Study Design and Study Setting
This paper uses secondary data collected from 17,232 women from an Integrated microfinance and health literacy (IMFHL) program that was implemented in rural Uttar Pradesh, India between 2012 and 2017 (please see  for a detail description of this survey). The program aimed to provide low-income women with maternal and newborn health literacy delivered through a microfinance platform (38,48,52). Under the program, a quasi-experimental survey design was used to collect cross-sectional survey data in two rounds (round 1 in 2015; round 2 in 2017) to evaluate the program’s impact on knowledge and health behaviours of women during pregnancy, delivery and post-delivery in Uttar Pradesh(38,48,52).
The IMFHL Program
The IMFHL program was a community-driven and rapidly scalable program that integrated health promotion activities in a microfinance platform across low developmental districts of UP (44,52). Details of the IMFHL implementationand health intervention have been published elsewhere (44,52). The health promotion component sought to address community-related barriers mainly related to low health literacy and poverty and to encourage women to adopt preventive health behaviours known to reduce maternal and neonatal mortality (44,52). The IMFHL program was built on previous participatory community programs such as the Makwanpur trial in Nepal and the Shivgarh trial in Uttar Pradesh that showed a reduction in maternal and newborn mortality achieved through the adoption of essential maternal and newborn care practices in households (29,53). Under the IMFHL program, the maternal health literacy component targeted eligible women at different stages of their pregnancy and provided them with information to recognise pregnancy-related complication signs in order to reduce delays in seeking care from a health facility in the event of any pregnancy-related complication (38,44). A SHG member, trained as a health volunteer, facilitated health discussions involving pregnant and recently delivered women using key program strategies such invitation to SHG meetings, reminder letters to pregnant and new mothers with key health messages, house visits, and exposure to community health video shows developed by the program (44,52). Furthermore, the IMFHL program created three apparent groups of beneficiaries. The baseline tier was composed of pregnant women in SHG member households that received health messages directly through program strategies (IMFHL intervention group). Since villages are comprised of member and non-member households living in close proximity and communicating with each other, a process of community-based diffusion in knowledge sharing was expected from SHG plus Health (IMFHL intervention group) to tier I and to tier II households respectively, to supplement direct program intervention efforts as depicted in Figure 2.
Diffusion of knowledge was expected to occur in SHG program villages through a process of collective socialisation in which heath literate SHG members serve as role models and help other non-members internalise biomedical norms around pregnancy and childbirth (30-41). This research examined the above-stated assumption to determine if the SHG platform encourages the sharing of health information from members receiving health literacy to non-members
IMFHL Program Implementation: Selection of Intervention and Comparison Blocks
Under the IMFHL program, implementation districts were selected comprising high maternal and neonatal mortality burden with a higher percentage of scheduled caste (SC)/tribe (ST) and low literacy (52). SC/ST are members from communities designated by the Indian Government to have historically faced ‘extreme social, educational and economic backwardness arising out of the traditional practice of untouchability' and afforded legislative protection and entitlements (54).
Within districts, 120 implementation blocks were selected and separately,83 comparison blocks, were roughly matched to the intervention blocks as per the percentage of SC/ST (38,44), Lastly, ‘pure control’ blocks that received no SHG or health intervention were chosen to observe secular change in health indicators in rural Uttar Pradesh. Eligible women in households were identified as SHG members in villages using a community participatory approach and inclusion criteria (52). The most disadvantaged households were often represented by landless poor households with low literacy, lower social class (and caste) with multiple (social) deprivations (44,52). In these villages, one woman from an eligible household was allowed to join SHGs that were nurtured by the field staff. Other households who would not be facing similar credit constraints as poorer households from lower castes in the same villages, were not targeted by the SHG program for membership (44,52).
Survey Sampling Approach and Study Population
Sampling Strategy for Survey
With a total population close to 200 million, Uttar Pradesh state is administratively subdivided into 75 districts, 822 blocks and 98,000 Gram panchayats (GP) (52). The IMFHL program was implemented in 203 blocks while the survey data was collected from 70 blocks in 20 districts as a representative sample from the IMFHL program’s coverage area (44,52). In India, GPs are the smallest unit of administration within blocks where SHGs were established. Where data were collected, a GP may be classified as a larger main village with smaller peripheral villages attached to it and may have fewer houses attached as hamlets. The surveys followed a three-stage sampling approach for selecting blocks, GPs and finally, households, based on the state’s administrative hierarchy, and as depicted in Figure 2 below.
