Improving Birth Preparedness and Complication Readiness Through an Integrated Micronance and Health Literacy Program:Evidence From Rural India

Background:Despite maternal mortality rate almost halving over the past two decades, approximately 295,000 deaths occurred annually in 2017 globally. They occur in high burden regions where most preventable deaths are precipitated by delays in maternal care provision. Recently, a novel rapidly scalable community-based program combining maternal health literacy delivery through micronance-based women-only self-help groups (SHG) was implemented in rural India. We evaluated the impact of the IMFHL program on BPCR practice by women in one of India’s poorest states-Uttar Pradesh, adjusting for community, household and individual variables. We also investigated if there is a diffusion of knowledge of BPCR from SHG members receiving the health literacy intervention to non-members in program villages. Method: Using secondary survey data from the IMFHL project, we analysed data from 17,244 women across SHG and non-member households in Uttar Pradesh. We performed multivariable logistic regressions to estimate adjusted IMFHL program’s effects on maternal BPCR practice. Results: Membership in SHG alone is positively associated with BPCR practice with 17 percent higher odds compared to women in villages without the programs. Furthermore, the odds of practising complete BPCR increases by almost fty percent when a maternal health literacy component is added to the SHGs. A diffusion effect was found for BPCR practice from SHG members to non-members only when an added health literacy component was integrated into the SHG model. The main effect of the IMFHL program on BPCR practice remained stable in the presence of confounders suggesting that the delivery of health messages is undiluted by other non-programmatic factors at the individual, household and community level. Among confounders, only the poorest households as captured by the wealth quintile were signicantly negatively associated with BPCR practice. Conclusion: The public health literature promotes the adoption of BPCR practices to improve maternal health outcomes. This study shows that SHGs exert both a dissemination effect of planned health behaviour within members, as well as facilitate a diffusion effect of the natural transfer of BPCR from members to non-members when SHGs are enriched with a health literacy component The micronance-based an in routine of maternal health the integrated micronance and health literacy (IMFHL) program that took the micronance platform to provide pregnant and recently delivered women with health messages. Information was provided to identify dangers related to pregnancy, childbirth and post-delivery complications, and to adopt birth preparedness and complication readiness steps in order to reduce delays associated with seeking care. The IMFHL program hypothesised that women in households receiving nancial access additional health information through the Self Help Groups(SHG) would be able to reduce health care-seeking delays during maternal complications. Micronance based SHG platform in UP and other low-development regions of India have shown that integrated health literacy and SHG interventions can improve routine maternal health practices such as antenatal care utilisation, institutional deliveries also improve newborn health practises at home such as cord care and timely breastfeeding(58, 59). However, the impact of membership in a SHG based health program on the practice of BPCR among SHG members has not been evaluated. Also, the potential of the program to diffuse health information from members to non-members in where SHGs are established is untested. Diffusion of health messages as dened as the unplanned natural spread of health information and practices in a community is differentiated with the dissemination or planned spread of health information under programs(60, 61). This study evaluated both the impact of membership on the dissemination of BPCR practice among members, as well as the diffusion of health practice from SHG economic and area-level confounders. The individual model coecients were estimated using the maximum likelihood estimation method. The Odds Ratio (OR) values are reported, as measures of the strength of association, along with their associated 95 per cent condence intervals (95% CI) and p values. ANC with key tests done in the last ANC visit reported lower odds of BPCR practice compared to women who received less than four visits and with incomplete tests (OR=0·90, 95%CI:0·82-0·99, p<0·05). Other variables reecting individual health status such as parity, pregnancy complication experience and health system variables such as contact with front line workers were found be statistically insignicantly associated with BPCR holding other variables constant up to a p-value of 10 per cent. Separate multivariate analyses conducted in this study (results not shown), found that more health variables became statistically associated with BPCR, albeit negatively when round and the interaction term, IMFHL by round, were removed from the full regression models· For example, in the absence of round and IMFHL by round, women with any history of previous pregnancy loss, and those who experienced pregnancy complication were now almost 8 per cent less likely to practice BPCR compared to those with no previous pregnancy loss history (OR=0·92, 95%CI: 0·86-0·99, p<0·05) and experience of complication (OR=0·94, 95% CI: 0·88-1·00,p<0·10). These results suggest that the round variable and the interaction term of IMFHL by round are exerting a subduing effect on the association of health status variables with the outcome in this analysis. As health behaviours, such as BPCR, may also be inuenced by wider social, demographic and economic factors, in model IV tested the effect of these confounders that are described next. The study evaluation focussed rst on the IMFHL program’s main effect on BPCR practice, independent of potential confounders. The evaluation then adjusted for potential confounders and separately assessed the association of multiple confounders on household decision making for BPCR adoption. The results show that SHGs exert both a dissemination effect of planned health behaviour within members, as well as a diffusion effect of the natural transfer of BPCR from members to non-members but only when a health literacy component is added to SHGs. Based on the study ndings, the authors would recommend integrating health literacy components into SHG networks and similar community groups to promote household’s adoption of complication readiness plans to reduce delays in seeking health care during maternal complications. The inclusion of health literacy in a micronance platform presents an example of a novel community-based strategy that has the potential to interrupt the cycle of household poverty and low maternal healthcare-seeking. Lessons from the IMFHL program implementation in India provide the opportunity to adapt SHG models implemented elsewhere by adding health literacy components. Beyond India, micronance is promoted globally as a poverty alleviation program in low resource settings where the burden of maternal mortality remains high. This can assist other low resource settings in accelerating progress towards the maternal health goal of the Sustainable Development Goals (SDG)


