Evaluating the impact of a maternal health voucher programme on service use before and after the introduction of free maternity services in Kenya: A quasi-experimental study

BMJ Global Health, Volume 3, No. 2, Year 2018

Interprétation

Introduction From 2006 to 2016, the Government of Kenya implemented a reproductive health voucher programme in select counties, providing poor women subsidised access to public and private sector care. In June 2013, the government introduced a policy calling for free maternity services to be provided in all public facilities. The concurrent implementation of these interventions presents an opportunity to provide new insights into how users adapt to a changing health financing and service provision landscape. Methods We used data from three cross-sectional surveys to assess changes over time in use of 4+ antenatal care visits, facility delivery, postnatal care and maternal healthcare across the continuum among a sample of predominantly poor women in six counties. We conducted a difference-in-differences analysis to estimate the impact of the voucher programme on these outcomes, and whether programme impact changed after free maternity services were introduced. results Between the preintervention/roll-out phase and full implementation, the voucher programme was associated with a 5.5% greater absolute increase in use of facility delivery and substantial increases in use of the private sector for all services. After free maternity services were introduced, the voucher programme was associated with a 5.7% higher absolute increase in use of the recommended package of maternal health services; however, disparities in access to facility births between voucher and comparison counties declined. Increased use of private sector services by women in voucher counties accounts for their greater access to care across the continuum. Conclusions Our findings show that the voucher programme is associated with a modest increase in women’s use of the full continuum of maternal health services at the recommended timings after free maternity services were introduced. The greater use of private sector services in voucher counties also suggests that there is need to expand women’s access to acceptable and affordable providers. A quasi-experimental study was conducted with repeated cross-sectional surveys administered in May 2010–July 2011, July–October 2012 and July–August 2016. Data were collected in four intervention counties (Kiambu, Kilifi, Kisumu and Kitui) and three comparison counties (Makueni, Nyandarua and Uasin Gishu) selected to match the geographical, population and health facility characteristics (type of facility and ownership) of the intervention counties. To facilitate comparisons over time, one intervention county (Kilifi) was excluded from this analysis, as it was not surveyed in 2016. We included a map of the study counties in online supplementary appendix 1). bmjgh-2018-000726supp001.pdf The study used a multistage sampling design. In the first stage, a random sample of 14 sublocations were selected within each intervention county from those located within a 5 km radius of a facility accredited in the voucher programme. In comparison counties, 14 sublocations were selected among those within a 5 km radius of a facility that were comparable to the intervention facilities in terms of facility type and ownership. This was done to ensure that all surveyed women had similar physical access to the maternal health services offered under the voucher programme. At the second sampling stage, three villages were randomly selected within each sublocation. Given that the voucher programme intended to target poor women, the poorest households in each village were identified by local administrators and purposively selected for inclusion in the study. Within each household, women aged 15 to 49 years with at least one birth in the past 12 months or pregnant at the time of the interview were targeted for participation. In households with more than one woman meeting the target characteristics, the youngest woman was selected into the study. Additional details of the study protocol and sampling methods have been described previously.16–18 20 Face-to-face interviews were conducted during each survey round using a tablet-based structured questionnaire covering a range of topics including women’s sociodemographic characteristics, reproductive history and maternal health service utilisation. Each participant provided written informed consent to participate in the study. Table 1 defines the 10 indicators of maternal health service utilisation and sector of care examined in this study. In addition to examining use of individual services in each period, we also looked at the proportion of women receiving a complete package of all three services across the maternal health service continuum (complete care). We also estimated the proportion receiving complete care at the recommended timings, with the first ANC visit occurring during the first trimester