We conducted a retrospective, observational cohort study using administrative claims from the Healthcare Integrated Research Database. This database includes medical and pharmacy claims from 14 states in the United States.¹⁷ Member enrollment data, inpatient and outpatient medical care, and outpatient prescription drug use are tracked longitudinally for each member. Researchers accessed data in the format of a limited dataset, for which data use agreements were in place with the covered entities in adherence with the Health Insurance Portability and Accountability Act Privacy Rule, and therefore, institutional review board approval was not required. The full study protocol is available through the Real-World Evidence Registry: https://osf.io/yq6b3.

Patients with commercial health insurance coverage who had at least 1 inpatient or emergency department visit or at least 2 outpatient visits (at least 30 days apart but not more than 183 days apart) with a diagnosis of MDD during the patient selection period (January 1, 2017, to September 30, 2019) were included in the study. The earliest MDD diagnosis date was set as the index date. Patients included could not have any claims for MDD or MDD treatment in their history prior to their index date. Furthermore, patients could not have a diagnosis of schizophrenia, bipolar disorder, postpartum depression, substance use disorder, an intellectual disability, or a neurocognitive disorder during the study period (January 1, 2006, to September 30, 2021), ie, both before and after their index date during all available data. Patients were required to be aged between 18 and 62 years on the index date (ie, patient could not be aged ≥ 65 years during follow-up to exclude transitions into Medicare insurance) and have continuous medical and pharmacy benefit enrollment during the 1-year baseline and 2-year post-index follow-up periods. Patients whose pharmacy benefit was carved out, resulting in missing cost data, were excluded. See Supplementary Tables 1-3 (available in online article) for code lists and Supplementary Figure 1 for the study design diagram.

Patient demographics were assessed on the index date, whereas the Quan-Charlson Comorbidity Index score, other comorbidities, and medication use (glucocorticoids, narcotic analgesics, and non-narcotic analgesics) were assessed during the baseline period.¹⁸ Patients were linked to data from the 2017 American Community Survey 5-year estimates at the Census block group level based on their location of residence on the index date.¹⁹ Socioeconomic status (SES) index score was calculated for each patient based on a composite of 7 variables from the American Community Survey including unemployment rate, poverty rate, median household income, median home value, rate of no high school diploma, rate of college degree, and crowding.²⁰ A score of 4 indicates the patient is in the top 25% of SES, and a score of 1 indicates the patient is in the bottom 25% of SES using all Census block groups in the United States.

MDD severity was assessed using International Classification of Diseases, Tenth Revision, Clinical Modification coding for severity on the medical claims and categorized as mild, moderate, severe, or in remission (Supplementary Table 1). Initial MDD severity was assessed between the index date and 30 days post-index. For those with different severity assessments during that time, patients were assigned the higher severity (eg, a patient with both a moderate and severe code would be labeled severe). Post-index severity changes were assessed between days 31 and 730 post-index among those with at least 1 claim with severity specified during that time and at least 1 claim specified at index. Patients were categorized as having no severity change, worsened severity, or improved severity by comparing the initial and last available severity assessment.

Treatment was assessed during the 2-year post-index period. Nonpharmacologic treatment in the post-index period was identified using Current Procedural Terminology codes (Supplementary Table 3). Antidepressant treatments were identified from any medical or pharmacy claims with a Healthcare Common Procedure Coding System code or Generic Product Identifier code for a treatment of interest. Antidepressants included SSRIs, SNRIs, tricyclic antidepressants (TCAs), monoamine oxidase inhibitors, serotonin modulators and stimulators, bupropion, mirtazapine, esketamine, and ketamine. Patients ended their current line of treatment when they discontinued their existing antidepressant (defined as a gap of 60 days or more between the depletion date [ie, date of last fill plus days supply on hand] and their subsequent fill), switched antidepressants, or added (or removed) an antidepressant to (or from) their initial antidepressant regimen. Restarting an antidepressant (either the same or a different antidepressant) after discontinuation, switching antidepressants, or adding (or removing) an antidepressant to (or from) the regimen prompted the start of a new line of treatment. Treatment patterns were assessed at the class-level for the 5 classes of drugs (SSRIs, SNRIs, TCAs, monoamine oxidase inhibitors, and serotonin modulators and stimulators) and the drug-level for the 4 specified drugs (bupropion, mirtazapine, esketamine, and ketamine). Changes in medications within the same class were not considered.