Background
Improving maternal health is a global priority in the attainment of the Sustainable Development Health Goals (SDG). Health system reforms and improvements in overall socio-economic conditions in low resource countries has led to an almost fty per cent reduction in maternal deaths by the Millennium Development Goal's end (MDG) in 2015 compared to 1990s level (1)(2)(3). However, the rate of decline was inadequate to meet the global MDG target of 75 per cent reduction in maternal mortality rate (MMR), de ned as the number of maternal deaths during a given time period per 100,000 live births during the same time period (3,4). Consequently, more than 295,000 maternal deaths occurred in 2017 mostly due to preventable causes in the maternity period, namely due to hypertensive disorders of pregnancy (pregnancy-induced hypertension-PIH/Eclampsia), severe bleeding, obstructed labour and sepsis (4)(5)(6). Despite established treatment options for these causes, the high number of unnecessary deaths stemming from these re ect the residual gap from the MDGs. The current SDG era has an even stringent target of reducing MMR to 70 per 100,000 live births by 2030 for all countries (5). Achieving the SDG's maternal health goal requires concerted action in select high poverty and low resource countries in Sub Saharan Africa and South Asia, where almost 99 per cent of global preventable maternal deaths occurred (1,7,8). After Nigeria, India accounts for the world's second-highest maternal deaths with almost 45,000 maternal deaths annually (9,10). Although India has invested substantially in improving health service delivery for its predominantly rural population, high maternal deaths remains a challenge (6,8). Attaining the SDG goal requires speci c improvements in high burden regions like India to accelerate maternal mortality decline beyond what was observed in the MDG era, and to adopt novel strategies that bridge health system inadequacies with civil society actions (11,12).
The World Health Organisation (WHO) recommends skilled birth attendance and access to timely emergency obstetric care during maternal complications as crucial interventions to improve maternal and neonatal survival (13). Improving birth preparedness and complication readiness (BPCR) is a crucial strategy to reduce maternal mortality through increasing awareness and preparedness around emergency obstetrics care in low resource settings (13). WHO recommends BPCR as a process of planning by women, households and communities to adopt key guided steps that help families navigate complex decision-making during maternal complications about when to act and where to seek care promptly. According to the WHO, BPCR broadly comprises of the following preventative steps: decide on the place of delivery( home or health facility); identify the facility that could provide emergency care; identify people to accompany the woman to the hospital for delivery; identify people to take care of children at home; save/arrange money for delivery expenses or in case of emergency; make advance arrangement of transportation to go to the facility; identify institution where to rush in case of emergency (13).
Pregnant women and their households are expected to receive BPCR advice as part of the antenatal care services by the health worker or through home visits (13). While the evidence of BPCR effects on maternal mortality reduction is mixed, a 2017 systematic review showed that the adoption of BPCR as part of an intervention package has the potential to improve care-seeking and encourage timely use of health facility in maternal; and neonatal complications (14). A study from an urban slum in India showed that skilled attendance at birth was almost 30 per cent higher among well prepared compared to less prepared mothers (15). Another study from Uganda showed that past experience of maternal complication and awareness of maternal danger signs are predictors for higher BPCR knowledge (16). Moreover, despite demonstrated bene ts of BPCR, studies showed that BPCR is challenging to achieve in low resource settings where health system services are lacking, and households have low education and high poverty levels (2,12,(16)(17)(18)(19). Studies from other high maternal burden countries in Africa showed that women's BPCR knowledge levels remain low due to health system factors related to poor quality of health education, absence of functional referral system and failure of the health system to implement an adequate policy on BPCR (16,20).
While India has prioritised institutional delivery through a conditional cash transfer scheme, substantial gaps persist in early identi cation and care of maternal complications, especially in the antenatal and post-natal period for rural populations (10,21,22). Inadequate referral and limited Emergency Obstetric Services (EMOC) place women at higher risk of adverse maternal outcomes in rural areas (12,22,23). Heath system inadequacies are compounded by community-related delays in seeking care and reaching health facilities due to nancial barriers and lack of birth and complication readiness (22,(24)(25)(26)(27)(28). Furthermore, studies showed that women from poor households in rural areas, belonging to certain socio-demographic segments, have a higher risk of maternal deaths compared to wealthier women in urban areas. (29)(30)(31). In most instances in rural India, decision-making among households during maternal emergencies is in uenced by past experiences with the health system and a tendency to follow traditional cultural beliefs (29,31,32). Previously delays in care-seeking have been explained using the widely validated three-delays framework as an approach to develop interventions to improve care-seeking steps (19,24,31,(33)(34)(35). The framework identi es the recognition of danger signs in the household as the rst type of delay, delays in reaching an appropriate facility as the second delay and lastly, delays in receiving adequate care in the health facility (35). While some view household factors as mainly accounting for the rst and second delays, and attribute the third delay exclusively to the health system, in an optimal setting, adequate provision of maternal health services would ensure that none of the delays occurs (19,35).
Most avoidable maternal deaths occur in disempowered segments of society, among women who face multiple deprivations related to severe poverty, low education and social marginalisation (29,31,(36)(37)(38). Verbal autopsy studies in India with family members of deceased women in rural and tribal regions showed that in many settings, women lack adequate information about pregnancy-related danger signs and are inadequately prepared during pregnancy-related complications (25,29,39). In the absence of complications readiness and adequate referral services, families are forced to adopt a harmful wait and watch approach to gauge the severity of underlying maternal distress, often with adverse outcomes (25,29,39). Preparing for birth and complications has, therefore, emerged as a crucial intervention for rural communities to reduce subsequent delays in maternal care-seeking (24,40).
