Background User fees have been reported to limit access to services and increase inequities. As a result, Kenya introduced a free maternity policy in all public facilities in 2013. Subsequently in 2017, the policy was revised to the Linda Mama programme to expand access to private sector, expand the benefit package and change its management. Methods An interrupted time-series analysis on facility deliveries, antenatal care (ANC) and postnatal care (PNC) visits data between 2012 and 2019 was used to determine the effect of the two free maternity policies. These data were from 5419 public and 305 private and faith-based facilities across all counties, with data sourced from the health information system. A segmented negative binomial regression with seasonality accounted for, was used to determine the level (immediate) effect and trend (month-on-month) effect of the policies. Results The 2013 free-maternity policy led to a 19.6% and 28.9% level increase in normal deliveries and caesarean sections, respectively, in public facilities. There was also a 1.4% trend decrease in caesarean sections in public facilities. A level decrease followed by a trend increase in PNC visits was reported in public facilities. For private and faith-based facilities, there was a level decrease in caesarean sections and ANC visits followed by a trend increase in caeserean sections following the 2013 policy. Furthermore, the 2017 Linda Mama programme showed a level decrease then a trend increase in PNC visits and a 1.1% trend decrease in caesarean sections in public facilities. In private and faith-based facilities, there was a reported level decrease in normal deliveries and caesarean sections and a trend increase in caesarean sections. Conclusion The free maternity policies show mixed effects in increasing access to maternal health services. Emphasis on other accessibility barriers and service delivery challenges alongside user fee removal policies should be addressed to realise maximum benefits in maternal health utilisation.
We used a retrospective interrupted time-series (ITS) design—which is one of the quasi-experimental designs.21 In this design, the data observed before the policies (intervention) would be used as control for the data observed during the intervention period. The Kenyan government is a devolved government system consisting of the national government and 47 semiautonomous counties.22 Health service delivery is devolved and consists of public, for-profit private and the not-for-profit private sector. The latter is mainly faith-based. Healthcare facilities are organised in a hierarchical manner with six levels; community health services (level 1), dispensaries (level 2), health centres (level 3), first referral subcounty hospitals (level 4), second referral county hospitals (level 5), and tertiary referral hospitals (level 6).23 In relation to health financing, Kenya has had several user fees reforms over the years, that have had an impact on maternal and child health, illustrated in figure 1. Soon after independence in 1965, user fees were abolished in all public facilities, later in 1989, they were brought back but suspended in 1990 and reintroduced in 1991.13 24 Level 4 and 5 public hospitals charged between KES 3000 and 6000 (US$30–60) for caesarean sections, while level 1–5 public facilities would charge patients between KES 700 and 2500 (US$7–25) for a normal delivery. Patients on average were charged KES 150 (US$1.5) for ANC and PNC services in public facilities. In 2004, there was an introduction of the 10/20 policy where user fees were abolished at the primary level and a registration fee of KES 10 and 20 (US$ 0.1 and 0.2) was levied in public dispensaries and health centres, respectively, and services for children under 5 years as well as those with special conditions such as malaria and tuberculosis were exempted from payment.25 Later in 2007, there was free deliveries in all public healthcare facilities, however, the extent to which this policy was implemented is unknown.25 A health sector services fund was established in 2010 that served to compensate healthcare facilities on lost revenues associated with the user fee removal policies.13 Later in June 2013, there was a presidential declaration that led to the abolishment of the 10/20 policy and the introduction of the free maternity policy. Subsequently, the Linda Mama free maternity programme was established in 2017. Timeline of user fee reforms in Kenya. ANC, antenatal care; PNC, postnatal care. This study was conducted across the 47 counties of Kenya and included public, private and faith-based facilities. Specifically, we included 5061 public level 2–3 facilities, 358 public level 4–6 facilities, 210 level 2–3 private and faith-based facilities and 95 level 4–5 private and faith-based facilities. Facilities with 100% missing data across all the utilisation indicators were excluded from the analysis. Private and faith-based facilities were included if listed by the NHIF as offering Linda Mama services.26 Five of the counties had universal health coverage (UHC) initiatives in the public sector; four were pilot sites for the country’s UHC programme since December 2018, and the other had a local county run UHC programme, that began in October 2016. The data source for the variables of interest was the Kenya Health Information System which is an open source web-based health information system used for reporting routine data.27 The level of missing data for the intervention outcomes was on average, 27% for caesarean sections, 29% for ANC visits, 51% for normal deliveries and 55% for PNC visits. Imputation of the missing data was done at the facility level using structural model and Kalman smoothing approach in instances where data were available for at least 50% of the time points. This imputation approach is recommended for longer and more complex time series that have trend and seasonality, as it very often produces accurate results.28 We analysed the level and trend changes in intervention outcomes (maternal health utilisation). Specifically, the intervention outcomes were ANC visits, normal deliveries, caesarean sections, and PNC visits. The intervention series covered a period of 89 monthly time points from January 2012 to May 2019. The time periods varied based on outcome and facility type due to the phased introduction of the Linda Mama programme that occurred at three different time points as illustrated in figure 2 (ie, phase 1: Introduction of deliveries in faith-based and private facilities in April 2017; phase 2: Introduction of deliveries in public facilities in July 2017 and phase 3: Introduction of ANC and PNC across all types of care in March 2018). Interrupted time-series study period. ANC, antenatal care; PNC, postnatal care. For instance, for deliveries in the private and faith-based facilities, the time periods included 17 months for the prefree maternity policy, 46 months for the period between the onset of the free maternity policy to phase 1 of the Linda Mama policy, and 26 months for the postphase 1 Linda Mama policy period. Deliveries in the public facilities included 17 months for the prefree maternity policy, 49 