and the PNC check occurring within 48 hours of delivery (recommended care). Indicator definitions ANC, antenatal care; PNC, postnatal care. Respondents were asked to report on all of their births within the 5 years prior to the survey; data from the three cross-sectional surveys were pooled and reshaped to allow us to perform analyses on all reported births. We categorised these births into three periods according to when they occurred. Period 1 (May 2005–December 2009) refers to the pre-intervention and roll-out phase of the programme. Period 2 (January 2010–May 2013) refers to the post roll-out phase, when the programme was implemented at full intensity. Lastly, Period 3 (June 2013–August 2016) refers to the period when both the voucher programme and the free maternity services policy for all government facilities were being implemented simultaneously. For the data collected in 2016, a glitch in the survey programming resulted in 23% of women who reported giving birth at least once in their lifetime having a missing response to the question, ‘During the last 5 years, how many children have you given birth to?’ This question was missing for less than 1% of respondents in both the 2010 and 2012 surveys. Based on the skip pattern of the instrument, only women who reported giving birth to one or more child in the past 5 years were asked subsequent questions about the key outcomes of this study related to maternal health service utilisation for each child born within the period. Women who reported zero births or had missing information on their number of births in the past 5 years were not asked these questions; we are therefore missing outcome data for births that occurred within the past 5 years to women with missing information for the aforementioned question. We conducted analyses to explore for any evidence of systematic biases in our estimates relating to the pattern of missing data in the question about the number of live births 5 years prior to the survey (online supplementary appendix 1). We found that after controlling for all relevant sociodemographic characteristics, both marital status and county had strong effects on the odds of having missing data. The observed effect of county is due to the fact that the data manager identified the glitch during the course of fieldwork and corrected it; the proportion of missing data therefore declined after the instrument was updated (Table A2.1). The mechanism behind the effect of marital status is unclear and may be due to chance. These findings suggest that the data are not missing completely at random and might either be missing at random (MAR) conditional on both county and marital status or missing not at random. However, because we know that the missing data mechanism was due to a software issue that is unrelated to the underlying values of the our outcomes of interest, we have assumed the data to be MAR and have conducted a complete case analysis controlling for both county and marital status.21 22 Less than 1% of responses were missing for all other variables across all three surveys. bmjgh-2018-000726supp002.pdf We performed Wald tests to assess cross-sectional differences in background characteristics between all surveyed women in voucher and comparison counties for each period. We used logistic regression models, adjusted by background characteristics, to estimate cross-sectional differences in women’s maternal health service utilisation for births that occurred in voucher and comparison counties. Our analysis of women’s background characteristics used a logistic regression models adjusted for multistage clustering at the sublocation and village levels. Outcomes related to service utilisation additionally accounted for clustering at the mother level, as some women reported more than one live birth within the 5 years prior to the survey. We used a difference-in-differences approach with mixed-effects linear regression models to approximate the impact of the voucher programme and introduction of free maternity services on maternal health service utilisation and sector of care with random effects included for county sublocation, village and mother. To assess the impact of the voucher programme, we estimated differences in the change over time in outcomes between births that occurred in voucher and comparison counties before (Period 1) and after (Period 2) the voucher programme was fully implemented. We further assessed whether any benefits of the voucher programme persisted after free maternity services were introduced by estimating the difference in the change in outcomes between births in voucher and comparison counties before (Period 2) and after (Period 3) user fees were removed. We present these voucher programme impact results controlled for key potential confounders, including location (urban/rural), wealth quintile, year of childbirth, insurance enrolment, mother’s parity, education, marital status and employment status. We used STATA IC V.15.1 (StataCorp LLC) to conduct this analysis.