The IMFHL program collected data in both survey rounds from three types of blocks, depending on the IMFHL program exposure: i) intervention blocks where households received health intervention through SHG program, ii) comparison blocks where households received SHG program only, and iii) pure control blocks where households did not receive any program exposure reflecting the natural change in health indicators. While the IMFHL program used SHGs in both intervention and comparison blocks, only households in intervention blocks received additional health intervention (see Figure 3). Moreover, both intervention and comparison blocks had SHG membership of similar duration (average duration of 18 months). In the first stage, the intervention (SHG plus health) blocks were first arranged in ascending order of their associated percentage of Scheduled Caste (SC) and Scheduled Tribe (ST) population (SC/ST), a critical parameter for development (54). The required number of intervention blocks were then equally selected by random sampling within each SC/ST-based stratum (38,44). Comparison (SHG only) blocks were selected within the same district (or from a geographically adjacent district if comparison block were not available in the same district) to reduce the effect of socio-cultural diversity between study blocks (38,44). Although comparison blocks comprised of roughly similar proportions of SC/ST as intervention blocks, these comparison blocks were, however, not one-to-one matched pairs and were selected independently of the intervention blocks (44,52). The average proportion of SC/ST population in the intervention and comparison blocks were similar (45 per cent and 44 per cent). Lastly, pure control blocks (no SHG and no health intervention) were also selected based on block percentage of SC/ST as the criterion for matching with intervention and comparison blocks in the same districts (44,52).
In the second stage, Gram Panchayats (GPs) were selected within comparison and intervention blocks as per SHG population coverage, and village population size in pure control blocks as no SHGs were established in these blocks (44,52). In intervention and comparison blocks where SHGs had been established, GP’s were drawn in equal numbers from three strata of SHG coverage: 5-15 per cent, 16-30 per cent and 30-60 per cent. Outlier GPs with coverage of SHGs < 5 per cent and > 60 percent were excluded (44,52). Whereas GPs in pure control blocks (no SHG, no health) were selected based on GP population size and with a similar proportion of main village and hamlets as comparison arm (with SHG, no health intervention) to ensure similar population characteristics in these villages (44,52).
In the final stage, households were selected from all three categories of blocks following a house listing and mapping exercise to develop a sampling frame to identify the eligible woman in intervention and comparison block (44,52). While eligible women in SHG member and non-member households were selected in intervention and comparison blocks, only eligible women from non-member households were selected in pure control blocks (44,52). The listing and mapping exercise in intervention and comparison blocks showed that the number of SHG households with an eligible woman was almost equal to the sample size requirement; therefore, all SHG households with eligible women were selected for an interview in these blocks (38,44).
As SHG programs enrol one member from each household only, each individual woman in the survey represents a household; a random procedure was used to select the eligible respondent where more than one eligible woman was found in a household (44,52). Furthermore, the house listing and mapping exercise in pure control blocks also provided a sampling frame to select non-member households following a systematic random sampling (44,52).
In the successive survey rounds, data was collected from the same GPs, but not the same households or women (44,52). Furthermore, as the survey used different selection criteria at the higher level (stratified and matched block selection using SC/ST) and at a lower level (stratified GP selection based on population coverage by SHGs members and non-members), this survey analyses all eligible women in households across the sampled GPs.
Separate questionnaires were used to collect individual, household and village data. The eligible respondents comprised currently married women aged 15 to 49 years who had had a baby in the 12 months preceding the survey; household head; and village representatives. Trained data collectors administered interviews in the local language (Hindi) after obtaining verbal informed consent from respondents using computer-assisted personal interview (CAPI) package designed in the Census and Survey Processing System (CSPro), a public domain software used for census and survey data (44,52).
Data collection from eligible women used a structured survey questionnaire with open-ended questions. Knowledge of danger signs was spontaneously recalled by women and then marked against danger signs options in the questionnaire. The questionnaire separately elicited probed responses for danger signs, however, for the purposes of this research paper only spontaneously provided responses by the woman were considered. The survey instrument is provided [see Additional file 1].