months for the period between the onset of the previous free maternity policy to phase 2 of the Linda Mama policy and 23 months for the postphase 2 Linda Mama policy period. ANC and PNC in public, private and faith-based facilities included 17 months for the prefree maternity policy period, 57 months for the period between the onset of the previous free maternity policy and phase 3 of the Linda Mama policy, and 15 months for the period postphase 3 of the Linda Mama policy. A total wash out period of 16 months was included, at different time points, that captured eight nationwide health workers strikes from 2012 to 2017.29 30 These washout periods were March 2012; September 2012–October 2012; December 2012–February 2013; December 2013; December 2016–March 2017 and June 2017–October 2017. Two interruptions were placed for each intervention outcome. The first interruption represented the introduction of the original free maternity policy and the second interruption represented one of the three phases of the Linda Mama programme introduction; June 2013 and April 2017 for deliveries in private and faith-based facilities; June 2013 and July 2017 for deliveries in public facilities and June 2013 and March 2018 for ANC and PNC services in private, faith-based and public facilities. We estimated the level and trend changes in the intervention outcomes. The analysis included the following independent variables; time (T) which was coded sequentially from 1 to 89; intervention status (X) which was defined as 0 for prefree maternity period, 1 for the period between the onset of free maternity policy and before Linda Mama, 2 for the periods were there were nationwide healthcare workers strikes and 3 for the period after the onset of the Linda Mama policy. The health workers strike period was used as a wash out period proxy. This is because during this time, utilisation of maternal health services was disrupted and does not reflect the true effect of the free maternity policies. The outcome variables were plotted against time to visually inspect the data for outliers, trends and seasonality. Initial analyses suggested overdispersion of the outcome variables, therefore a negative binomial distribution was assumed for the outcomes. Three different models were fitted for each of the four intervention outcome variables, one that had all 89 time points and did not define the wash out period, another that defined the wash out period and the last that excluded the wash out period. Adjustments were made for any seasonal effects by using harmonic terms based on the month of the year.31 We used the Akaike Information Criteria (AIC) to determine the best fitting model. The final model was expressed as follows: Where B0 is the baseline level at time 0, B1 represents the trend change in the preintervention phase, B2 represents the level change following the intervention and B3 represents the trend change following the intervention. We considered four forms of secondary analyses. The first assessed whether the observed level and trend changes were attributable to the implemented policies. This involved the use of non-intervention outcome as a control. Out-patient day (OPD) visits was included to control for ANC and PNC visits, while inpatient admissions in public facilities and faith-based facilities was chosen as a control for normal and caesarean sections. The free maternity policies were aimed at maternal health and therefore the controls were not directly targeted by the policies. An additional variable (Z) was added that denotes the type of outcome (whether treatment or control), and as a result, we fitted a controlled interrupted time-series (CITS) model of the form: Where B4 is the difference in intercept at time 0, B5 defines the trend difference between the intervention and control group in the preintervention period, B6 defines difference between the change in level in the control and intervention group associated with the intervention and finally B7 is the difference between the trend change in the control and intervention group associated with the intervention. We interpret B6 and B7 to infer any causal effects. In the second form of the secondary analysis, we conducted a combined ITS comparing the difference between the level and trend change in the five counties that had UHC initiatives in the public sector and the remaining counties that did not have any UHC initiatives. Third, we did an available case analysis where a separate ITS for the intervention outcomes without imputing for any missing data was done. Finally, the fourth secondary analysis involved the use of pseudo-start periods (replacing the true intervention start dates with other start dates along the preintervention period) for the intervention outcomes which had a pre-existing trend prior to the intervention. The AICs of the different regression models for both the separate intervention outcome models and the single CITS are shown in online supplemental table 1. For all the outcomes, the best-fitting model excluded the health worker strike periods, with or without accounting for seasonal trends. bmjgh-2020-003649supp001.pdf The model diagnostics included examination of residuals, autocorrelation function as well as partial autocorrelation function. table 1 reports the best-fitted negative binomial model estimates for the separate ITS for the intervention outcomes in public, private and faith-based facilities, while figures 3 and 4 visualise predicted numbers from the model that account and do not account for seasonality. The final model estimates for the control outcomes in the separate ITS analysis are reported in online supplemental table 2, while online supplemental table 3 reports on the model estimates for the single CITS. Interrupted time-series analysis for intervention outcomes in public facilities. ANC, antenatal care; PNC, postnatal care. Interrupted time-series analysis for intervention outcomes in private and faith-based facilities. ANC, antenatal care; PNC, postnatal care. Final negative binomial estimates for the intervention outcomes in the separate ITS analysis All segmented regression used a log link function with negative binomial distribution and p values are derived from z tests. Values in bold represent a strong evidence of an effect at a 0.05 level of significance. ANC, antenatal care; ITS, interrupted time series; PNC, postnatal care. bmjgh-2020-003649supp002.pdf bmjgh-2020-003649supp003.pdf In all the fitted and interpreted models in the primary analysis, the residual (see online supplemental figure 1), partial autocorrelation and autocorrelation (online supplemental figures 2 and 3) plots showed no evidence of autocorrelation in the data. bmjgh-2020-003649supp004.pdf bmjgh-2020-003649supp005.pdf bmjgh-2020-003649supp006.pdf Ethics approval to conduct the study was obtained from the Kenya Medical Research Institute/Scientific and Ethics Review Unit (KEMRI/SERU/CGMR-C/132/3735). We also obtained approvals from the Council of Governors, National Commission for Science, Technology and Innovation, the respective county department of health, and the health facilities management. No patients were involved in this ITS analysis.
N/A