Patients were further categorized into 1 of 4 treatment modalities based on antidepressant and nonpharmacologic treatment utilization at any time in the post-index period: (1) those who received antidepressant treatment only (Rx-only); (2) those who received nonpharmacologic treatment only (non–Rx-only); (3) those who received both antidepressant and nonpharmacologic treatment (combination); or (4) those who received no treatment (none) during the post-index period.

All-cause HCRU and health care costs were assessed during the post-index period by treatment modality. All-cause costs were adjusted to 2021 US dollars based on the most recent medical price index information provided by the Bureau of Labor Statistics and calculated as the combined health-plan-paid, member-paid, and coordination-of-benefits (third-party payer) costs.²¹

Demographics, clinical characteristics, treatment patterns, HCRU, and health care costs were presented using descriptive statistics. A Sankey diagram showing the antidepressant treatment lines was created for up to 4 lines of treatment. HCRU was presented as number and percent of patients with any use during the post-index period, whereas health care costs were presented as monthly mean costs calculated by taking the sum of costs over the post-index period and dividing by the total number of months during the post-index period (ie, 2 years). Multinomial logistic regression was used to identify associations between baseline patient/provider characteristics (exposure) and treatment modality (outcome; reference category = no treatment). Multivariable logistic regression was used to identify associations between baseline patient/provider characteristics and treatment modality (exposures) and MDD severity (outcome; reference category = no change or worsened severity). Results from these regressions are presented as forest plots. Multicollinearity was assessed using variance inflation factors with inflation factors greater than 10 indicating potentially harmful levels of collinearity.²² Bidirectional stepwise regression models were used to select variables to be included in the models. For entry and stay in the model, explanatory variables had to have P values of 0.25 or less. Age and sex were retained in all models regardless of significance. Logistic regression model fit was assessed with the Hosmer-Lemeshow test and C-statistic. A threshold of P < 0.05 was used to define statistical significance, with no adjustments made for multiple testing. Instant Health Data software and SAS Enterprise Guide 8.3 were used to conduct all analyses.

In total, 12,657 patients met cohort selection criteria (Supplementary Figure 2) with a mean age of 35.6 years and 60% female. Most were diagnosed with MDD by a primary care physician (PCP) (36%) or nonphysician clinician (34%) such as a psychologist, nurse practitioner, or licensed clinical social worker (LCSW). Furthermore, most (86%) were diagnosed in an outpatient setting. Patients also had low Quan-Charlson Comorbidity Index scores (mean score: 0.18) and a low prevalence of other comorbid conditions (Table 1).

Among all patients with MDD, 4,282 (34%) received Rx-only treatment, 3,121 (25%) received non–Rx-only treatment, 3,554 (28%) received combination treatment, and 1,700 (13%) received no treatment over 2 years of follow-up. Of the 7,836 patients who had any antidepressant medication treatment, 54% had 1 treatment line, whereas 46% had 2 or more treatment lines during the post-index period. The most common first-line treatment was SSRI monotherapy, and this remained the most common treatment regimen throughout the first 4 lines of treatment (Supplementary Figure 3). The most common nonpharmacologic treatment was individual, family, or group psychotherapy, used by 90% (6,033/6,675) of patients with any nonpharmacologic treatment (data not shown).

In adjusted multinomial logistic regression (Figure 1; Supplementary Table 5), significant sociodemographic associations with treatment modality were identified. Older patients were less likely to receive Rx-only, non–Rx-only, or combination treatments than no treatment, and compared with male patients, female patients were more likely to receive nonpharmacologic treatment than no treatment. Compared with those living in the South, those living in the Northeast had higher odds and those in the Midwest had lower odds of receiving nonpharmacologic treatment over no treatment. Those living in the Northeast also had higher odds of receiving combination treatment over no treatment than those living in the South. Those living in the Northeast and West had lower odds of receiving Rx-only treatment vs no treatment than those living in the South. Treatment modality was also associated with rurality; those living in rural locations were less likely to receive non–Rx-only vs no treatment than those living in urban locations and more likely to receive Rx-only vs no treatment, but there was no statistical difference between receiving combination vs no treatment. Those with higher SES (quartile of 4 or 3 vs 1) had higher odds of non–Rx-only treatment vs no treatment and combination vs no treatment.