To capture the individual, households, and community/area level influence on maternal danger signs among member and non-member women, this analysis merged individual-level, household, and village level sub-datasets across rounds.
Outcome and Explanatory Variables
i) Outcome variable: The main outcome variable captured eligible woman’s self-reported knowledge of maternal danger signs. A binary variable capturing no or partial knowledge (=0), and complete knowledge of danger signs (=1) was created for this study.
In the IMFHL program, the survey collected data from women on self- recalled knowledge of key danger signs in the last pregnancy retrospectively. In both rounds, eligible women irrespective of place of delivery were asked to recall multiple responses to the question “What problems/complications can a woman face during pregnancy or delivery or within 42 days of delivery which require immediate medical attention?” The responses in the interview were marked against fourteen common medical danger signs occurring across the maternity period, such as severe headache, blurred vision, loss of foetal movement etcetera. As some of the fourteen signs also comprised signs that were common to normal pregnancy, a clinical review was undertaken by a medical doctor for this study to select key danger signs indicative of serious direct (preventable) causes of maternal mortality for this analysis. The dangers signs selected identify complications in pregnancy, delivery and the post-partum period that are associated with main direct causes of mortality in India and other high maternal burden countries namely haemorrhage, eclampsia, severe maternal infections, prolonged labour.
ii) Main Explanatory variables: The main explanatory variable, the IMFHL intervention, was categorised into five levels based on household’s exposure to IMFHL program. An ordinal variable was created to capture the program’s main effect on women’s knowledge, that is, the change in women’s knowledge across the program levels of exposure: intervention (SHG plus health), comparison (SHG only) and control (no SHG, no health) villages. Woman’s SHG membership is determined by the eligible woman being herself a member of the SHG or belonging to a household where someone else (e.g., mother, mother-in-law, sister in law) is an SHG member. The coding of the IMFHL intervention variable with description is shown as follows:
- Group 0: Comprised of households that were not SHG members (non-members) and were in villages without any program intervention (pure control households).
- Group 1: Comprised of households that were not SHG members (non-members) but were in program villages where the SHG program alone was implemented (diffusion-control households).
- Group 2: Comprised of households that were not SHG members (non-members) but were in program villages where the SHG program plus Health intervention was implemented (diffusion-control households).
- Group 3: Comprised of households that were SHG members in program villages where only SHG program was implemented (comparison households).
- Group 4: Comprised of households that were also SHG members but were in villages where both the SHG program and additional health intervention was provided. Only these households received health intervention through the SHG (intervention households).
iii) A survey round variable was created to assess the effect of program intervention on women in round II (2017) compared to round I (2015).
iv) Confounding Variables: The analysis included a comprehensive set of confounding variables that were identified from the maternal health literature and captured at the individual and community levels. They represented socio-demographic, health, and community factors. The individual maternal health variables included parity, history of past pregnancy-related complication, number of previous pregnancy loss, knowledge of minimum ANC visits required, and place of last delivery. Whereas maternal health system utilisation variables were the quality and number of ANC received, duration of stay in a hospital after delivery and intensity of contact with the frontline worker in last pregnancy. Socio-demographic variables represented the type of family (nuclear versus joint/extended), household head’s religion, social caste, and the years of education attained by an eligible woman and her husband. Economic variables capturing household poverty included the working status of eligible woman and a composite variable of household wealth quintile. The wealth quintile variable was constructed for this analysis using Polychoric Principal Component Analysis (PCA) combining household assets and amenities to evaluate program impact across five gradients of poverty from the poor (reference category) to the poorest.
Under the IMFHL program, the total sample size was calculated considering a seven percentage point increase in primary health indicators, for example, institutional delivery, number of antenatal care visits and others, after program implementation, with an 85 percent power to detect changes, the usual 5 percent level of significance and a design effect of 2 (44,52). In this study, a sample comprising 17,232 eligible women was used, out of which 41 per cent or 7,144 women were SHG members.
Statistical Data Analyses and Models
This study examined the effect of an IMFHL intervention on knowledge levels of maternal danger signs in women from households that were either SHG members or non-members in rural Uttar Pradesh.
