden. Binary indicators of comorbidities were based on responses to a series of questions asking whether respondents had ever been diagnosed with specific diseases. Cardiovascular disease (CVD) thus indicates a positive response to diagnosis with coronary heart disease, angina, myocardial infarction, or other heart diseases. Race/ethnicity, age, education, health insurance, and marital status were each divided into three categories (see Table 1). Census region and metropolitan statistical area (MSA) were based on residence as of the end of the year. Income was defined as a percentage of the poverty level and was grouped into four categories.

Table 1. Sample Demographics by Depression Categories Among Adults with Diabetes
Full size table

Statistical Analyses

Descriptive statistics were used to determine sociodemographic and comorbidity variables across the four depression categories among adults with diabetes over the 8-year pooled dataset. Unadjusted mean and standard pairwise comparisons were used to compare the total healthcare expenditures by depression categories.24 Mean expenditures between the six paired groups were compared by sociodemographic and comorbidity variables as well as by year.

Adjusted healthcare expenditures were estimated using a two-part model.25,26 In the two-part model, a probit model was estimated for the probability of observing a zero versus positive medical expenditure. Positive medical expenditure is defined as any healthcare expenditure greater than zero.26 Conditional on having a positive medical expenditure, a generalized linear model (GLM) was estimated.27,28 GLM was used instead of log OLS [ordinary least squares] regression, since it relaxes the normality and homoscedasticity assumptions and avoids bias associated with re-transforming to the raw scale.27 The results of the modified Park test verified that the use of a gamma distribution with a log link was the best–fitting GLM for consistent estimation of coefficients and marginal effects of medical expenditures.28,29 Multicollinearity was checked for predictors of the two-part model, taking into account the complex survey design. The variance inflation factor (VIF) for all predictors used in the two-part model indicated no multicollinearity problems. The F-test for the two-part regression models was found to be significant, which indicated the overall significance of the regression model.

Two-part models have been widely employed in situations where, due to a large number of non-users of health services, there are excess zeros in the resource use or cost data, and the assumption of normality of the error term is not satisfied.30 Marginal effects were then estimated from the combined first and second parts of the final model.27 To control for confounding, sociodemographic factors including age, sex, race, marital status, education, health insurance, metropolitan statistical area status, region, income level, and comorbidities were included in the fully adjusted model. All analyses were performed at the individual level using STATA 13 software (StataCorp LP, College Station, TX, USA). Statistical significance of p < 0.05 was used.

RESULTS

Demographic Characteristics

The resulting dataset was an unweighted sample of 15,548 adults with diabetes (weighted sample of 17,465,579). Among the adults with diabetes in the pooled sample, 10,473 (67.3 %) individuals had no depression, 1579 (10.2 %) had unrecognized depression, 2117 (13.6 %) had asymptomatic depression, and 1379 (8.9 %) had symptomatic depression. Demographic characteristics overall and by depression category are shown in Table 1. Unrecognized depression was more likely in the 65–85 age group, minorities, those with less than high school education, and low-income individuals.

Unadjusted Cost Differences for Depression Categories Across Years

The results of unadjusted mean medical expenditures attributable to the four depression categories over time are shown in Table 2. The overall mean medical expenditures for patients with diabetes were $10,016 (95 % CI 9589–10,442) for those with no depression, $15,155 (95 % CI 13,587–16,723) for those with unrecognized depression, $16,134 (95 % CI 14,885–17,382) for those with asymptomatic depression, and $20,105 (95 % CI 18,103–22,106) for patients with symptomatic depression. Pairwise comparison tests showed significant differences in mean expenditures between all paired groups except those between asymptomatic and unrecognized depression.

Table 2. Total Healthcare Expenditures (Mean and 95 % CI) by Depression Category Among Adults with Diabetes (Reported in 2014 Dollars)
Full size table

Figure 1 shows the mean expenditures per person per year by the four depression categories. The no-depression category showed little change across the years, and remained the lowest of the four categories. Symptomatic depression showed a decrease over time, but a higher sustained cost than any of the other categories. Unrecognized and asymptomatic depression showed decreases from 2004/2005 to 2006/2007, but gradual increases towards 2010/2011.

Figure 1
figure 1

Medical expenditure by depression and diabetes categories, 2004–2011.

Full size image

Figure 2 shows the mean expenditures per person per year by the four depression categories broken down by total, inpatient, outpatient, prescription, office-based, and emergency room costs. No depression was consistently the lowest cost category, and symptomatic depression the highest in all cost types. Unrecognized depression showed costs similar to but lower than asymptomatic depression, with the exception of inpatient and outpatient costs. The two highest cost types were inpatient and prescription costs.

Figure 2
figure 2

Annual mean expenditures by depression categories, 2004–2011.

