Postpartum Depression in Reproductive-Age Women With and Without Rheumatic Disease: A Population-Based Matched Cohort Study

METHODS

The UMass Chan Medical School Institutional Review Board considered our study not to be human subject research, as it used a secondary, deidentified database (H00018231).

Data source. We used data from the IBM MarketScan Commercial Claims and Encounters Database (January 1, 2013, to December 31, 2018), one of the largest commercial medical claims databases in the USA. It is based on a convenience sample of a large US population with employer-sponsored commercial insurance plans, including insured employers, their spouses, and their children. This claims-only database captures information on patient demographics, enrollment in insurance plans, and medical and pharmacy claims submitted from outpatient office visits, inpatient hospital stays, and specialty care.¹⁷

Identification of axSpA/PsA/RA. A previously validated algorithm (using International Classification of Diseases, 9th and 10th revisions [ICD-9/10] codes)^18-20 identified patients with axSpA, PsA, or RA. Patients were eligible if they had satisfied 1 of the following criteria between January 1, 2013, and December 31, 2018: (1) ≥ 2 outpatient diagnoses of axSpA, PsA, or RA on different dates ≥ 7 days apart; (2) ≥ 1 outpatient diagnosis of the disease confirmed by a rheumatologist; (3) ≥ 1 outpatient diagnosis of axSpA, PsA, or RA and ≥ 1 dose of a disease-modifying antirheumatic drug (DMARD; biologic [bDMARD] or nonbiologic); or (4) ≥ 1 diagnosis of axSpA, PsA, or RA that occurred during an inpatient hospitalization. The diagnostic codes to identify cases of axSpA, RA, and PsA are provided in Supplementary Table S1 (available with the online version of this article).

Baseline depression status and PPD. A validated algorithm was used to identify patients with clinical depression prior to delivery.²¹ Patients were considered to have depression if they had ≥ 1 inpatient or ≥ 2 outpatient claims for depression in 1 year. A binary variable (yes or no) was created to record the presence or absence of clinical depression prior to delivery. The same algorithm was used to identify women with PPD during the follow-up period after delivery. Within the first 12 months following delivery, patients were required to have ≥ 1 inpatient claims for depression or ≥ 2 outpatient claims to be identified as a case of PPD.²¹ The ICD-9/10 codes used to identify cases of depression are provided in Supplementary Table S2 (available with the online version of this article).

Study design. The study design is displayed in Figure 1. The date of delivery was defined as the index date. The period from 6 months before the date of the last menstrual period (LMP) to the delivery date was used as the baseline period. The LMP date was estimated using a previously validated algorithm^22,23 by subtracting the gestational age (obtained using Z3A codes [weeks of gestation; ICD-10] and gestational age codes [ICD-9]) from the claims date (Supplementary Table S3, available with the online version of this article). When gestational age codes were not available, LMP date was calculated as the delivery date (index date) minus the standard gestational age in weeks (live birth: 39; mixed birth [(≥ 1 livebirth]: 35).^24-28 The period from the index date to PPD diagnosis, patient-specific end of continuous enrollment, or 12 months postdelivery—whichever came first—was used as the follow-up period. The earliest date that the patient had a claim satisfying the above criteria for PPD was used as the date of diagnosis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Study design. axSpA: axial spondyloarthritis; b/w: between; LMP: last menstrual period; PsA: psoriatic arthritis; PPD: postpartum depression; RA: rheumatoid arthritis.

Source population and cohort identification. Figure 2 presents the inclusion and exclusion criteria used and the eligible sample for analysis. The initial source population included 2,175,353 women with a delivery date between January 1, 2013, and December 31, 2018. We then restricted the source population to women whose pregnancy ended in either a live birth or mixed birth (≥ 1 live birth; n = 1,678,886; Supplementary Table S4, available with the online version of this article).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Flowchart of study sample. axSpA: axial spondyloarthritis; DMARD: disease-modifying antirheumatic drug; ICD-9/10: International Classification of Diseases, 9th and 10th revisions; LMP: last menstrual period; PsA: psoriatic arthritis; RA: rheumatoid arthritis; RD: rheumatic disease.

For the group henceforth referred to as the axSpA/PsA/RA cohort, we first identified women in the source population who met the operational definitions for axSpA, PsA, or RA (n = 15,809). Among these, we then restricted women who had a recorded ICD-9/10 code for only 1 of the 3 rheumatic conditions (ie, only axSpA, only PsA, or only RA; n = 15,752). For example, those who had 1 axSpA diagnostic code with 1 RA diagnostic code were excluded. The remaining sample was then restricted to women with only axSpA, PsA, or RA and no other RD (n = 14,198; Supplementary Table S5, available with the online version of this article) and whose date of axSpA, PsA, or RA diagnosis preceded the LMP date by ≥ 90 days (n = 3884). Further, patients were required to be continuously enrolled within the MarketScan database for medical and pharmacy coverage (1) for ≥ 6 months prior to the LMP date and (2) throughout pregnancy (ie, between the date of LMP and the index date; n = 2695). Because the ICD-9/10 codes for pregnancy are only applicable to women between the ages of 12 and 55 years, only patients who were aged ≤ 55 years on the index date were included in our analysis (n = 2682). For women with multiple eligible births, the first event was selected. Finally, we excluded 15 patients because we were unable to identify appropriate comparators for matching from the non-RD comparison cohort. The remaining 2667 women comprised the final analytic sample.

