The framework used for evaluating the economic burden of adults with MDD drew largely on the approach used in our prior study . The main elements of the methodology, including any key changes, are described in the following sections. A schematic outlining the methodological framework of this study is available in the Electronic Supplementary Material (ESM)-1.
Prevalence and Major Depressive Disorder (MDD) Severity Data
The National Survey on Drug Use and Health (NSDUH), a national probability sample of the US adult civilian, noninstitutionalized population, was used to compare prevalence rates of MDD by sex, age, employment, and treatment status for 2010 and 2018. In the NSDUH depression module, adult respondents were asked questions adapted from the National Comorbidity Survey Replication to assess the rate of individuals with a past-year major depressive episode according to Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV criteria (for respondents in 2010) and DSM-5 criteria (for respondents in 2018) [9,10,11].
The depression module of the NSDUH also included questions from the Sheehan Disability Scale (SDS) to assess the impairment of individuals with MDD. This self-report tool evaluates functional impairment in the domains of work/school, social life, and family life and measures severe impairment as a maximum score of ≥ 7 in any SDS domain [10, 12].
Cost Data: Sample and Control Group
The OptumHealth Reporting and Insights administrative claims database was used to compare the characteristics and costs of patients with MDD against those without MDD for 2010 and 2015 (the most recent year for which comprehensive cost and disability data were available). This private, de-identified insurance database was chosen for its robustness and breadth. In particular, it includes comprehensive information regarding patient age, sex, enrollment history, plan type, medical diagnoses, dates and place of service, payment amounts, and pharmacy claims for over 19 million beneficiaries (i.e., employees, spouses, and dependents) from 84 large, self-insured US companies in a broad range of industries with locations in all census areas of the USA . The prior research relied on claims data where diagnoses were recorded based on International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM) codes, but the current study identified patients with MDD for analysis if they had at least two ICD-9-CM or ICD-10-CM claims for MDD occurring on different dates (ICD-9-CM codes: 296.2, 296.3; ICD-10-CM codes: F32.0–F32.5, F32.9, F33.0–F33.42, F33.9). Patients in each study year were required to have continuous healthcare eligibility. Patients with MDD who had health maintenance organization, capitated, or Medicare coverage were excluded from the analysis because payment information may not have been complete among these groups of patients. To create the control group, patients without any diagnosis of MDD and without any prescription claim for antidepressant, antipsychotic, or antimanic drugs were selected using similar criteria.
To compare characteristics and costs between the two groups, patients with MDD were matched one-to-one with control patients using a combination of direct-characteristic matching (e.g., age, sex, region, insurance type, employment status, relationship to primary beneficiary, and Charlson Comorbidity Index [14,15,16]) and propensity score analysis. Each patient’s propensity score was calculated based on a logistic regression that controlled for general physical comorbidities that were observed to be statistically different at baseline between patients with MDD and control patients but that were not known to be directly related to MDD (e.g., hypertension). Patients with MDD were matched to control patients using a caliper of 0.25 within the standard deviation. Detailed comparisons of characteristics of patients with MDD and control patients before and after patient matching are presented in ESM 2 and 3.
Direct Costs Estimation
Consistent with the prior study, average costs were calculated for both patients with MDD and control patients for three direct cost categories: (1) MDD costs (i.e., medical costs incurred on the same day and in the same location as a medical claim with a diagnosis for MDD, as well as pharmaceutical costs for antidepressant, antipsychotic, and antimanic drugs); (2) other depression costs (i.e., including medical costs incurred on the same day and in the same location as a medical claim with a diagnosis for another type of depression (e.g., bipolar depression) but not MDD specifically, as well as pharmaceutical costs for antianxiety and anticonvulsant drugs); and (3) nondepression costs (i.e., all medical and pharmaceutical costs not captured in either of the first two categories). Incremental costs of patients with MDD were calculated by subtracting average costs of matched controls from those of patients with MDD (ESM 4 presents the detailed calculations for employed and treated adults with MDD in 2010 and 2015). The first cost category was the basis for estimating the direct costs of MDD, whereas all three categories combined were used to estimate the direct costs of individuals with MDD.
