Setting
Sehore District, which is located in the central Indian state of Madhya Pradesh, has a population 1.3 million, of whom 81% live in a rural area and 68% are literate [19]. In Madhya Pradesh, maternal health services are readily accessible: 90% of mothers were immunized for tetanus prior to their last birth and 81% of births are in health facilities [20]. In collaboration with the Programme for Improving Mental Health Care (PRIME) research consortium [21], the state government started integrating mental health care services for depression into the primary health care sector, which includes perinatal health services, in Sehore District in 2013 [22]. Starting in 2016, nurses in district hospitals across the state are receiving training to detect perinatal depression and provide basic psychological interventions.
Community Study
A detailed description of the Community Study is available [23]. Briefly, the Community Study was a multi-round, population-based survey which aimed to estimate the change in treatment coverage among adults (age ≥ 18 years) affected by depression or by alcohol use disorder after implementation of PRIME. In the baseline (pre-implementation) round, research assistants selected 89 villages randomly in Sehore District, selected one voting polling station randomly from each village (range 1–5 stations), and downloaded the list of registered voters for that station from the government election commission website. At the time the study was designed, we assumed that the voter registration list was fairly comprehensive and suitable for use as a sampling frame for the population of potential health service users. After the baseline survey occurred, the implementation area was redrawn, reflecting a decision to focus programme activity on 188 villages in one sub-district. The follow up survey sample included all 188 villages, and all voter lists in each village. In both rounds, from each voting list, the research assistant selected adults randomly to recruit into the study. The baseline survey (n = 3220) was conducted between May 2013 and March 2014, the follow-up survey (n = 2914) was conducted between October and January 2017, and independent samples were drawn in each round.
Facility Study
The Facility Study aimed to estimate the change in detection of depression and of alcohol use disorders by primary care clinicians working in three public clinics in one sub-district of Sehore District. Research assistants recruited a systematic random sample of adults (age ≥ 18 years) who were exiting their clinical consultations. The baseline (pre-implementation) survey (n = 760) was conducted between August and October 2013, the follow up (post-implementation) survey (n = 817) was conducted between September and October 2016, and independent samples were drawn in each round.
Data collection and measures
In both studies, Hindi-speaking interviewers used a structured questionnaire application administered on an Android tablet device to ask participants about their sociodemographic characteristics (i.e. age, religion, caste, educational attainment, marital status, housing quality, land ownership, and parity, and age and sex of living children), and health-related status (i.e. pregnancy, depression, disability, and recent suicidal ideation). The response options for housing quality are widely understood in this setting to correspond to homes which are low quality, intermediate quality and high quality.
Interviewers screened participants for depression with the 9-item Patient’s Health Questionnaire (PHQ9) [24], which has been validated in India [25]. Each item relates to a DSM-IV criterion for major depressive disorder and is scored from 0 (“not at all”) to 3 (“nearly every day”) with a recall period of 2 weeks. Four score categories have been established: 0–4 normal, 5–9 mild depressive symptoms, 10–14 moderate depressive symptoms ≥ 15 moderately severe to severe depressive symptoms [26, 27]. We considered a total PHQ9 score of 10 or more to be a positive screen for depressive symptoms. A meta-analysis of PHQ9 validation studies found that this cut off score has 85% sensitivity and 89% specificity for detecting depression [28]. The Cronbach’s alpha for the PHQ9 for this sample was 0.67.
To assess disability status over the past 3 months, interviewers administered the 12-item World Health Organization Disability Assessment Schedule (WHODAS) 2.0 [29], and administered an adaptation of the Composite International Diagnostic Interview (CIDI) suicidality module [30] to assess suicidal ideation in the past 12 months.
In the Community Study, interviewers asked PHQ9-positive participants whether they had sought treatment (i.e. from the specialist, generalist or complementary health provider, and the nature of the treatment). In the Facility Study, interviewers asked PHQ9-positive participants about their clinical consultation (i.e. any diagnoses made, and the nature of any advice, treatment or referral provided).
As part of the recruitment process, interviewers gave adults oral and written information about the study and participating adults affirmed their consent with a signature or thumb print. Participants who affirmatively responded to questionnaire items about suicidality received a referral to a research psychiatrist. The study protocols were approved by Sangath (Goa, India), the University of Cape Town (South Africa) and World Health Organization (Geneva, Switzerland).
Statistical analysis
For this secondary analysis, we included women from the two studies who were in the perinatal period, defined as being pregnant or had a child of 12 months or younger [1].
First, we describe the sociodemographic characteristics, health-related status and study base of participants within the Community Study, within the Facility Study, and for the pooled sample, using percentages within each category. We created tertiles of age and education. For parity, we created three categories: antenatal, postnatal with only daughters, postnatal with any sons, which was informed by previous literature about links between gender preference and depression [16]. For disability, we created tertiles of WHODAS scores to represent those with lower (score 12–15), average (16–19) or higher (≥ 20) levels of disability.
Second, with data pooled from both studies, we evaluated whether the distribution of PHQ9 scores varied by sociodemographic characteristics, health-related status, or study base (i.e. community or facility). As the distribution of PHQ9 scores was skewed we could not use linear regression to test for associations. Instead, we report the median and interquartile range for each stratum and used the Kruskal–Wallis test to test for an association between the measure and the PHQ9 score. For the test of association between suicidal ideation and PHQ9 score, we excluded the final item on the PHQ9 from the total outcome score, as this item also pertains to suicidality. We repeated the associations analysis with a binary PHQ9 score, reported the proportion of participants in each category who screen positive (PHQ9 ≥ 10), and used Fisher’s exact to calculate P values for the differences in proportions. And then we repeated the second part of this analysis for each separate study sample.
Third, we report on the primary outcomes of the two parent studies as they pertain to this sub-sample of perinatal women. From the Community Study, we report the change, from baseline to follow-up, in the proportion of PHQ9-positive participants who reported seeking care from a health care provider for symptoms relating to depression. And from the Facility Study, we report the change from baseline to follow-up, in the proportion of PHQ9-positive participants who reported a diagnosis of depression in their clinical consultation. We tested for differences in these proportions using Fisher’s exact test.
We completed all analyses in Stata SE 14.2 (College Station, TX, USA) (Stata code in Additional file 1).