National Trends in the Use of Antidepressants Between 1997 and 2009 and the Role of Medicare Part D Implementation

We used multiyear data (1997–2009) from the Medical Expenditure Panel Survey Household Component (MEPS-HC), which includes five rounds of interviews over two and a half years involving a nationally representative sample of noninstitutionalized U.S. residents (11). The data include information on health care utilization (type, frequency, expenditure, and payer) and sociodemographic characteristics of respondents and their families.

We identified individuals with depression by using MEPS-HC medical conditions files. The survey asks respondents to describe their medical conditions. From this self-reported information, professional coders recorded individuals’ medical conditions by using the International Classification of Diseases, ninth revision, Clinical Modification (ICD-9-CM). Persons with ICD-9-CM codes 296 or 311 recorded for any round of interviews was defined as having had depression during that calendar year (12).

The MEPS-HC prescribed medicines file provides self-reported information on medication use. We identified antidepressants by drug names, including TCAs (amitriptyline, amoxapine, clomipramine, desipramine, doxepin, imipramine, maprotiline, mirtazapine, nortriptyline, and protriptyline); SSRIs (citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, and sertraline); and others (bupropion, duloxetine, nefazodone, selegiline, trazodone, and venlafaxine) (12). We constructed a binary variable representing whether the respondent had at least one of these medications during a calendar year.

A key variable of the analysis was insurance coverage. We categorized the type of insurance into four mutually exclusive groups: Medicare, Medicaid, dual coverage for Medicare and Medicaid (dual eligible), and private coverage. We excluded uninsured people, whose pattern of health care service use may be so different from that of insured populations as to make ready comparison difficult.

Covariates used in the analysis were age, race (white, black, and Hispanic), gender, education, family income, and health risk. The education variable was classified as zero to 12 years, 13 to 15 years, and 16 or more years. Family income was grouped as ≤124% of the federal poverty level (FPL), 125%−199% of the FPL, 200%−399% of the FPL, and 400% or more of the FPL. Health risk was measured by chronic condition indicators and the number of prescription drugs taken by a patient. [Descriptive statistics of all covariates reported by insurance group are available online in a data supplement to this report.]

We analyzed the trends in antidepressant use over time by using logistic regression controlling for sociodemographic factors. We included dummy variables for year to capture the trend in treatment between 1997 and 2009. Using 1997 as the base year, we obtained the adjusted odds ratios (AORs) of antidepressant use for all subsequent years. The analysis was conducted for each insurance group. Standard errors were clustered within an individual in all analyses to account for repeated interviews of the individual that occurred because of the panel survey design.

Next, we examined the effect of Part D on the use of antidepressants among Medicare beneficiaries by estimating a logistic model with an interaction term between the Part D and Medicare indicators. The Part D indicator equaled 1 if the year was 2006 or after and 0 if the year was before 2006. The Medicare indicator equaled 1 if the patient was covered by Medicare and 0 if not. Our main interest was the coefficient of this interaction, which indicated whether the changes in antidepressant use after Part D were significantly different among Medicare beneficiaries than among other groups. This is known as the difference-in-difference (DD) estimator. We obtained DD estimates for Medicare and all other insurance groups combined as well as for Medicare and each of the other three insurance groups. We controlled for all patient characteristics, and we adjusted standard errors for clustering within an individual.

The percentage of people using antidepressants increased in all insurance groups between 1997 and 2009. The Medicare group had a large increase in 2006 but showed a somewhat stable trend after that, while the privately insured group had a consistent increase from the early 2000s up to 2009. The Medicaid group showed a relatively slow increase compared with the other insurance groups. [A figure depicting unadjusted trends in antidepressant use by insurance coverage is available online as a data supplement to this report.]

Table 1 presents the results of a regression analysis of the trend in use of antidepressants by insurance group. The estimates by year dummies indicated increasing use of antidepressants during the study period in all groups. The odds of use of antidepressants in the Medicare group significantly increased in 2006 relative to 1997 (AOR=2.52), potentially reflecting the impact of Part D. However, AORs remained relatively stable from 2007 through 2009 (between 2.69 and 3.30). The odds of antidepressant use increased significantly in the Medicaid group starting in 2005 and in the dual-eligible group starting in 2004. AORs in the private coverage group increased continuously from 2002 up to 2009 and reached their highest levels in 2008 (AOR=4.81) and 2009 (AOR=6.63).

For other covariates, older people were more likely than younger people to take antidepressants in the Medicaid and private coverage groups. Whites had significantly greater odds of using antidepressants than blacks in the Medicare and privately covered groups. The effects of education and income were mostly insignificant.

Table 2 shows the results from the regression examining whether use of antidepressant treatment after Part D increased more in the Medicare group than in the other groups. With Medicaid as the control group, the analysis indicated that the interaction term between Part D and Medicare indicators was significant (AOR=1.35). This implies that the increase in the odds of taking antidepressants after Part D was significantly larger among Medicare beneficiaries than in the Medicaid group. The interaction term in other analyses was insignificant.

The use of antidepressants between 1997 and 2009 increased in all insurance groups. This overall increasing trend is encouraging; however, our analysis indicates that the rate of utilization and the degree of the increase over time differed by insurance group. The Medicaid group showed a lower rate of utilization and a slightly slower increase in utilization over time than were seen in other groups. Our study did not explore reasons for these differences or the appropriateness of utilization in each group. However, the assessment of the treatment pattern is the first step toward ensuring the use of effective services. Further research is necessary to assess the need for antidepressant treatment and the level of appropriate use in each insurance group.

Our analysis showed that Part D led to a greater use of antidepressants among Medicare beneficiaries, compared with the Medicaid group. This positive effect of Part D on antidepressant treatment is consistent with the prior literature (9,10). However, the large increase seems to have occurred mostly during the initial year of Part D, which suggests that further evaluations of Part D are needed to better understand its long-term impact.

We did not find a significant increase in the use of antidepressants among Medicare beneficiaries after Part D compared with dual-eligible or privately covered populations. This suggests that there may have been interventions to improve antidepressant use in those non-Medicare groups during the post–Part D period. Identifying mechanisms through which other insurance (non-Medicare) groups improved drug utilization is an important topic for future research.

A few limitations of the study should be noted. First, we relied on self-reported medical conditions to identify patients with depression, which raises a concern that receiving antidepressant treatment was a determinant of self-reporting depression. If this were the case, we would have observed an increase in the number of people with depression but little or no increase in the rate of antidepressant use. However, our data indicate that the depression population did not grow as antidepressant use increased over time, suggesting that such a concern was unlikely.

Second, our analysis was limited to the insured population. The use of depression treatment is likely to be low among the uninsured, a population that needs extensive attention and interventions. However, we excluded persons without insurance because they may exhibit systematically different characteristics that do not allow for appropriate comparison with the insured.

Third, although MEPS is a methodologically rigorous survey, it is subject to reporting errors, potentially because of recall problems among the respondents. However, there is no conclusive evidence that those errors systematically differ among insurance groups or change over time, and the impact of reporting errors on our findings of the trends, if any, was likely to have been insignificant.

The use of antidepressants among people with depression increased continuously in all insurance groups up to 2009. Increases after Part D were larger among Medicare beneficiaries than among Medicaid beneficiaries. The different patterns of utilization and trends over time among insurance groups suggest that further research is needed to explore the reasons for those differences and to assess levels of appropriate use of antidepressants by insurance group.

The authors thank Dennis Shea, Ph.D., for thoughtful comments on an earlier version of this article.

The authors report no competing interests.