Résumé de l’IA

Study Justification:

– The study aimed to evaluate the impact of a maternal health voucher program and the introduction of free maternity services in Kenya.
– The concurrent implementation of these interventions provided an opportunity to understand how users adapt to changes in health financing and service provision.

Highlights:

– The study used data from three cross-sectional surveys conducted between 2010 and 2016.
– It focused on changes in the use of antenatal care visits, facility delivery, postnatal care, and maternal healthcare among predominantly poor women in six counties.
– The voucher program was associated with a 5.5% increase in facility delivery and increased use of private sector services.
– After the introduction of free maternity services, the voucher program was associated with a 5.7% increase in the use of recommended maternal health services.
– Disparities in access to facility births between voucher and comparison counties declined.
– The study highlights the need to expand women’s access to acceptable and affordable providers.

Recommendations:

– Expand the maternal health voucher program to increase women’s access to care across the continuum.
– Strengthen the implementation of free maternity services to ensure equitable access for all women.
– Address disparities in access to facility births between different counties.
– Improve access to acceptable and affordable providers for maternal health services.

Key Role Players:

– Government of Kenya: Responsible for implementing and expanding the maternal health voucher program and free maternity services.
– County Health Departments: Involved in the implementation and monitoring of the voucher program and free maternity services.
– Health Facilities: Provide the maternal health services covered by the voucher program and free maternity services.
– Non-Governmental Organizations (NGOs): Collaborate with the government to support the implementation and evaluation of maternal health programs.

Cost Items for Planning Recommendations:

– Voucher Program Expansion: Budget for increasing the coverage and reach of the maternal health voucher program, including subsidies for poor women.
– Free Maternity Services Implementation: Allocate funds for the provision of free maternity services in all public facilities, including staffing, equipment, and supplies.
– Training and Capacity Building: Invest in training healthcare providers to deliver quality maternal health services and ensure adherence to standards.
– Monitoring and Evaluation: Allocate resources for monitoring and evaluating the impact of the voucher program and free maternity services to inform future improvements.
– Communication and Awareness: Budget for public awareness campaigns to inform women about the availability and benefits of the voucher program and free maternity services.

Force de la preuve

The strength of evidence for this abstract is 7 out of 10.
The evidence in the abstract is moderately strong, but there are some areas for improvement. The study design is quasi-experimental, which provides some level of evidence, but it would be stronger if it were a randomized controlled trial. The study used data from three cross-sectional surveys, which allows for comparisons over time, but a longitudinal study design would provide stronger evidence. The study conducted a difference-in-differences analysis to estimate the impact of the voucher program, which is a suitable method, but it would be beneficial to include more details on the statistical analysis and potential confounders that were controlled for. To improve the strength of the evidence, future studies could consider using a randomized controlled trial design and collecting longitudinal data. Additionally, providing more details on the statistical analysis and potential confounders would enhance the robustness of the findings.

