Participants included 540 self-identified black or African American and Latina or Hispanic postpartum mothers who delivered between April 2009 and March 2010 at a large tertiary inner-city hospital located in East Harlem in New York City. The Program for the Protection of Human Subjects (the Institutional Review Board) at Mount Sinai School of Medicine approved this study. The patient sample was identified through an electronic documentation system for the labor and delivery unit. Eligible participants were black or African American, or Latina or Hispanic, aged 18 years or older, had infants with birth weights of 2500 grams or higher, and 5-minute Apgar scores of 7 or greater. Maternal race, maternal ethnicity, maternal age, infant Apgar scores, and infant birth weight are contained in the electronic system.
To insure eligibility, two bilingual clinical research coordinators asked additional questions of mothers who met initial requirements from the electronic record. Final eligible mothers self-identified as black or African American, or Latina or Hispanic, spoke English or Spanish, and had a working telephone. The clinical research coordinators approached mothers between their delivery day and postpartum day #2 (recruitment was done Mondays through Fridays). Patients were assigned a number between 1 and 20 in order of delivery date and time and then approached in the order set by that day’s list of random numbers generated by the project statistician. Our study team recruited between 12 and 14 patients per week.
After obtaining informed written consent, enrolled participants completed a 20 minute baseline in-hospital survey. Patients were randomized to conditions in four stages. The statistician carried out randomization using a computerized procedure that used a random number generator to assign participants to intervention or enhanced usual care. At the end of each stage, the randomization allocation could be potentially adjusted based on important covariates (e.g. race/ethnicity, EPDS score). No adjustment to randomization was initiated at any time. The research clinical coordinators were blinded to study arm assignment.
Patients randomized to the intervention arm were given a 2-step behavioral educational intervention. The in-hospital component of the intervention involved a 15-minute, in hospital review of a patient education pamphlet and partner summary sheet by the mother with a masters-trained bilingual social worker. The pamphlet represented each potential trigger of depressive symptoms as a “normal” aspect of the postpartum experience, and provided specific suggestions for management. For example, the prevalence of moderate or heavy vaginal bleeding immediately postpartum was depicted by 8 of 10 female silhouettes colored red; only 1 of 10 was red 3 months post delivery. Simple “to do” statements (Rest; Use pads) were listed between the two rows of figures. Postpartum and 3 month rates and intermediate “to do” lists also were provided for c-section site pain, episiotomy site pain, urinary incontinence, breast pain, back pain, headaches, hair loss, hemorrhoids, feeling sad and blue/depressive symptoms, and infant colic. A separate page was dedicated to social support and “helpful organizations” were listed. The partner summary sheet spelled out the typical pattern of experience for mothers postpartum, i.e., it was designed to “normalize” the feelings and behaviors experienced and enacted by most mothers postpartum and stressed the importance of social support for the patient. The social worker reviewed the patient education pamphlet and partner summary sheet with the patient during her postpartum hospital stay and answered questions.
The second and final component of the intervention was a two-week post delivery call in which the social worker assessed patients’ symptoms, skills in symptom management, and other needs. The “to do” lists to help alleviate symptoms were reviewed when needed and patient and social worker created action plans to address current needs including accessing community resources. Fidelity of the intervention was maintained by repeated training and review of scripts for both the in person in-hospital and telephone components of the intervention. Approximately 5% of both in-hospital sessions and two-week telephone needs assessment were observed by a physician or project manager on the team.
The content of the intervention was based on responses from our prior studies, focus groups with postpartum mothers, obstetricians, psychiatrists, social workers, and our community advisory board. The intervention was pilot tested with 50 diverse postpartum mothers and revised based on feedback from pilot study participants, the results from two focus groups conducted with black and Latina postpartum mothers, and feedback from community advisory board members (community representatives from Harlem). Content, pictures, wording, and length were revised according to input from these sources. Materials were translated to Spanish and back translated for accuracy and consistency of meaning.
Enhanced usual care patients received routine postpartum hospital education, (i.e. discharge materials, television educational programs on infant care, breastfeeding, and peripartum care). To insure equivalent contact, patients assigned to enhanced usual care control received a two-week post delivery call to inform them of future surveys and a list of health-related and community resources was mailed to them.