For the non-RD comparison cohort, we first identified patients in the source population with a delivery date (index date) between January 1, 2013, and December 31, 2018, and who did not have an ICD-9/10 diagnosis code for any RD (as listed in Supplementary Table S1 and Supplementary Table S5, available with the online version of this article; n = 1,634,882). The non-RD comparison cohort was sampled using the same inclusion and exclusion criteria as the axSpA/PsA/RA cohort. Four patients from the non-RD comparison cohort were matched to each case in the axSpA/PsA/RA cohort.²⁹ Covariates for matching were selected a priori, including: (1) history of depression (yes or no), (2) duration of depression before the index date, and (3) maternal age at delivery. The earliest date at which the patient met the criteria of clinical depression was used as the date for the diagnosis of depression. When matching the duration of depression prior to delivery, we first calculated the time period in days between the date of depression and the index date. This calculated duration ± 30 days was then used to match patients from the non-RD comparison cohort to those in the axSpA/PsA/RA cohort. For example, if a patient from the axSpA/PsA/RA cohort was diagnosed with depression 230 days prior to the index date, they would be matched with 4 patients from the comparison cohort with a diagnosis of depression between 200 to 260 days prior to the index date. Matching without replacement was used to match the cases to the comparator patients.

Prescribed medications at baseline and follow-up period. Prescribed antidepressants were assessed (1) during the 6 months prior to the LMP date, (2) between the LMP date and the index date, and (3) during the 12 months postdelivery. Antidepressants included selective serotonin reuptake inhibitors (SSRIs), tricyclic antidepressants (TCAs), monoamine oxidase inhibitors (MAOs), serotonin and norepinephrine reuptake inhibitors (SNRIs), and others (Supplementary Table S6, available with the online version of this article). Initially, 5 binary variables were created to determine prescriptions filled for each medication class during each of the 3 time periods. We then created a composite, 4-level variable consisting of (1) no antidepressant use in the baseline period, (2) antidepressant use during the 6 months prior to pregnancy only, (3) antidepressant use during pregnancy only, and (4) antidepressant use throughout the baseline period (ie, in the 6 months before pregnancy and during pregnancy) for analytical purpose. Binary variables (yes or no) for prescriptions filled for bDMARDs, nonbiologic DMARDs, nonsteroidal antiinflammatory drugs (NSAIDs), corticosteroids, and opioids were also determined before, during, and after pregnancy (Supplementary Table S6, available with the online version of this article).

Other covariates. Several potential factors associated with PPD were evaluated using data from the baseline period including maternal age (15-24, 25-34, or ≥ 35 years), geographic region (Northeast, North Central, South, West), employment status of the primary beneficiary (active full-time, active part-time or retiree, other or unknown), and health insurance plan type (health maintenance organization, preferred provider organization [PPO], consumer-directed or high-deductible healthcare plans, or other).³⁰ A binary variable indicating the presence of comorbidities was derived based on the Charlson comorbidity index score^31-33 using medical comorbidities identified with ICD-9/10 codes from inpatient and outpatient claims during the baseline period. The proportion of women with prenatal care visits during pregnancy was documented and a 4-level variable (0, 1-5, 6-9, or ≥ 10) was created to determine the frequency of prenatal care visits (Supplementary Table S7, available with the online version of this article). Further, the mode of delivery (vaginal delivery or cesarean section) and maternal complications during pregnancy, including gestational diabetes mellitus, pregnancy-induced hypertension, preeclampsia, and eclampsia, were also explored (Supplementary Table S8, available with the online version of this article). Initially, 4 binary variables were created to indicate the presence of each pregnancy complication. We then created a composite variable to evaluate the presence of pregnancy complications (yes or no).

Statistical analyses. Baseline demographic and clinical characteristics were described and compared for the 2 study cohorts. Survival time was reported in days and was calculated for each patient from the date of delivery or index date to the date of PPD diagnosis, patient-specific end of continuous enrollment, or 12-months postdelivery—whichever came first.³⁴ Kaplan-Meier estimates were used to graph the probability of developing PPD between the axSpA/PsA/RA cohort and non-RD comparison group. Univariable and multivariable Cox frailty proportional hazards models, including random intercepts of matched subjects, were used to estimate the crude and adjusted hazard ratio (aHR) and 95% CI of incident PPD among women with axSpA, PsA, or RA compared with those without any RD. For each model, we ruled out deviations from the proportional hazards assumption using Schoenfeld residuals goodness of fit test followed by log-log survival curves.³⁵