The Impact of i-PUSH on Maternal and Child Health Outcomes, Health Care Utilization and Financial Protection: A Cluster Randomised Controlled Trial Based on Financial and Health Diaries Data

Background : Universal Health Coverage (UHC) ensures access to quality health services for all, with no financial hardship when accessing the needed services. Nevertheless, access to quality health services is marred by substantial resource shortages creating service delivery gaps in low-and middle-income countries (LMICs), including Kenya. The Innovative Partnership for Universal Sustainable Healthcare ( i -PUSH) program, developed by AMREF Health Africa and PharmAccess Foundation (PAF), aims to empower low-income women of reproductive age and their families through innovative digital tools. This study aims to evaluate the impact of i -PUSH on maternal and child health care utilization, women’s health including their knowledge, behavior and uptake of respective services, as well as women’s empowerment and financial protection. It also aims to evaluate the impact of the LEAP training tool on empowering and enhancing CHVs’ health literacy and to evaluate the impact of the M-TIBA health wallet on savings for health and health insurance uptake. Methods: This is a cluster randomised controlled trial (RCT) study that uses a four-pronged approach – including year-long weekly financial and health diaries interviews, baseline and endline surveys, a qualitative study and behavioral lab-in-the-field experiments – in Kakemega County, Kenya. In total, 240 households from 24 villages in Kakamega will be followed to capture their health, health knowledge, health-seeking behavior, health expenditures and enrolment in health insurance over time. A random half of the households live in villages assigned to the treatment group where i -PUSH will be implemented after the baseline, while the other half of the households live in control village where i -PUSH will not be implemented until after the endline. The study protocol was reviewed and approved by the AMREF Ethical and Scientific Review Board (AMREF-ESRC). Research permits were obtained from the National Commission for Science, Technology and Innovation (NACOSTI) agency of Kenya. Discussion: health care through the mobile-phone based “health wallet” , it enhances enrolment in subsidized health insurance through the mobile platform – M-TIBA – developed by PAF, and it seeks to improve health knowledge and behavior through Community Health Volunteers (CHVs) who are trained using the LEAP tool –AMREF’s mHealth platform. The findings will inform stakeholders to formulate better strategies to ensure access to UHC in general, and for a highly vulnerable segment of the population in particular, including low-income mothers and their children.


Background
There has been a renewed international commitment to Universal Health Coverage (UHC) aiming at ensuring all people have access to the health services they need without suffering from financial hardships. However, there are sustained resource shortages and service delivery gaps in many countries that prevent them from meeting the Sustainable Development Goal (SDG) 3.8 related to UHC (i.e., achieve universal health coverage, including financial risk protection, access to quality essential health-care services and access to safe, effective, quality and affordable essential medicines and vaccines for all). Achieving UHC is particularly important in achieving SGDs goals related to maternal and child health (i.e., reducing the global maternal mortality ratio to less than 70 per 100 000 live births (SGD 3.1) and ending preventable deaths of newborns and children under 5 years of age (SGD 3.2). All countries are committed to reduce neonatal mortality to at least as low as 12 per 1000 live births and under-5 mortality to at least as low as 25 per 1000 live births) by 2030. Evidence still shows that more than half of the world's population lacks access to health care of sufficient quality [1] and about 100 million people fall into extreme poverty each year due to ill-health [2], particularly in lowand middle-income countries. To translate commitment to UHC into a reality, still need to undertake systemic reforms that require strong management and organization.
Despite being classified as a middle-income country in 2014, Kenya still remains among the 25% poorest countries strongly affected by social and health inequality in the world [1,2].
Inequalities in access to healthcare, in particular maternal and child care are still rampant despite major improvements made through targeted policies over the past few years [4]. The poorest mothers are still far behind in terms of coverage of essential reproductive and maternal and child health services. The Government of Kenya has included UHC as one of its 'Big Four Agenda'-action points, which is anticipated to lead the transformation of the country by 2022 [3]. The objective is to achieve a 100% cost subsidy for essential health services and to reduce out-of-pocket health expenditures by half. Low-cost health insurance schemes, eHealth and mobile health (mHealth) services are among other opportunities to achieve this goal. Consequently, the non-governmental organisations AMREF Health Africa and PharmAccess Foundation (PAF) are both supporting the Government to achieve the goal of UHC in many counties. This includes Kakamega-a county in Western Kenya-where AMREF and PAF are jointly implementing their Innovative Partnership for Universal Sustainable Healthcare (i-PUSH) program.
The i-PUSHprogram utilizes innovative digital tools developed by both partners to enhance access to affordable and quality health care to low-income women of reproductive age (WRA) and their families. Through the i-PUSH, women receive the National Health Insurance Fund (NHIF) SupaCover at subsidized premiums on their mobile phone, using the PAF's so-called "health wallet". The "health wallet" runs on the digital platform M-TIBA, which registers health care utilization at participating clinics, connecting patients, providers and clinics on one platform. The PAF is supporting the implementation of UHC by registering households using M-TIBA and developing a socio-economic mapping of the population. One such tool, Connected Diagnostics for malaria, was recently piloted in Kisumu County (Kenya). Another tool, designed to ease communication between doctors and women during pregnancy and postnatal period, has been piloted in Nairobi and there are plans to scale up in other counties.
Community Health Volunteers (CHVs) are the first point-of-contact for women in the program.
They make use of AMREF's Mjali (Mobile Jamii Afya Link) tool for digital registration of household information and the mobile phone-based LEAP tool for training. The LEAP tool employs a mobile learning approach to train and empower CHVs using their mobile devices operating from any phone [5]. This enables the CHVs to learn at their own pace, and with their own mobile devices while in the community, providing both interpersonal and community aspects of learning.
In order to maximize the effectiveness of this program, it is important for stakeholders to have a deep understanding of current access to health services, households' health-related decision-