The dependent variable in this analysis was a binary variable that represents the knowledge of maternal danger signs about obstetric complications among women. Consequently, the program impact was evaluated using multivariable logit regressions to establish the program’s effect while controlling for other confounders in models. While model I established the main effects of both the intervention and the survey round, model II included an interaction term to draw out the interaction effect of program exposure with survey round. Confounders related to individual health and health system were included in model III, while the full model IV included sociodemographic-economic and area-level variables. The results are reported as Odds ratios (OR) with associated 95% confidence intervals, and the model of best fit was evaluated using the Consistent Akaike Information Criteria (CAIC) and Bayesian Information Criteria (BIC), two valid measures of model fit, parsimony and model selection (55). Estimates of effects were reported with associated 95% confidence intervals as suggested by NEJM guidelines(56). The a priori level of significance was set at the usual 5% alpha and all p-values reported in Tables using the asterisk convention: ***: p < 0.01; **: p < 0.05; *: p < 0.10 (57), with the last category meant to show that a “trend towards statistical significance” has to be noted (58-60). All analyses were performed using Stata 16 (Statacorp, USA).
Table 1 presents the summary statistics of the outcome variable used in this analysis.
A detail list of the knowledge of all danger signs that were retrospectively collected from eligible women is provided in Table 1, which shows the frequency and percentage of danger signs across different phases of pregnancy/delivery and 42 days post-partum. The table, categorised across member and non-member households showed that most women (37 per cent) recalled severe abdominal pain, while only a minority (1 per cent) recalled danger signs related to placental expulsion or umbilical cord issues. Moreover, the distribution of danger signs showed that women were more likely to recall those danger signs that occurred in the pregnancy and delivery period as compared to the post-partum period. The Table 1 also shows that across member and non-member households, the majority (>69 per cent) of women knew danger signs in all three phases of pregnancy, delivery and post-delivery with only a minority (<15 per cent) of all women reporting no knowledge.
Table 2 presents the summary statistics of the explanatory variables for eligible women categorised across (SHG) member and non-member households that capture all type of households’ program exposure and associated characteristics and relevant factors (individual and community levels).
The sampled women in this analysis comprised of more women that were non-member (59 per cent) than SHG members. However, an almost equal proportion of women spread across both survey rounds, and a proportionate number of households were allocated within each level of program exposure.
In both groups, eligible women reported a mean parity close to 2.4 reflecting near current Indian fertility rates (median 2.2, range 2.1-4), and among them, close to 50 per cent women experienced an obstetric complication in their last pregnancy/delivery or post-partum period Furthermore, a quarter of all women across both groups reported experiencing a pregnancy loss either due to induced or spontaneous abortion. Although 83 per cent women reported delivering in an institution for their last pregnancy (public or private), only a minority (26 per cent) had received the minimum four antenatal care (ANC) visits in their last pregnancy with all required tests done in the last check-up, that are: urine, blood pressure, weight, abdominal and ultrasound tests. The summary statistics showed that among those women with an institutional delivery, the majority (70 per cent) were discharged within 12 hours after delivery which is less than the recommended 48 hours stay post-delivery for normal deliveries (61). Furthermore, only 10 per cent of total women reported receiving the recommended minimum of three post-natal care visits within the crucial first seven days post-delivery when maternal and neonatal complications commonly occur.
The villages were on average 5.4 kilometres (km) from the closest Primary Health Centre (PHC) and about 1.47 km to the closet town.
The summary statistics reflecting household’s economic status showed that only 16 per cent of women reported they were working to earn in cash or kind at the time of the survey, while almost 45 per cent belonged to households that were living below the poverty line -an income-based measure of household poverty.
Moreover, most women (60 per cent) were living in joint/extended households which is common in rural Uttar Pradesh. Almost 92 per cent of all women identified themselves as being part of Hindu households, the majority religion in Uttar Pradesh and India. Moreover, 45 per cent of women reported belonging to Scheduled Caste and Tribes (SC/ST), a proxy for social margination in the rural Indian content. Women across member and non-member households were comparable in relation to age (mean 26 years), and level of education. While 66 per cent of all women interviewed reported having some level of education, the summary statistics showed that husbands overall were more likely (88 per cent) to have had received some level of schooling. For the dependent variable, knowledge of danger signs in women, most women (70 per cent) had full knowledge of danger signs in all phases of pregnancy/delivery and post-delivery.