Maternal death reviews showed that select high poverty and low development states in the north of India contribute the majority of maternal deaths (8,10,30). Notably, the state of Uttar Pradesh(UP) which has a population of 200 million and high concentration of rural population(79 per cent), and poverty(31 per cent live below the poverty line), has the highest number of maternal deaths in the country (8,10,32). MMR in Uttar Pradesh stands at 201 per 100,000 live births compared to the national MMR of 170 per 100,000 live births (10,42).
UP also accounts for a quarter of the neonatal mortality burden of India and shares similar socio-economic, demographic and health system characteristics with other high mortality regions in India and South Asian countries (10). In 2016, UP reported 43 per cent home deliveries, among which only 26 per cent were attended by skilled birth personnel (43,44). Also, among the 33 per cent of pregnant women in rural UP who reported receiving the minimum three antenatal care(ANC) visits (now revised to four), only eight per cent used the recommended 100 tablets of Iron Folic Acid (IFA) suggesting gaps in ANC service delivery (22). Studies from India and rural UP showed that, despite higher institutional deliveries, de ciencies in ANC limits transmission of key maternal health information such as BPCR to pregnant women (22,43).
In a rural setting where antenatal and referral services are inadequate to ensure monitoring of pregnancies irrespective of risk, households' adoption of BPCR, therefore, assumes greater importance (22,43).
Limitations in health systems in low resource settings have given rise to alternative approaches incorporating community efforts to mobilise, educate and empower women to improve maternal health outcomes. While many community mobilisation programs showed success as pilot programs under controlled environments, scaling of these programs to a larger population(s) is challenging. Mainly, due to dilution effects of resources when going from small-scale pilot in a controlled environment to a population level program with real-life conditions (44). Moreover, when community health programs are implemented, access to information and resources is often limited to direct program bene ciaries, potentially excluding speci c population subsections or harder to reach groups (45,46). Studies have found that the spread of health information beyond the program participants in a community is in uenced by the number and quality of social networks that enhance interpersonal contact to improve health message uptake (47). The role of social networks and community groups is especially pertinent in rural Uttar Pradesh where studies show that up to 70 per cent of women can have no access to traditional sources of media such as radio, television and newspaper (48). These disadvantages in receiving information are compounded by low levels of education that can affect the processing of health information as well as changing health behaviour based on health information (47,49). In such settings, social networks that are accepted within the community can facilitate access to health information and reduce disparities among different population groups (50,51). Evidence also shows that health programs led by the community promote the sharing of new knowledge that can expand the reach of community interventions (50,52,53). However, limited research has been conducted in low resource settings to examine whether health information provided through community program facilitates health promotion beyond the primary bene ciaries of the program.
Recently, a promising community-based intervention that builds on the concept of empowering poor rural pregnant women by providing them with health literacy and nance-based assistance through women-only micro nance groups was implemented in rural Uttar Pradesh. The micro nance-based program builds on other community-based programs that showed an improvement in routine indicators of maternal health (54)(55)(56)(57). This study evaluated the integrated micro nance and health literacy (IMFHL) program that took advantage of the micro nance platform to provide pregnant and recently delivered women with health messages. Information was provided to identify dangers signs related to pregnancy, childbirth and post-delivery complications, and to adopt birth preparedness and complication readiness steps in order to reduce delays associated with seeking care. The IMFHL program hypothesised that women in households receiving nancial access and additional health information through the Self Help Groups(SHG) would be able to reduce health care-seeking delays during maternal complications. Micro nance based SHG platform in UP and other low-development regions of India have previously shown that integrated health literacy and SHG interventions can improve routine maternal health practices such as antenatal care utilisation, institutional deliveries and also improve newborn health practises at home such as cord care and timely breastfeeding (58,59). However, the impact of membership in a SHG based health program on the practice of BPCR among SHG members has not been evaluated. Also, the potential of the program to diffuse health information from members to non-members in villages where SHGs are established is untested. Diffusion of health messages as de ned as the unplanned natural spread of health information and practices in a community is differentiated with the dissemination or planned spread of health information under programs (60,61). This study evaluated both the impact of membership on the dissemination of BPCR practice among members, as well as the diffusion of health practice from SHG -members to non-members within program villages.
This research builds on limited evidence on the effectiveness of membership in integrated micro nance program with health literacy component on birth preparedness and complication readiness practice among women in rural areas (58,59,(62)(63)(64). The objective of the study is to generate evidence for community health program managers and policymakers on the role of the program as a potentially scalable model for improving maternal health outcomes in India. The study also examines the diffusion effect generated by the program that permits diffusion of BPCR practices and knowledge from members to non-members in rural India.