Study design and randomization procedure
The survey design is a longitudinal cluster randomized controlled trial (RCT). Randomization occurs at the level of villages in Khwisero. The "treatment" and "control" groups are constructed, comparing villages where i-PUSH will be rolled out after the baseline with village where i-PUSH will not be rolled out until after the endline.
The research team used community-level socio-demographic and infrastructure indicators from baseline data to form pairs of similar villages and determine the exact matching indicators. In keeping with robustness of the cluster RCT, the procedure hence followed four steps for matching of the treatment" and "control" villages: (i) purposive selection of the Sub-county (Khwisero) where the intervention will roll out; (ii) random selection of 24 villages; (iii) pairmatching of villages based on relevant background characteristics and outcomes of interest; and (iv) randomization of treatment and control villages within each pair of the villages by flipping a coin. Pairing villages before randomization reduces the risk of a bad draw during the randomization process. Randomisation without pairing will, in expectation, also lead to similar control and treatment groups, but it is also possible that the random draw produces a control and treatment group with very different characteristics by chance [6]. This risk is reduced through pairing. We used the Euclidean distance for our matching process, which corresponds to the absolute difference between the standardized values of all of the covariates for a possible pair of matches. We conducted the matching within each of the four health clinic catchment areas. Thus, each village was matched with one of the other five villages in the vicinity of the health clinic. This was done to ensure that each health clinic had an equal number of treatment and control villages in their catchment area. Hence we computed this distance measure between each village and all other villages within the same health clinic catchment area; "pair" the two villages with the minimum distance and remove them from the list; repeat the distance calculation excluding the pair made; and continue until all villages were paired.
After the matching process, the randomization assignment was carried out in the presence of key stakeholders including PAF, local liaison persons and village representatives, upon explaining all steps. Consent for the procedures was obtained from local government officials before the random assignment. The following steps were followed: papers with paired village names were folded and put in a bag; and two village representatives from each paired village discussed on whom to pick the paper and after the other group members verified that the names could not be seen, one paper was picked. A Kenya Shilling 10 coin was used to decide which group the picked village belonged to by flipping the coin. The village representatives had decided that the head of the coin should represent the control group, justifying that Kenyatta (1 st president of Kenya) was a "controlling village", and the shield to represent the intervention group. The process of choosing the folded paper and flipping of the coin was repeated for all paired villages.
The treatment group thus consists of the target population living in the randomly assigned 12 villages. i-PUSH roll-out in the treatment villages includes training of their CHVs with the LEAP tool, who will subsequently introduce the health wallet to eligible women living in the treatment villages, and offer them the subsidized insurance scheme on their mobile phone. The CHVs working in the control villages (as well as the remaining non-sampled villages on the longlist) will not receive training on the LEAP tool yet, nor will women in the control villages be offered the health wallet and subsidized insurance on their mobile phone. They constitute our comparison group. We randomized at the village level because 1) villages are served by one CHV each, who are either trained or not trained on LEAP (hence, the LEAP intervention cannot be varied within villages); 2) to avoid contamination between households within the same villages regarding health-related knowledge and behavior; and 3) moreover, it was deemed politically unfeasible to offer the health wallet and subsidized health insurance to some eligible households in a village but not to other eligible households in that same village. Upon on roll out of the subsidized services, eligible households are encouraged to use the services, though they are given the right to opt out at any time.