Abstrait

A quasi-experimental study was conducted with repeated cross-sectional surveys administered in May 2010–July 2011, July–October 2012 and July–August 2016. Data were collected in four intervention counties (Kiambu, Kilifi, Kisumu and Kitui) and three comparison counties (Makueni, Nyandarua and Uasin Gishu) selected to match the geographical, population and health facility characteristics (type of facility and ownership) of the intervention counties. To facilitate comparisons over time, one intervention county (Kilifi) was excluded from this analysis, as it was not surveyed in 2016. We included a map of the study counties in online supplementary appendix 1). bmjgh-2018-000726supp001.pdf The study used a multistage sampling design. In the first stage, a random sample of 14 sublocations were selected within each intervention county from those located within a 5 km radius of a facility accredited in the voucher programme. In comparison counties, 14 sublocations were selected among those within a 5 km radius of a facility that were comparable to the intervention facilities in terms of facility type and ownership. This was done to ensure that all surveyed women had similar physical access to the maternal health services offered under the voucher programme. At the second sampling stage, three villages were randomly selected within each sublocation. Given that the voucher programme intended to target poor women, the poorest households in each village were identified by local administrators and purposively selected for inclusion in the study. Within each household, women aged 15 to 49 years with at least one birth in the past 12 months or pregnant at the time of the interview were targeted for participation. In households with more than one woman meeting the target characteristics, the youngest woman was selected into the study. Additional details of the study protocol and sampling methods have been described previously.16–18 20 Face-to-face interviews were conducted during each survey round using a tablet-based structured questionnaire covering a range of topics including women’s sociodemographic characteristics, reproductive history and maternal health service utilisation. Each participant provided written informed consent to participate in the study. Table 1 defines the 10 indicators of maternal health service utilisation and sector of care examined in this study. In addition to examining use of individual services in each period, we also looked at the proportion of women receiving a complete package of all three services across the maternal health service continuum (complete care). We also estimated the proportion receiving complete care at the recommended timings, with the first ANC visit occurring during the first trimester and the PNC check occurring within 48 hours of delivery (recommended care). Indicator definitions ANC, antenatal care; PNC, postnatal care. Respondents were asked to report on all of their births within the 5 years prior to the survey; data from the three cross-sectional surveys were pooled and reshaped to allow us to perform analyses on all reported births. We categorised these births into three periods according to when they occurred. Period 1 (May 2005–December 2009) refers to the pre-intervention and roll-out phase of the programme. Period 2 (January 2010–May 2013) refers to the post roll-out phase, when the programme was implemented at full intensity. Lastly, Period 3 (June 2013–August 2016) refers to the period when both the voucher programme and the free maternity services policy for all government facilities were being implemented simultaneously. For the data collected in 2016, a glitch in the survey programming resulted in 23% of women who reported giving birth at least once in their lifetime having a missing response to the question, ‘During the last 5 years, how many children have you given birth to?’ This question was missing for less than 1% of respondents in both the 2010 and 2012 surveys. Based on the skip pattern of the instrument, only women who reported giving birth to one or more child in the past 5 years were asked subsequent questions about the key outcomes of this study related to maternal health service utilisation for each child born within the period. Women who reported zero births or had missing information on their number of births in the past 5 years were not asked these questions; we are therefore missing outcome data for births that occurred within the past 5 years to women with missing information for the aforementioned question. We conducted analyses to explore for any evidence of systematic biases in our estimates relating to the pattern of missing data in the question about the number of live births 5 years prior to the survey (online supplementary appendix 1). We found that after controlling for all relevant sociodemographic characteristics, both marital status and county had strong effects on the odds of having missing data. The observed effect of county is due to the fact that the data manager identified the glitch during the course of fieldwork and corrected it; the proportion of missing data therefore declined after the instrument was updated (Table A2.1). The mechanism behind the effect of marital status is unclear and may be due to chance. These findings suggest that the data are not missing completely at random and might either be missing at random (MAR) conditional on both county and marital status or missing not at random. However, because we know that the missing data mechanism was due to a software issue that is unrelated to the underlying values of the our outcomes of interest, we have assumed the data to be MAR and have conducted a complete case analysis controlling for both county and marital status.21 22 Less than 1% of responses were missing for all other variables across all three surveys. bmjgh-2018-000726supp002.pdf We performed Wald tests to assess cross-sectional differences in background characteristics between all surveyed women in voucher and comparison counties for each period. We used logistic regression models, adjusted by background characteristics, to estimate cross-sectional differences in women’s maternal health service utilisation for births that occurred in voucher and comparison counties. Our analysis of women’s background characteristics used a logistic regression models adjusted for multistage clustering at the sublocation and village levels. Outcomes related to service utilisation additionally accounted for clustering at the mother level, as some women reported more than one live birth within the 5 years prior to the survey. We used a difference-in-differences approach with mixed-effects linear regression models to approximate the impact of the voucher programme and introduction of free maternity services on maternal health service utilisation and sector of care with random effects included for county sublocation, village and mother. To assess the impact of the voucher programme, we estimated differences in the change over time in outcomes between births that occurred in voucher and comparison counties before (Period 1) and after (Period 2) the voucher programme was fully implemented. We further assessed whether any benefits of the voucher programme persisted after free maternity services were introduced by estimating the difference in the change in outcomes between births in voucher and comparison counties before (Period 2) and after (Period 3) user fees were removed. We present these voucher programme impact results controlled for key potential confounders, including location (urban/rural), wealth quintile, year of childbirth, insurance enrolment, mother’s parity, education, marital status and employment status. We used STATA IC V.15.1 (StataCorp LLC) to conduct this analysis.