Method Study Aim and Design
This study evaluated the impact of membership in an integrated micro nance and health literacy (IMFHL) program on the practice of birth preparedness and complication readiness (BPCR) by women during their last pregnancy in rural Uttar Pradesh. The study also, evaluated whether the IMFHL program contributed to the adoption of BPCR among women that did not participate in the IMFHL program as SHG members through a process of diffusion of behaviours from member households to neighbouring non-member households in villages where the program was implemented.
The study uses survey data, collected under the IMFHL project in two rounds, capturing the program's pre-intervention characteristics in round I in 2015 followed by round II of data collection two years into the program in 2017 (58). Under the IMFHL project, a quasi-experimental survey design was used to collect cross-sectional survey data in each round to evaluate the program's impact on mothers' knowledge and health behaviours in low-income households in villages across UP (58)

IMFHL Program Context
The IMFHL program integrated a maternal and neonatal health literacy component within a micro nance based SHG platform to provide pregnant and recently delivered women with health messages in rural UP. The program was implemented over ve years (2012-2017) through a consortium comprising a technical lead, an implementing partner, a research partner and an evaluation partner (58). A detailed description on IMFHL program context, selection of intervention, comparison and control blocks, is available elsewhere (58).
The design of the health literacy intervention was informed by the three delays framework and from successes of the previous nonmicro nance based participatory community health programs such as the Shivgarh trial in UP, and the Makwanpur trial in Nepal (35,52,55). Health messages were directly provided to women by the IMFHL program using various inter-personal communication strategies, such as an invitation to SHG meetings for health discussion; postal letters to pregnant and new mothers with key health messages timed to delivery and neonate care; reinforced messages through house visits; and exposure to community health video shows developed by the program (58).
Woman's SHG membership was determined when the woman was a SHG member herself or belonged to a household where someone else, usually a mother in law or sister in law, was a SHG member (58). In the latter case, perfect communication was assumed between the family members. These women (SHG members) represented the rst tier of bene ciaries who directly received health information from the SHG, encouraging them to adopt desired health behaviours to promote maternal and neonatal health. The IMFHL project investigators also expected health information to reach non-member women in the same village where SHGs were established and health literacy provided. As members and non-members lived in the same village/environment, the IMFHL project investigators postulated that women receiving health information as SHG members would communicate new knowledge to neighbouring non-member households mainly through informal communication networks that would gradually lead to the adoption of desired behaviours. Previous social network analysis studies conducted on the IMFHL program showed that SHG membership increased recently delivered mother's health advice networks and linkages with community health workers (65). The extent of these social networks in the transfer of health information and practices to non-members is also evaluated in this paper. This diffusion effect of knowledge from members (tier I) to neighbouring non-members (tier II) is depicted in gure 1 below.

Study Setting and Participants
Sampling Procedure The survey sampling design followed Uttar Pradesh's administrative hierarchy and collected data from households in gram panchayats (GP) or villages from 70 blocks in 20 districts, aiming for a representative sample from the program's coverage area (66). Moreover, while the same gram panchayats, blocks and districts, were visited in both survey rounds; different households were sampled and interviewed in each round. A GP is the smallest administrative region in blocks where SHGs were established and from where household survey data was collected. The survey design used a three-stage sampling approach to select survey blocks, GPs or villages and nally, household as depicted in gure 2. A detailed description of the selection strategy is available elsewhere (58).

Sample Size
Different outcome measures were used by the IMFHL project investigators to calculate sample size based on the net increase in health behaviours due to program exposure with an expected response rate of 90 per cent (58). The sample size was estimated considering a net increase of seven percentage point in key reproductive maternal, neonatal and child health indicators (RMNCHN), for example, institutional delivery, the number of antenatal care visits and others, after program implementation with 85 per cent power to detect the change, with the usual 5 per cent level of signi cance and a design effect of 2 (58). Moreover, in order to detect the diffusion from SHG households to nonmember households in program villages, the project investigators also separately determined the sample size based on an estimated ve per cent net change in RMNCHN indicators over the two rounds with 80 per cent power, 95 per cent con dence interval, and a design effect of 1.5 (58).
The sample used in this study comprised a total of 17,244 eligible women, of whom 59 per cent or 10,097 women did not belong to SHGs, and 41 per cent or 7,147 women were SHG members. The data was collected at the individual, household and community levels using separate structured questionnaires. The survey collected data from currently married women aged 15 to 49 years who had delivered an infant in the 12 months preceding the survey, and from the household head and village representatives. The interviews were conducted by data collectors trained by evaluation partner who administered interviews in the local language, Hindi, after obtaining verbal informed consent from respondents using computer-assisted personal interview (CAPI) package designed the Census and Survey Processing System (CSPro), a public domain software used for census and survey data (58). As this study sought to capture individual, household and community level in uences on BPCR practice among members and non-member women, we customised our study dataset by combining individual-level, household and village levels sub-datasets across rounds to capture the different levels of in uence on birth preparedness and complication readiness.