Additionally, the association between diagnosing provider specialty and treatment modality varied (Figure 1; Supplementary Table 5). Those diagnosed by a psychiatrist, emergency medicine physician, or nonphysician clinician had higher odds of receiving non–Rx-only treatment vs no treatment compared with those diagnosed by a PCP. Similarly, those diagnosed by a psychiatrist or nonphysician clinician also had higher odds of receiving combination treatment vs no treatment than those diagnosed by a PCP. Conversely, those diagnosed by an emergency medicine physician or nonphysician clinician had lower odds of receiving Rx-only treatment vs no treatment than those diagnosed by a PCP. Finally, location of diagnosis was associated with treatment modality; those diagnosed in an outpatient setting were more likely to receive non–Rx-only, Rx-only, or combination treatments vs no treatment than those diagnosed in an emergency department or inpatient setting.

Among the total patient population, 6,444 (51%) had MDD severity specified on their index diagnosis claim, with 26% coded as mild, 54% as moderate, and 20% as severe MDD. All patients whose MDD severity was specified on the index claim also had a specified diagnosis in the post-index period. Most (76%) patients did not change MDD severity between the index and last specified severity in the post-index, but 18% improved severity while 6% worsened severity (Table 2).

In multivariable logistic regression examining associations with MDD severity improvement, neighborhood-level SES, diagnosing provider specialty, index severity, and post-index treatment modality were significantly associated with severity improvement (Figure 2; Supplementary Table 6). Those who lived in neighborhoods of high SES were more likely to improve severity than those living in neighborhoods of low SES. Those diagnosed by a psychiatrist, psychologist, or LCSW all had lower odds of improving severity than those diagnosed by a PCP. Index severity was also associated with severity improvement, with those who had moderate MDD at index having 3.33 times the odds (95% CI = 2.66-4.18) and those who had severe MDD at index having 8.37 times the odds (95% CI = 6.51-10.77) of severity improvement in the post-index period than those who had mild MDD at index. Finally, post-index treatment modality was significantly associated with severity improvement. Those who received Rx-only were 2.03 times more likely (95% CI = 1.37-3.02) and those who had combination treatment were 3.26 times more likely to improve severity (95% CI = 2.20-4.82) compared with those who had no treatment.

In the post-index period, those who received no treatment had the numerically highest HCRU and mean health care cost. Specifically, 29% of those who had no treatment had at least 1 inpatient hospitalization in the post-index period whereas only 11%, 10%, and 16% of those who had Rx-only, non–Rx-only, and combination treatment modalities, respectively, had at least 1 inpatient hospitalization. Prevalence of emergency department utilization in the post-index period was similarly higher among those who received no treatment; 46% of those who had no treatment had an emergency department visit, whereas 28%, 22%, and 30% of those with Rx-only, non–Rx-only, and combination treatments, respectively, had an emergency department visit. Finally, the mean monthly all-cause costs were $1,292 in the no treatment group in the post-index period, whereas costs were $792, $633, and $786 for those with Rx-only, non–Rx-only, and combination treatments, respectively (Figure 3; Supplementary Table 4).

These findings should be interpreted with consideration of several limitations. First, this study was an exploratory analysis, and the reported associations should not be interpreted as causal effects. In particular, the contemporaneous assessment of treatment modality and MDD severity precludes causal inference. Second, certain characteristics, such as provider preferences, health plan benefit design including the structure of behavioral health coverage, and patient characteristics (eg, individual health-related social needs, family history), were not available, and their effects were not included in our analysis. Third, administrative claims are collected for the purposes of payment and may not reflect true diagnoses and treatment, as coding issues may occur or medications may not be taken as prescribed. Fourth, certain medications (eg, TCAs and bupropion) may have multiple indications, and we were unable to discern for which indication these were prescribed. Fifth, the treatment modality analysis did not consider intensity of treatment, nor did it examine whether patients who received both antidepressant and nonpharmacologic treatment did so concurrently. Sixth, some patients’ follow-up time overlapped with the beginning of the COVID-19 pandemic, which may have impacted their access to and utilization of treatments and other health care services. Finally, the study results may not be generalizable to the national level because patients who have commercial health insurance may have different characteristics from those with different types of or no insurance.