Matériaux

N/A

Innovations

The recommendation to improve access to maternal health based on the study is to expand women’s access to acceptable and affordable providers. This can be done by increasing the number of healthcare facilities, improving the quality of care, and ensuring that services are affordable for all women, especially those from low-income backgrounds. Additionally, efforts should be made to address the disparities in access to facility births between different regions or counties.

AI Innovations Description

The recommendation that can be developed into an innovation to improve access to maternal health based on the study is to expand women’s access to acceptable and affordable providers. The study found that the maternal health voucher program was associated with a modest increase in women’s use of the full continuum of maternal health services after free maternity services were introduced. However, there were disparities in access to facility births between voucher and comparison counties. The study suggests that increasing access to private sector services in voucher counties contributed to greater access to care across the continuum. Therefore, to further improve access to maternal health, it is recommended to expand women’s access to acceptable and affordable providers, both in the public and private sectors. This can be done by increasing the number of healthcare facilities, improving the quality of care, and ensuring that services are affordable for all women, especially those from low-income backgrounds. Additionally, efforts should be made to address the disparities in access to facility births between different regions or counties.

AI Innovations Methodology

The methodology used in the study to simulate the impact of the main recommendations on improving access to maternal health involved a quasi-experimental design with repeated cross-sectional surveys. Here is a brief description of the methodology:

1. Study Design: The study used a quasi-experimental design, comparing intervention counties (Kiambu, Kilifi, Kisumu, and Kitui) with comparison counties (Makueni, Nyandarua, and Uasin Gishu). The counties were selected to match geographical, population, and health facility characteristics.

2. Sampling: A multistage sampling design was used. Sublocations were randomly selected within each county, and three villages were randomly selected within each sublocation. The poorest households in each village were purposively selected for inclusion in the study. Women aged 15 to 49 years with at least one birth in the past 12 months or pregnant at the time of the interview were targeted for participation.

3. Data Collection: Face-to-face interviews were conducted using a tablet-based structured questionnaire. The questionnaire covered topics such as women’s sociodemographic characteristics, reproductive history, and maternal health service utilization.

4. Data Analysis: The study analyzed changes over time in the use of antenatal care visits, facility delivery, postnatal care, and maternal healthcare across the continuum. A difference-in-differences analysis was conducted to estimate the impact of the voucher program on these outcomes, and whether the program’s impact changed after the introduction of free maternity services.

5. Statistical Analysis: Logistic regression models were used to estimate cross-sectional differences in women’s maternal health service utilization between voucher and comparison counties. Mixed-effects linear regression models were used to approximate the impact of the voucher program and the introduction of free maternity services on maternal health service utilization.

6. Control Variables: The analysis controlled for key potential confounders, including location (urban/rural), wealth quintile, year of childbirth, insurance enrollment, mother’s parity, education, marital status, and employment status.

7. Software: STATA IC V.15.1 was used for data analysis.

By analyzing the data collected through surveys and using statistical models, the study was able to estimate the impact of the voucher program and the introduction of free maternity services on maternal health service utilization. This allowed the researchers to assess the effectiveness of the recommendations and identify areas for improvement in access to maternal health services.

Auteurs et co-auteurs

Statistics:

Citations: 12

Authors: 7

Identifiers:

DOI: 10.1136/bmjgh-2018-000726

Research Areas:

Community Interventions, Disparities, Health System and Policy, Maternal Access, Maternal and Child Health, Quality of Care, Sexual and Reproductive Health, Social Determinants

Study Countries:

Kenya

Study Design:

Cross Sectional Study, Quasi Experimental Study, randomised-control-trial

Study Approach:

Quantitative

Participants Gender:

Female

Partagez ceci :