Statistical Analysis
Multivariable logistic regression modelling was used to evaluate the intervention program's impact on the level of birth preparedness and complication readiness practice among women in rural Uttar Pradesh. The main explanatory variable, IMFHL intervention, was categorised according to four levels of household's exposure to IMFHL program. Covariates or other confounding variables were chosen based on the broader maternal and child health literature to represent individual, household, and community-level characteristics. Four separate regression models were tted to the data with the rst model (Model I) establishing the program's main effect without any confounding variables but including the survey round variable. The second model (Model II) included an interaction term (IMFHL intervention by survey round) to draw out the change over time, in the effect of IMFHL program exposure. Confounders related to individual health were included in the Model III, while the full model, Model IV included socio-demographic, economic and area-level variables.
Outcome, explanatory and confounding variables i. Outcome variable: The outcome variable, birth preparedness and complication readiness (BPCR) practice captured eligible women's selfreported BPCR steps that were taken during the last pregnancy. A binary variable was constructed with '0' representing partial or no BPCR preparation, and '1' representing complete BPCR de ned as when households practised all birth preparedness and complication readiness steps during last pregnancy as outlined by the WHO. In the IMFHL program, women belonging to SHGs in the intervention villages (SHG and health literacy) were provided with the guidance of key steps required for complete BPCR in the pregnancy. The survey collected data from all women on self-recalled steps of BPCR undertaken in the last pregnancy. In both survey rounds, eligible women, independent of the place of delivery, were asked to recall multiple responses to the question "What advance preparation did you/your family members make to manage in case of any pregnancy/delivery complications?". The interview responses were marked against eight key steps required to ful l birth and complication readiness such as: 'Decided on the place of delivery-home or health facility' 'Knew the facility that could provide emergency care'; 'Identi ed institution where to rush in case of emergency'; 'Identi ed people to accompany the woman', 'Identi ed people to take care of children at home', 'Saved/arranged money for delivery expense or in case of emergency', 'Advance arrangement of transportation to go to the facility' and 'others -cloth, soap'. Although individual steps corresponded to either birth preparedness or only complication readiness were practised, complete BPCR was determined when a household followed all steps, in accordance with current WHO guidelines on BPCR.
According to WHO, complete BPCR was de ned when the woman and household ful lled all the steps that provided families a greater opportunity for readiness in the event of a maternal emergency.
ii. Main explanatory variable. The exposure variable, the IMFHL intervention, comprised of four levels based on households' exposure to the IMFHL program. An ordinal variable was created allowing to evaluate the program's main effect on the practice of BPCR, that is, the magnitude of change in BPCR practice across levels of the IMFHL program exposure: intervention (SHG plus health), comparison (SHG only) and pure control (no SHG, no health) households. The coding of the IMFHL explanatory variable with description is shown below: 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 either the SHG program alone or SHG program plus Health intervention was implemented (diffusion-control households).
Group 2: Comprised of households that were SHG members in program villages where only SHG program was implemented (comparison households).
Group 3: 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 literacy intervention through the SHG (intervention households).
iii. Additionally, a survey round variable was created to assess the effect of intervention program over time (change in women's BPCR in round II in 2017 compared to round I in 2015 when the program was yet to be implemented).
iv. Confounding variables: A comprehensive set of confounding variables capturing individual, household and community-level characteristics were identi ed from previous maternal and child health literature and included in models III and IV. Confounding variables in model III represented eligible woman's maternal health status and broader maternal health service utilisation indicators such as parity, the experience of pregnancy-related complication in a previous pregnancy, number of previous pregnancy loss, knowledge of minimum ANC visits required, place of last delivery. Additional socio, demographic and area-level characteristics were included in the nal Model IV, namely: type of family (nuclear versus joint or extended), religion, household's social caste, education level of the eligible woman and her husband. Model IV also included a household wealth quintile, constructed for this analysis using Polychoric Principal Component Analysis (PCA), combining all household assets and amenities to evaluate BPCR association across ve wealth gradients extending from marginally poor (reference category) to the poorest households.
Summary Results Table 1 provides results of individual steps undertaken in the study sample towards birth preparedness and complication readiness (BPCR) practice in last pregnancy among women aged 15-49 in rural Uttar Pradesh (n=17,244). Table 1 and Figure 3 reveals, only 24 per cent of all women in this analysis had not practised any steps for birth or complication readiness; however, 76 per cent women had some level of BPCR readiness with 49 per cent had undertaken full BPCR in last pregnancy. Also, 6 per cent of women had taken steps ful lling only birth preparedness while only 21 per cent of all women had undertaken the three steps required for complication readiness. Table 2 presents the summary results for eligible women categorised across (SHG) member and non-member households across all variables used in this paper. The study sample comprised of more women from non-member households (59 per cent) but with an equal allocation of households within various levels of IMFHL program exposure levels. Surveyed women had a mean parity of 2·4(median 2, range 0-9) which similar as in rural UP, but, it is higher than the current Indian fertility rate (median = 2·2, range: 2·1 -4)(44). Moreover, close to 48 per cent of all women reported experiencing a pregnancy complication in their last pregnancy, while only a quarter (25 per cent) had suffered a spontaneous or induced abortion. Furthermore, women on average had four contacts with frontline workers in the previous pregnancy, with 26 per cent of women receiving four ANC with all tests during the last pregnancy and only nine per cent women receiving the vital three post-natal visits in the rst week after delivery. Contrary to this, 83 per cent of all women reported having an institutional delivery in a public or private facility. However, women reported a shorter duration of stay in a health facility after delivery with almost 67 per cent discharged earlier than the Indian government mandated 48 hours stay for normal (non-emergency or caesarean) deliveries (22). In terms of area-level characteristics, villages were at an average distance of 5·4 kilometres (km) to the closest primary health centre (PHC). Lastly, about 46 per cent of households had access to the Below Poverty Limit Card (BPL) By design, the IMFHL program was implemented in poor rural areas of Uttar Pradesh which the statistics show with only 16 per cent of women working to earn cash or kind. Moreover, within households, the majority (92 per cent) of households belonged to the Hindu religion, the predominant religion in UP and India, while 45 per cent identi ed themselves as belonging to Scheduled caste and tribes. The sampled women had a mean age of 25 years, with 66 per cent of all women have received some form of education, while husband's education level showed 83 per cent had received some form of schooling. Finally, according to the levels of BPCR, the outcome variable, 52 per cent of all SHG households had a complete BPCR plan in place in the last pregnancy, while 46 per cent of women in non-member households had all BPCR steps in place in the last pregnancy.