Study population
The study population consists of eligible households living in the selected study villages.
Eligible households included those with at least one woman of reproductive age (WRA) (18-49) who: a) had at least one child below 4 years living with her at baseline; or b) was pregnant at baseline. It hypothesized a confidence interval of 95%, a margin-of-error of 5% and a power of 80%.
The cluster size per arm, total women per clusters was therefore 10 and 12 women, respectively. Hence: , with ES (effect-size) the ratio between the mean difference and pooled standard deviation, Z 1− α 2 and Z 1−β are the values from the normal distribution holding 1-α 2 ,1-β below it, respectively. α and β are levels of significance and power, respectively. Thus, By fixing the clusters per arms to k=12, the required sample size per arms is given by the following formula: In the formula above, a feasibility check must be done so that k > nρ. This condition is satisfied If we assume an attrition rate of 10%, then: To keep sample size at par, households that dropped out of the study before the start of the intervention were replaced with new eligible households on a rolling basis for a maximum period of six months, or until program start.

Description of the i-PUSH program
i-PUSH is a comprehensive intervention that ultimately aims to improve the utilization of access to RMNCH services as well as improving the quality of care of RMNCH services. In the original i-PUSH program that is the focus of this evaluation, households receive the first year of health insurance premium for free, and they are stimulated and supported to save for a 50% co-payment in the second year and a 100% premium payment thereafter. The free provision in year one is expected to show the benefits of insurance to the selected households.
The 50% co-payment and the support for savings in the second year is expected to install a habit of savings. 1 This evaluation study will focus on the first two of these three spheres of interest: knowledge and (financial) access; the third sphere of interest-quality upgrades at the health facilities, cannot be evaluated with our study design because all i-PUSH clinics in our study area were already upgraded at baseline. A rough sketch to the implementation of enrolment, intervention and assessment of the programme is indicated in Annex 1. and behavior on pre-specified topics. We will also assess whether CHVs' time spent on LEAP, number of training modules completed, and scores on the LEAP quizzes predict impact on women and men in the communities.
Subsidized access to NHIF SupaCover health insurance through the M-TIBA health wallet The improved knowledge on health and health financing of women and men is also expected to translate into improved attitudes towards insurance and saving for health and insurance. To support these changes in knowledge and attitudes, WRA will receive the first year of their NHIF insurance premium at 100% subsidy and the second year at a 50% subsidy. The subsidies are expected to enhance initial enrolment in NHIF such that enrollees can experience first-hand the benefits of insurance. Moreover, this will allow women to be acquainted with regular savings for the insurance premium for the next year, such that they will get into the habit of recurrent savings and increasingly be able to frequently set aside small amounts of money.

Expected outcomes
Overall, the outcomes of interest for this evaluation study include the following:

Healthcare utilization
• Healthcare utilization for curative care (conditional on illness and/or injury): • Healthcare utilization for preventive care (immunization, growth monitoringchildren below 5 only).
• Continuum of maternal care (perinatal care -pregnant women only): • Proportion of women who attended at least four times antenatal care during pregnancy.
• Proportion of women given folic acid/iron supplementation during pregnancy.
• Proportion of births attended by skilled birth attendant.
• Proportion of postnatal care visits after delivery.
• Proportion of common mental illnesses in adults.
• Proportion of other illnesses (infectious, non-communicable diseases, etc.). The study will provide additional insights into the following digital tools that are currently being rolled-out by AMREF or PAF:

Data collection instruments and techniques
Data collection consists of four main components: 1) a qualitative baseline study, 2) baseline and endline household surveys, 3) weekly financial and health diaries interviews with all adults and emancipated minors in the households, and 4) behavioral lab-in-the-field experiments.
Both quantitative and qualitative tools were piloted in Nairobi slums and debriefed to the research team including the field team.

Qualitative baseline study
The study uses qualitative data collection methods to get a deeper understanding of the perceptions and behaviors of the population on health insurance and health care utilization, to feed into the quantitative instrument design and to complement findings from the quantitative surveys. This will help understand the motivations, drivers and obstacles to savings for insurance in the target population. The qualitative baseline study is based on in-depth interviews (IDIs, n = 20) and focus group discussions (FGD, n = 4) with different stakeholders who were purposively sampled after CHVs' mobilization to willingly participate in the study.
Another qualitative study may be conducted after the quantitative endline survey has been completed to provide additional under-the-skin description and/or for improved interpretation if the impact evaluation yields unexpected results.