Interviewers were blinded to study arm assignment. All study participants were interviewed by phone at 3-weeks (mean days = 27.6, SD 5.8; median days = 27, range of 20 to 52), 3-months (mean days = 83.5, SD 7.8; median days = 81, range of 76 to 124) and at 6-months (mean days = 175.6, SD 8.9; median days = 174, range of 165 to 218) to assess depressive symptoms and contributing and buffering factors. Postnatal depressive symptoms were assessed using the 10-item Edinburgh Postnatal Depression Scale (EPDS). The EPDS is a common measure used in research to assess symptoms of postpartum depression in both treatment and prevention trials, and has been validated in many postpartum populations and different languages.16–29
The recommended cut-off score of 10 or greater has sensitivities of 0.59–0.81 and specificities ranged from 0.77–0.88 for major and minor depression.2
The Patient Health Questionnaire-9 was also administered and a comparison was made between the EPDS and PHQ-9 scores over time in a secondary analysis. Survey items also included questions on sociodemographics, clinical characteristics such as antepartum complications, comorbid conditions (e.g. diabetes, hypertension, asthma, thyroid disease, heart disease), past depression history, anxiety, social support, and healthcare factors. Medical charts were reviewed for parity, delivery type, insurance, past medical history, maternal complications, delivery complications, and infant outcomes.
After patients were consented and completed the baseline survey, women reporting high levels of depressive symptoms (EPDS of 13 or greater, PHQ-9 of 20 or greater, or suicidal ideation) were referred for inpatient psychiatric assessment and possible treatment. To address risk over time, all mothers, including those referred for inpatient assessment at baseline, were retained in the study, and mothers who reported high levels of depressive or suicidal symptoms at any time point were referred for psychiatric assessment/treatment. Analyses excluding mothers referred for baseline psychiatric intervention during hospitalization were planned a-priori.
Sample size was powered to detect a clinically meaningful difference in reduction of symptoms of postpartum depression 3 weeks post randomization for women in the intervention arm in comparison with women in enhanced usual care. Assuming that 46% of black and Latina women would report depressive symptoms (EPDS of 10 or greater) in enhanced usual care (based on published literature and our studies of postpartum depressive symptoms among inner-city black and Latina mothers3,4
), our recruitment target of 460 patients (230 per arm) met 90% power based on a two-sided 0.05 level chi-squared test to detect a clinically meaningful 15% reduction in depressive symptoms (from 46% to 31%, a relative risk reduction of about 33%). To allow for decreased power due to patient loss at follow-up we planned to enroll 540 participants.
Data were collected in-person at baseline and by telephone during follow-up interviews by bilingual clinical research coordinators blinded to intervention status. All data were entered directly into a laptop computerized instrument linked to a Microsoft Access database. Participants in the intervention arm (N=270) and enhanced usual care (N=270) were compared at baseline on demographic and clinical characteristics using t-tests and their non-parametric equivalents, chi-square tests, as appropriate. Group differences were summarized by 95% confidence interval estimates. Overall study attrition rate was low and equivalent across treatment groups. The amount of missing data for our primary outcome measure, EPDS, was low at 3-weeks (13%; 71/ 540), 3-months (12%; 62/540) and six months (22%; 117/540), and similar between groups at each time point.
The primary analysis examined the efficacy of the intervention in reducing the likelihood of positive postpartum depression screens. Planned subgroup analyses examined whether the intervention reduced the likelihood of a positive depression screen among mothers who scored below 13 on the EPDS and were therefore not referred for baseline psychiatric intervention. Prior to conducting these analyses, missing data were imputed, i.e. estimated using a multiple imputation approach.30
Data were assumed to be missing at random as baseline characteristics for those missing outcomes at 3-weeks were very similar to those for whom outcomes were available at 3-weeks, with the only significant differences (p<.05) being that those with missing outcomes were more likely to be born in the US, to speak English, and have past history of depression. The regression models used baseline EPDS scores and past depression history. Five complete data sets were created using multiple imputations and analyzed and combined using the MIANALYZE procedure in SAS. The final parameter estimate and the associated standard errors account for both with-in and between-imputation variance. As imputation yielded similar results to the observed trial results both for the total sample and the subgroup analysis including the mothers not referred for intervention (92% of the sample), we present observed results, the more conservative estimate of effect, for both. The logistic regression analyses tested the effect of the intervention on positive depression screens at 3 weeks, 3 months, and 6 months postpartum. We conducted adjusted analyses for the subgroup of mothers not referred for treatment at baseline. We included independent variables that were associated with a positive depression screen at baseline. To assess change over time across groups for the primary outcome, mixed-effects models using maximum likelihood procedures were employed. Significance level of 0.05 was used for the primary outcome of postpartum depression and the final model was adjusted for baseline depression screen.