Results
The results from multiple logistic regression models are presented in Table 3 with models I and II providing the main effects of program exposure levels and survey round on BPCR practice. Models III and IV in Table 4 shows the results of IMFHL program exposure survey round on BPCR adjusting for socio-demographic, economic and area-level confounders. The individual model coe cients were estimated using the maximum likelihood estimation method. The Odds Ratio (OR) values are reported, as measures of the strength of association, along with their associated 95 per cent con dence intervals (95% CI) and p values.

Model I: IMFHL program's effects on birth preparedness and complication readiness
Women belonging to SHG alone, and those receiving health intervention from SHG, both reported higher odds of practising birth and complication readiness in last pregnancy compared to the non-member in control(non-program)villages (Table 3). While, women in SHG households were 17 per cent or 1·17 times more likely to have practised BPCR in their last pregnancy (OR=1·17, 95% CI: 1·07-1·29, p<0·01) compared to non-members in control villages. SHG members who received additional health literacy were 48 per cent or 1·48 times more likely to practice BPCR (OR=1·48, 95% CI: 1·35-1·63, p<0·01) compared to non-members in control villages. However, results also showed that non-member women living in villages where either SHG or SHG plus health intervention was implemented, were 11 per cent less likely to practise BPCR compared to non-member women in control villages, that is, where no SHG nor health was implemented (OR=0·89, 95% CI:0·82-0·97, p<0·01). The models also evaluated the effect of program implementation overtime on BPCR using the variable, survey round. The results showed that women in households interviewed in round II were almost 63 per cent or 0·37 times less likely to have practised BPCR compared to households interviewed in the round I given all the levels of program exposure (OR=0·37, 95% CI:0·35-0·39, p<0·01).

Model II: Interaction effect of IMFHL program and Survey Round
The program exposure is expected to mature in implementation between round I in 2015 when the maternal health literacy component was started, and in 2017, two years into the intervention when round II was conducted. The program effect on BPCR, therefore, is expected to vary over round I and II. Model II, therefore included an interaction term of program exposure with survey round to capture the project's in uence on BPCR over time.
Model II showed that the inclusion of the interaction term changed the direction and magnitude of the impact of the program exposure (for women in SHG plus health households) on the practice of BPCR compared to the same household type in model I. The results showed that these women now reported a statistically insigni cant four per cent lower odds of practising BPCR compared with non-member women in control villages (OR=0·96, 95%CI: 0·84-1·11, p>0·10). The shift in direction and magnitude observed for women in SHG plus health households between models I and II are suggestive of an underlying signi cant interaction effect of program exposure and survey round.
The results from the two-way interaction in Model II showed that women in SHGs that received health literacy were 2·21 times more likely to practice birth preparedness and complication readiness in the second round compared to same household type in the rst round (OR=2·21, 95%CI: 1·82-2·68, p<0·01). This means that women in the intervention group were roughly 2·122 (=0·96*2·21) (95%CI:1·52-2·97) times more likely to practice BPCR in survey round II compared to women in control villages. Similarly, the results showed that non-members in program villages reported 1·73 times higher odds of BPCR practice in round II compared to the same household type in the rst round (OR=1·73, 95%CI: 1·46-2·05, p<0·01). The difference in odds ratio suggests that the health literacy intervention impacted BPCR practice after program implementation as re ected in round II. These results indicate that a diffusion effect occurred from SHG women receiving health literacy to non-member women in the same villages, with women in this group being roughly 1·194 (=0·69*1·73) (95%CI:0·89-1·57) times more likely to practice BPCR. However, the results of the two-way interaction effect revealed that women in SHG only households in round II did not report a signi cant change in BPCR practice in last pregnancy compared to the same group of women in round I as shown in Table 3. Therefore, the results show that the source of the diffusion of BPCR practices in non-members is due to the health literacy intervention in SHG villages, and not due to the SHG only network.

Model III and IV: Effects of IMFHL program adjusted for confounders
The effect of program exposure was further evaluated in model III and IV in Table 4 when confounders were added. Overall, the effect of program exposure on BPCR observed previously in model II remained unaffected when adjusted for confounders in model III and IV, suggesting that the outcome is mainly in uenced by program variables rather than confounders. However, the results of key confounders in model III and IV holding the main program exposure variable and others constant are reported as follows.

Effects of individual health and health system variables
Among variables re ecting the continuum of maternal care, women with the correct knowledge of ANC visits were more likely to practise BPCR compared to those with incorrect knowledge (OR=1·06, 95%CI: 0·99-1·13, p<0·10). Similarly, women who were discharged earlier from the health facility after delivery reported higher odds of practising all steps of BPCR in last pregnancy (OR=1·15, 95%CI: 1·00-1·33, p<0·05).
Contrastingly, women who reported receiving four ANC with key tests done in the last ANC visit reported lower odds of BPCR practice compared to women who received less than four visits and with incomplete tests (OR=0·90, 95%CI:0·82-0·99, p<0·05). Other variables re ecting individual health status such as parity, pregnancy complication experience and health system variables such as contact with front line workers were found be statistically insigni cantly associated with BPCR holding other variables constant up to a p-value of 10 per cent. Separate multivariate analyses conducted in this study (results not shown), found that more health variables became statistically associated with BPCR, albeit negatively when round and the interaction term, IMFHL by round, were removed from the full regression models· For example, in the absence of round and IMFHL by round, women with any history of previous pregnancy loss, and those who experienced pregnancy complication were now almost 8 per cent less likely to practice BPCR compared to those with no previous pregnancy loss history (OR=0·92, 95%CI: 0·86-0·99, p<0·05) and experience of complication (OR=0·94, 95% CI: 0·88-1·00,p<0·10). These results suggest that the round variable and the interaction term of IMFHL by round are exerting a subduing effect on the association of health status variables with the outcome in this analysis. As health behaviours, such as BPCR, may also be in uenced by wider social, demographic and economic factors, in model IV tested the effect of these confounders that are described next.