Baseline and endline survey instruments
The quantitative evaluation starts with a baseline survey before the rollout of the i-PUSH  (Table 1). Data are collected through digital tools that allow to analyze data as they are recorded, and improve data collection tools mid-course of the process. The short recall period drastically reduces recall bias and ensures that both major and minor illness episodes, including those with foregone care are reported. Because interviewers visit households weekly, they build a relationship of trust, which enables the diaries to capture also more sensitive health events.
This is of particular importance in relation to capturing maternal and child health experiences.
Moreover, in contrast to what is common in standard RCT, the shorter research cycles, and report results on a regular basis to allow program managers to learn and take action for continuous improvement.
Qualified fieldworkers were recruited and trained on data collection tools and techniques.
Diaries data are collected through personal interviews in a conversation-like manner.

Data management and analysis
Trained team leaders are situated in the field to supervise real time data collection. Data quality is ensured by regular spot checks and sit-ins to approximately 5-10% of each fieldworker's daily work to verify authenticity of the data collected. Data collection is done electronically using tablets/phones, with spot checks for quality control. The field supervisors certify the quality of the data through editing the data before they are transferred to the database. Once the data collection is completed and synchronized in centrally located database, all inconsistencies are resolved prior to data analysis. An automated routine to check on the data completeness, correctness and consistency runs on 100% of the collected data. A discrepancy report is generated to help in following up on any inconsistencies or errors in the data with the responsible interviewer. Similarly, the quality of qualitative data will be ensured through recruitment and training of qualified field interviewers with experience in qualitative data collection. A qualified transcriber will transcribe the interviews verbatim and double coding of about 10% of the transcripts will also be done to ensure consistency in the application of the codes. Access to data is granted for all research teams in respective research organizations.
Data auditing is carried out by the research team on a progressive basis on weekly basis throughout the study duration.
Quantitative data will be analyzed using Stata version 14 statistical software. The first set of analysis will consist of descriptive statistics and will summarize and compare using measures of central tendency and dispersion (mean (SD), range and median). This will allow us to detect similarities and/or differences between participants' characteristics across the different subgroups. In other words, we will compare some baseline measurements between the control and intervention groups using t-test adjusted for clustering at the village unit for continuous variables, and cluster-adjusted chi-square for binary variables. We will first check whether the outcomes and covariates in the control group and treatment group are comparable at baseline.
The second set of analysis will consist assessing the causal effect of the i-PUSH intervention via an analysis of covariance (ANCOVA) based on intention-to-treat analysis (all respondents who are randomized will be included in the statistical analysis and will be analyzed according to the group they were initially/originally assigned). Based on our research questions, we will explore several econometric models as outlined in Appendix 2. Interim data analyses are done using baseline and weekly diaries data.

Risks and measures to minimize them
As the information collected will be on health service delivery, financial and insurance information, we do not anticipate any risks to participants, nevertheless. We will aim to minimize the risks by being as forthcoming as possible on the project description and the ethical process. In the case of minimal discomfort as a result of personal and sensitive questions, the research team will endeavor to ensure that the participant is given ample time to compose themselves, reassure them of confidentiality and ability to stop the interview if they are not able to continue with the interview. Ancillary and post-trial care may not be indicated as this is a community trial that does not involve any significant harm to the participating households.