Employment and MDD treatment status were accounted for in the estimation of direct costs per patients with MDD: (1) for employed and treated patients, costs were estimated using the claims data; (2) for employed and not treated patients, MDD costs were set equal to 0 and non-MDD costs (comprising other depression costs and nondepression costs) were set equal to those incurred by employed and treated patients; and (3) for nonemployed patients (either treated or not treated), costs were assumed to be 1.7 times those found in the employed population, an assumption based on the ratio of healthcare costs incurred by patients with MDD with Medicaid coverage compared with patients with MDD who were privately insured . (ESM 5 presents detailed calculations of ratios used to infer otherwise missing cost categories.) Because the claims data did not contain cost information for patients aged ≥ 65 years, costs for these patients were assumed to be equal to those observed in patients aged 50–64 years.
Societal direct costs were extrapolated by multiplying NSDUH estimates of the number of people with MDD by the direct cost estimates per patient for each of the three cost categories noted above, stratified by age and sex. Because the NSDUH data did not contain the exact age for all respondents, the age stratifications in our study relied upon age categories available in the NSDUH (i.e., ages 18–25, 26–34, 35–49, and ≥ 50 years). These categories are consistent with those used in our prior study, allowing for direct age-related prevalence comparisons.
Suicide-Related Costs Estimation
Suicide-related costs for 2010 and 2018 were estimated using the human capital method, stratified by age and sex. The total number of suicides for each year was obtained from the Centers for Disease Control . Based on prior literature, our study attributed 50% of suicides to MDD in our cost estimates [19,20,21]. The present value of lifetime earnings was estimated using mortality rates and life expectancies from the National Vital Statistics Report as well as earnings data from the National Bureau of Economic Research [22,23,24]. A 3% discount rate was applied to the mortality cost estimates to express future earnings in present value terms . (ESM 6 presents the number of suicides and estimated loss of lifetime earnings due to MDD in 2010 and 2018.)
Workplace Costs Estimation
Workplace costs incurred by individuals with MDD included attention to both absenteeism due to missed days of work and presenteeism due to reduced productivity while at work. These costs were estimated for the same three categories as described for direct costs (i.e., MDD costs, other depression costs, and nondepression costs). Three different categories of absenteeism costs were considered: (1) injury/illness, (2) discretionary time off, and (3) disability. For absenteeism due to illness or injury, time away from work was imputed for the employed and treated subgroup based on the claims data: outpatient visits on workdays were counted as a half-day of missed work, and inpatient or emergency department visits on workdays were counted as full days missed. Discretionary time off was estimated based on NSDUH data in which respondents reported the number of workdays missed “because the respondent didn’t want to be there”. Costs of absenteeism associated with these first two categories were estimated by multiplying the cumulative number of days absent from work, from both categories, by that employee’s daily wage, which was included in the claims data. Costs of absenteeism associated with the third category, disability, were assessed directly from the claims data for the employed and treated subgroup. Absenteeism costs for individuals who were employed but not treated were assumed to be 44% of the costs incurred by those in the employed and treated group based on the ratio of workdays missed reported by each group in the NSDUH. Presenteeism costs were assumed to be 6.1 times the costs of absenteeism due to illness or injury based on previous literature estimates . (ESM 5 presents detailed calculations of ratios used to infer otherwise missing cost categories.) Societal workplace costs were extrapolated using the approach described for estimating total direct costs.
For the analyses of NSDUH data, p values for the changes across years in prevalence, employment, and treatment status were calculated using a two-sample z test for independent proportions. In the analyses of the OptumHealth claims data, statistical tests were performed using chi-squared tests for categorical variables and t tests for continuous variables.
Because our study relied on a combination of original estimates and estimates based on previously published literature, we performed sensitivity analyses for all estimates that incorporated literature-based estimates. Key parameters derived from the literature were increased and decreased by 10% to determine the extent to which the outcomes would change in either direction.
Analyses of the NSDUH and OptumHealth claims data, including propensity score matching, were performed using SAS Enterprise Guide version 7.15 (SAS Institute Inc., Cary, NC, USA). Costs were adjusted to $US, year 2020 values, using the medical care index of the Consumer Price Index.