Effect of socio-demographic, economic and area-level variables
Model IV results revealed that the impact of program exposure on birth and complication readiness was unaffected by the adjustment for socio-demographic variables. A wealth quintile variable was constructed based on household asset ownership for this study and included in the regression to describe the economic status of the household beyond the use of the BPL card variable. Results from the wealth quintile showed that the likelihood of BPCR practice reduced as the level of household poverty increased from marginally poor to the poorest households, with the poorest households in the fth quintile, 23 per cent or 0·87 times less likely to practice birth preparedness and complication readiness compared to women in marginally poor households, or rst quintile. Maternal education which has previously been found in the literature to be strongly correlated with routine maternal health behaviours, in contrast, was negatively associated with BPCR practice in this analysis. Mothers who had completed primary education, and more are seven per cent less likely to practice BPCR compared to mothers without any schooling (OR=0·93, 95% CI: 0·87-0·99, p<0·05). The increasing level of husband's education was, however, found to be positively associated with BPCR practice, but the results were not statistically signi cant (OR=1·02, 95% CI:0·95-1·11, p>0·10).

Discussion
This study makes a signi cant contribution to the literature in investigating the main effect of the IMFHL program membership on birth preparedness and complication readiness (BPCR) practice among women who have recently given birth in rural UP. The study provides evidence of the bene ts of SHG membership alone and when integrated with health literacy on improving BPCR practice. This evidence suggests that care-seeking is potentially improved during maternal complications among SHG members, and supports the claim that micro nance based women's groups are an effective strategy to provide health literacy to marginalised rural populations for enabling behaviour change. No previous study in the Indian context or elsewhere has examined BPCR in the context of a SHG program alone or in an integrated SHG and health literacy program. This is the rst study to our knowledge that demonstrates the presence of a diffusion effect of health behaviour, BPCR practice, from SHG members receiving health literacy to non-members in program villages. Overall, the results reveal that SHGs promote the adoption of BPCR practice among members and the odds of BPCR adoption almost doubles when a health literacy component is integrated into the model. However, the diffusion effect of BPCR practice is only observed when SHGs have an integrated health literacy component. Therefore, these ndings suggest that self-help group requires the addition of a maternal health literacy component to achieve a higher coverage of a behavioural health intervention among members and non-members in villages. The underlying mechanism of diffusion requires additional research to explain pathways; however, insights from previous studies suggest that either SHG members may themselves be promoting practices as 'model adopters' of new behaviour through interpersonal contact, or that SHGs serve as a better platform for facilitating linkages between non-members and community health workers when added with health literacy intervention (66,68,69). It is therefore feasible that when SHG members receive health literacy, they may be acting as change-agents to encourage other nonmembers to adopt desired health behaviours.
Prior, studies in Uttar Pradesh have shown no association of SHG membership alone with routine maternal health practices in pregnancy such as antenatal care, institutional delivery. However, our study, in contrast, found an improvement in BPCR practice for women in SHGs (63,68,69). Our ndings are supported by other studies from Uttar Pradesh and Bihar, also a state with low development indicators, that showed an increase in routine indicators of maternal health system utilisation( such as four ANC, institutional delivery etcetera) and home-based neonatal health care practices by 5-11 percentage points when health literacy intervention was provided through SHGs compared to membership in SHG alone (58,59). Our study showed that the peer effect of membership in SHG alone is associated with a 17 per cent higher likelihood of practising BPCR, which increases to 48 per cent when a health literacy component is delivered through it. The higher likelihood of BPCR practise among women in SHG's alone is explainable as a direct consequence of SHG membership which grants women with access to funds needed. This access ful ls one of the key steps for BPCR with regards to saving funds for a maternal emergency. Despite higher institutional deliveries in rural India which are nancially incentivised for women and health workers, studies in India have shown that incentivisation meets only 25 per cent of total maternal expense in rural areas (70). Studies from rural India showed that marginalised families are more likely to incur higher out of pocket expenditure (OOPE) ranging from USD 11-16 (INR 700-970) depending on whether the delivery is normal or complicated. Such expenses are considerable for populations analysed in this paper where the average monthly household income is USD 25. High maternal health expenses are counted as a strong factor for pushing vulnerable families at the risk of poverty into debt traps which reinforce cycles of debt borrowing and illness. Therefore, in such settings, the provision of health literacy and nancial access is expected to improve households' care-seeking preferences. Our results are further supported by previous studies that have shown that SHGs provide mutual support among members (59,67). Therefore, health literacy delivered through the peer network of SHG is likely to change health practices among marginalised women. In the IMFHL project, pregnant women in intervention villages received personalised health messages through house visits by SHG health volunteers, and through invitations to SHG meetings where discussions were set in community settings that allowed addressing community norms about health behaviours. Previously, a systematic review reported that combining home visits and community group meetings is an effective strategy to improve maternal and neonatal health outcomes (71). Yet scarce evidence exists that describe health outcomes when community programs combine micro nance provision and maternal health literacy, such as this program.
Our study also showed that adjusting for other socio-economic confounders BPCR was only associated with the poorest households in the fth wealth quintile, with these households least likely to practice BPCR. Moreover, the study showed that increasing years of education for neither the woman nor her husband were statistically associated with BPCR practice compared to those who had no education. This nding differs from another study in India that showed signi cant associations between BPCR practices in an urban slum and both maternal literacy and availing antenatal services (15). While the rural demographics of our study differ from the urban slum study environment, women included in the IMFHL program share some commonalities with the urban slum, since they represent a population with higher social marginalisation, higher mortality and low literacy, from areas of Uttar Pradesh where the quality of health services was lacking.