Full size image

Adjusted Incremental Cost Differences for Depression Categories Among Diabetes

The results of the adjusted two-part GLM on the incremental costs associated with the four depression categories, sociodemographic factors, comorbidities, and time trends are shown in Table 3. In the pooled sample, after adjusting for sociodemographic factors, comorbidities, and time trend covariates, the incremental cost of unrecognized depression increased by $2872 (95 % CI 1660–4084), asymptomatic depression increased by $3347 (95 % CI 2568–4386), and symptomatic depression increased by $5170 (CI 95 % 3610–6731) compared to patients with no depression.

Table 3. Two-Part Regression Model: Incremental Effects of Healthcare Spending by Depression Categories Among Adults Accounting for Healthcare Expenditure (Reported in 2014 Dollars)
Full size table

Estimated US Burden of Depression Among Adults With Diabetes

Finally, we estimated the aggregate cost during the period 2004–2011 among adults with diabetes. Based on the unadjusted mean, the average annual aggregate costs for the US population with diabetes were estimated at $21.3 billion for unrecognized depression, $42.9 billion for asymptomatic depression, and $30.3 billion for symptomatic depression. The adjusted total incremental costs increased by $4.0 billion for unrecognized depression, $9.2 billion for asymptomatic depression, and $7.8 billion for symptomatic depression, compared to those with no depression.

CONCLUSIONS

Based on a national sample of adults with diabetes, mean medical expenditures were $5000–6000 higher for those with unrecognized and asymptomatic depression than with no depression, and twice as high for those with symptomatic depression. Unrecognized depression showed higher expenditures than asymptomatic depression for inpatient and outpatient costs, while asymptomatic depression showed higher prescription and office-based costs than unrecognized depression. Adjusted analyses revealed that expenditures were still $2000–3000 higher for unrecognized and asymptomatic depression than for no depression, and $5000 higher for symptomatic depression. These higher medical expenditures persisted over time.

This study is the first of its kind, showing the impact of healthcare costs over time in patients with diabetes and different classifications of comorbid depression (unrecognized, asymptomatic, and symptomatic). The total economic burden of depression alone among the US population rose to an estimated $210.5 billion in 2010.31 Results of this study show the additional impact of unrecognized comorbid depression in patients with diabetes at the individual and national levels ($4 billion at the population level). Asymptomatic depression also costs significantly more than patients with no depression ($9 billion for the US population). This economic burden is important to consider given the high prevalence of depression in patients with diabetes, the epidemic of diabetes in the United States, and the persistent costs over time shown by this study. Estimates in this study show that depression is a costly comorbidity, with only CVD and stroke showing increased costs comparable to any of the depression categories for patients with diabetes. Recent work on diabetes alone and chronic kidney disease show that the costs for all individual comorbid depression categories are higher than those for diabetes alone ($2558 per year), but lower than the costs of chronic kidney disease ($8473 per year).32,33

These results support the call for improved recognition of depression through increased screening at the primary care provider level.1,13,34 Comorbid depression is associated with decreased adherence to treatment, possibly leading to poorer glycemic control, higher complication rates, and increased healthcare use and cost.1 A systematic review of depression treatment for individuals with diabetes found that interventions were cost-effective compared to usual care.35 Therefore, increased use of interventions to address depression in patients with diabetes may result in a reduction in costs over time for symptomatic patients, but may not be sufficient to remove symptom load. Unrecognized and asymptomatic depression costs individual patients nearly $3000 more per year, and are not as often considered in interventions. While collaborative care programs have been shown to be moderately cost-effective, the offset to costs has not been enormous.7,14,15,36 In addition, increased mental health visits for patients with diabetes and depression were effective in reducing inpatient costs, but the savings did not offset outpatient and pharmacy costs.37 Based on this study, some costs associated with asymptomatic depression are the result of receiving treatment (higher for prescription and office-based visits than unrecognized depression). While the benefit of treatment to individuals is worth the offsetting costs, providers and policymakers should strive to develop alternatives that allow for treatment but also provide savings to the system overall. Strategies are needed to focus interventions on identifying patients that are symptomatic and in need of more aggressive treatment, in addition to identifying patients with unrecognized depression—for example, co-located primary care and telemental health.37

This study is strengthened by the use of nationally representative data, a large sample size, and a novel cost methodology; however, some limitations do exist. As with all pooled cross-sectional data, this study cannot show causality. Diabetes prevalence was based on self-reports rather than medical diagnosis, although previous studies have acknowledged the reliability of self-reporting of diabetes.38,39 In addition, self-reported comorbidities may not include all relevant comorbidities, and may include those that are not medically diagnosed. Finally, the PHQ-2 is a screening tool for major depression and requires further assessment using a diagnostic interview. Nevertheless, it a validated screening method, and has shown high sensitivity and specificity for depression.

In conclusion, using a pooled dataset over 8 years, we found that unrecognized and asymptomatic depression in patients with diabetes each resulted in an average of $3000 higher costs per year compared to patients without depression. Higher medical expenditures persisted over time, with only symptomatic depression showing a sustained decrease.