Dissemination of findings
Scientific dissemination through peer-reviewed publications and implementing partners' it is still two-fold higher than the SDG target [11]. These figures are disproportionately higher in Western Kenya, such as in Kakamega.
Improving access to health care throughout pregnancy, childbirth and during childhood is key in improving maternal and child health. Experience over the past decade has shown that building capacities of individuals, families and communities to ensure appropriate self-care, prevention and care-seeking behavior improves maternal and child health outcomes [12].
However, this is more difficult to achieve in poor populations who have worse health outcomes than the non-poor. Barriers such as costs of care, lack of information and cultural beliefs impede access to health care among poor communities.
Since its independence in 1963, the government of Kenya has initiated policy, reforms and strategies towards UHC for all, including those in vulnerable situations such as low-income mothers and children. In 1998, the NHIF act was amended to enhance coverage among the poor, accelerate coverage of the informal sector, and enhance the benefit package [13,14]. The most recent reform along the same line was the introduction of free maternal health services in 2013 that included abolition of user fees at primary health care facilities [14]. Despite these positive steps, Kenya's implementation of UHC has been riddled with myriads of challenges including poor quality of care, utilization and catastrophic spending by households especially the poor and other vulnerable groups [16]. Overall, health insurance coverage in Kenya has only increased from 8% to 20% between 2009 and 2014, and those from wealthy households were 12 times more likely to have insurance compared to those in poor households. Similarly, those in the informal employment and rural settings were less likely to be insured [11].
The i-PUSH program thus has been initiated to accelerate the expansion of health insurance coverage among the low-income population, WRA and their family members, using innovative digital tools. This research investigates the reasons for low access to health services, in particular related to maternal and child services, where the problems are persistent. It also investigates in-depth to what extent costs of services hinder access, and whether expanding UHC through the i-PUSH program is an effective strategy to increase access.
Information generated from this study will be instrumental in improved implementation of policies supporting the roll-out of UHC. This research will also provide valuable information on UHC policies for the academic community, in particular, because we use high-frequency data (diaries) and analyze digital technologies that can support UHC.
To achieve these objectives, the study assumes the following: (i) no contamination across treatment and control groups; this is enhanced by a focus on villages and CHVs across villages, instead of individual women; (ii) County (Sub-county), AMREF and PAF will stick to the randomization plan that was independently developed by the research team but in agreement with stakeholders, and a memorandum of understanding (MoU) was signed among the partners; and (iii) Government does not unexpectedly and drastically alter its UHC policy plans in Kakamega (i.e., the government does not suddenly decide to offer free public care in Kakamega County, because that will take away many of the benefits of i-PUSH). Although there is little the research team can do in that case, this is the reason a flexible, high-frequency diaries design instead of standard (less flexible) RCT design is chosen. That is, there are more data points and if policies change half way, it is possible to actually capture this in the data.
This project is not free from limitations. One of the main limitations of the study is related to the exclusion of the following components in the impact evaluation, though they are an integral part of i-PUSH program. These include four components: (i) Capacity-building at the regional level: i-PUSH invests in improving the capacity of community-based organizations to increase community-wide dialogue on RMNCH, (ii) Health facility quality upgrades: the improved capacity of healthcare providers (implemented through SafeCare) to deliver RMNCH services is expected to lead to improved standards of care, improved quality of services and enhanced client satisfaction.
(iii) M-JALI household registration tool: AMREF has developed a digital tool to increase the capacity of CHVs to keep a census of the households in their target area, and collect household survey data on a rolling basis that can be used for the systematic reporting of community-level data, thereby potentially enhancing health-related decision-making and resource allocation at all policy levels.
(iv) M-TIBA digital health platform: PAF aims to increase the capacity of healthcare workers on learning, data capturing and reporting through the M-TIBA platform that allows amongst others for digital recording of health visits, diagnoses, treatments and services as well as payments/billing between patients, NHIF, Ministry of Health and providers.
These components are left out of the evaluation because our study focuses on the demandside (target population), and the described components are implemented either at the community-level or at the facility-level.

Conclusions
This study aims to evaluate the impact of i-PUSH program on maternal and child health care utilisation, women's health including their knowledge, behaviour and uptake of respective services, as well as women's empowerment and financial protection. It also aims to evaluate the impact of the LEAP training tool on empowering and enhancing CHVs' health literacy, and to evaluate the impact of the M-TIBA health wallet on savings for health and health insurance uptake. The findings will inform stakeholders to formulate better strategies to ensure access to UHC in general, and for those highly vulnerable segments of the population in particular. The findings of this research will provide valuable information on UHC policies for the academic community, policy-makers and other stakeholders to support the achievement of SDGs.

Trail status:
Protocol was registered in NIH -ClinicalTrials.org with registration number: AfricanPHRC; Trial ID: is the ITT effect or the impact of the i-PUSH program. For continuous outcomes, we will use an OLS model with standard errors clustered at the level of the community units. For binary outcomes, we will use a logit model. is the error term.
In addition to pretest measures, all other baseline covariates such as age, education, socioeconomic status (wealth quintile), crowding (persons per room), occupation, etc. will also be included in Equation (1). Thus, in Equation (1) the treatment effect of the I-PSUH program 2 assesses the treatment difference on post-treatment outcome adjusted for baseline.

Impact of the i-PSUH on women empowerment
We will assess whether the i-PUSH will significantly contribute in empowering women. For self-reported surveys on women empowerment, we will construct a total score 2 of women empowerment which is the sum of each domain when the woman indicates that she has sole or joint decision-making power within the household (for each domain, the decision-making binary indicator will be equal to 1 if the woman respondent makes the decision alone or jointly with her partner and 0 otherwise). This total score will therefore be used as the dependent variable and Equation (1) above will be used to estimate the impact of i-PUSH program on women empowerment. Furthermore, with regards to the field experiment, we will follow Almås et al. (2018). In other words, we will still use the same econometric model (Equation 1) after constructing the dependent variable emanating from the field experiment. This dependent variable will be the willingness-to-pay 3 which is the share that the woman is willing to pay when the experiment stops.