Study Strengths and Limitations
The key strengths of the study include: the community program strategy that combined micro nance and health literacy components; the use of a large sample size of 17,244 women representing marginalised rural populations; the quasi-experimental study design with pre-and postintervention measurements that is possibly the most e cient approach in a highly populated developing country's context, providing the opportunity to evaluate the incremental change across program exposure levels (impact of SHG only, SHG plus health intervention), and the potential diffusion effect (impact of SHG plus health intervention from members to non-members in areas without any program exposure).
However, our ndings need to be considered in the context of some limitations. First, the selection of program blocks under the IMFHL was based on the program's operational criterion, therefore, preventing randomisation. The quasi-experimental survey design, however, adopted a multistage sampling using different criterion to select blocks (SC/ST percentage) and villages (SHG coverage and population size) to limit selection bias and allow for comparisons of areas with similar socio-cultural and economic characteristics. Second, the study's crosssectional survey design prevents concluding causality.
A third limitation relates to the survey data used for the study, which was collected from women based on self-recall, which is susceptible to recall bias and social desirability. However, the potential for these errors is minimised as the women were interviewed within 12 months postdelivery and by trained interviewers who conducted the interviews. Finally, while the survey design selected women from SHG member and non-member households from the same villages, in some cases, the women sampled in a given village would be either all members or all nonmembers, which potentially limited the variation in the outcome variable within villages. This study therefore assessed the program impact on outcome across villages instead of within villages.

Policy and program implications
A programmatic strength of the IMFHL intervention was that it utilised the existing structure of the SHG program to embed the health literacy component, which provided evidence of a scalable model for disseminating health information. The ndings reported in this study are relevant for the rural Indian context where the Indian government is establishing a network of women-only SHGs as a national poverty alleviation program. Therefore, allowing the integration of the health intervention into SHG structures as done for the IMFHL program permits improving maternal health outcomes for the state and the country.
Another strength of the study, in terms of policy implications, is the design of the health outcome to re ect the complete practice of BPCR as opposed to limited BPCR steps re ecting partial practice used other studies. Complete BPCR allows inferring wider protection of care-seeking steps for women experiencing maternal complications and living in disadvantaged households.
Lastly, as health care-seeking behaviours are shaped by cultural practices in the community settings, the diffusion effect of health messages from members to non-members can reinforce desired health behaviours. Diffusion also increases program coverage of harder to reach populations. Future research is required to explore the pathways of diffusion effect of knowledge and practice from members to nonmembers in program villages.

Conclusions
The integrated micro nance and health literacy (IMFHL) program evaluated in this study promoted behaviour changes amongst some of the most marginalised households in India's most populated and a developmentally disadvantaged state, Uttar Pradesh. The study evaluation focussed rst on the IMFHL program's main effect on BPCR practice, independent of potential confounders. The evaluation then adjusted for potential confounders and separately assessed the association of multiple confounders on household decision making for BPCR adoption. The results show that SHGs exert both a dissemination effect of planned health behaviour within members, as well as a diffusion effect of the natural transfer of BPCR from members to non-members but only when a health literacy component is added to SHGs. Based on the study ndings, the authors would recommend integrating health literacy components into SHG networks and similar community groups to promote household's adoption of complication readiness plans to reduce delays in seeking health care during maternal complications.
The inclusion of health literacy in a micro nance platform presents an example of a novel community-based strategy that has the potential to interrupt the cycle of household poverty and low maternal healthcare-seeking. Lessons from the IMFHL program implementation in India provide the opportunity to adapt SHG models implemented elsewhere by adding health literacy components. Beyond India, micro nance is promoted globally as a poverty alleviation program in low resource settings where the burden of maternal mortality remains high. This can assist other low resource settings in accelerating progress towards the maternal health goal of the Sustainable Development Goals Declarations Ethics approval and consent to participate Ethics approval was granted by the Human Research Ethics Committee, at the University of Canberra. Previously, an independent ethics committee of the evaluation partner approved the primary data collection under the IMFHL project.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests
The author(s) declare(s) that they have no competing interests.

Funding
The paper is part of the PhD research of the rst author (Danish Ahmad) who is supported by the University of Canberra's Higher Degree by Research scholarship for his PhD. Open access funding for the publication is provided by the Bill and Melinda Gates Foundation who also funded the IMFHL program evaluated in the paper. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Author Contribution DA conceptualised the study, conducted the analysis, interpreted the results and drafted the study.IM guided the study design, the data analysis, the interpretation of the results and discussion. IM provided substantive feedback and reviewed all drafts of the paper. TN guided the study design, the data analysis and interpretation of results and discussion. TN provided substantive feedback and reviewed all drafts of the paper.

18.
Family  Note: Con dence intervals in parentheses; and signi cant p-value showing ***: p < 0·01, **: p < 0·05 and *: p < 0·10 levels. Log-likelihood and AIC/BIC values were also reported. Figure 1 Assumed pathway of knowledge diffusion from member households (tier I) to neighbouring non-member households (tier II) in the same villages Figure 2 Three-stage survey sampling approach comprising block, village and household selection