The impact of Maternal and Child Health and Nutrition Improvement Project on maternal and child health outcomes and service utilization in Ghana: An interrupted time series analysis

Improving maternal and child health outcomes (MCH) continues to be a major public health concern to governments in sub-Saharan Africa and the international development community. The Maternal and Child Health and Nutrition Improvement Project (MCHNP) was a nationwide project that sought to improve the utilization of maternal and child health services in Ghana through nancial incentive packages. The objective of this study was to determine the differential impact of MCHNP on maternal and child health outcomes. A retrospective longitudinal pre-test post-test study design was employed. The study used monthly data from the District Health Information Management System between January 2014 to December 2018. Interrupted time series analysis was applied to estimate the impact of MCHNP on MCH for each region of the country. post-intervention signicant statistically p-value 1.70

The introduction of MCHNP in Ghana was based on wider disparities that existed in the country which affected use of maternal and child health services especially in remote communities. To reiterate, the Ghana Health Service (GHS) and Demographic Heath Survey (DHS) indicated in 2015 that only 41% of pregnant women in rural Ghana received skilled deliveries [13], clearly indicating low service utilization, although the country saw an improvement over the last decade in its maternal and child health outcomes, where it recorded a 380 maternal deaths per 100,000 live births and 60 child deaths in 1,000 live births. [14].
PBF initiatives in recent times have received several critiques following the Cochrane collaboration systematic review on PBF in 2012, which indicates that PBF in low and middle-income countries are based on too weak evidence and very little experience for a large scale implementation, hence the tendency to weaken the existing health system [15,16]. The objective of this study, therefore, was to determine the differential impact of MCHNP on maternal and child health service utilization and outcomes in Ghana.

Study design
The study employed a retrospective pre and post-test study design, which is a longitudinal study design. Monthly data on outcome variables were collected over January 2014 to December 2018. The adoption of the interrupted time series design was as a result of the lack of a control group, given that the intervention was implemented nationwide.

Study setting
The study was conducted in all the ten (10)

Study population
The study populations were pregnant women, post-partum women who accessed services obstetric services as well as those who died in the course of receiving these services, and babies who died within the rst 28 days of life under the review periods in all 10 regions of Ghana.

Data collection
Data for the study were obtained from the national health database District Health Information Management System II (DHIMS2). Data on the outcome variables were extracted from the various datasets in the form of a query for the period January 2014 to December 2018 and for the ten (10) administrative regions in the country as of 2018 using the pivot table application in the DHIMS2 database. Results from the queries were downloaded into an excel template for cleaning and management.

Data management
Data extracted were validated to identify any inconsistencies, incompleteness, and inaccuracies. Inconsistent data were compared with data in the other datasets in the system as a data element may appear more than once in the system to make room for checks and balances during data entry or transcriptions. This was done to enhance reliability in the data as well as producing a credible study result.

Page 4/19
The data extracted were saved unto a hard disk drive and were kept in a password-protected le to prevent unauthorized personnel's from accessing and making changes to the data which might affect consistency in the study results. Again a copy of the data was drafted into a cloud storage space to enable access to the data in case of any catastrophe.

Statistical analysis
Background information on maternal and child health service coverage, utilization and outcomes were summarized and further segregated into pre-intervention summary (January 2014-December, 2015) and post-intervention summary (January 2016 -December 2018).
Normality tests were carried out prior to the impact analysis to indicate whether or not the data are normally distributed using histogram super-imposed with a normal distribution curve and the Shapiro-Francia Test of Gaussian distribution. The mean comparison of the outcome measures pre and post-intervention periods was estimated using unequal Welch t-test for each of the dependent variables.
The evaluated intervention is said to target population-level health outcomes because the intervention was implemented at the population level with data collected at regular time intervals (monthly) over a speci ed period of time indicating clearly the pre and post-intervention periods with outcomes analyzed at the regional level to achieve the study objectives. An interrupted time-series analysis (ITSA) regression model was xed to determine the differential impact of MCHNP on the observed health outcome variables using the underlying trends determined over the study period.
An interrupted time series analysis (ITSA) is applicable when taking into account the effect of an intervention at a population level and the available data are reported at an aggregated level and when several measurements on the outcome variables are obtainable pre and postintervention period as it offers a hypothetically high level of internal validity [17].
Time series data are said to encounter serial correlation or autocorrelation (due to similarity between observations at a different time) as well as heteroscedasticity (due to differences in variability among the various subpopulations) in the model's error term. Cumby-Huizinga test for autocorrelation was employed to overcome serial correlation and heteroscedasticity at speci ed lags.

Estimation technique
Interrupted time series analysis (ITSA) relies on ordinary least square (OLS) regression due to its exibility in the context of ITSA and its ability to account for autocorrelated errors [18].
A single-group ITSA must be employed when the study has no comparison group that is, used only one group [18,19], hence the adoption of the single-group ITSA with the regression equation presented as below; Where; Y is the outcome variable of interest measured, T is the time periods speci ed for the study, X is a dummy variable representing the intervention (pre-intervention periods 0, post-intervention 1), TX is an interaction term and represent the impact of the intervention and ∈ been the error term.
In the case of a single-group study, β 0 represents the intercept or starting level of the outcome variable. β 1 is the slope or trajectory of the outcome variable until the introduction of the intervention.β 2 represents the change in the level of the outcome that occurs in the period immediately following the introduction of the intervention (compared with the counterfactual). β 3 represents the difference between preintervention and post-intervention slopes of the outcome variables" [18].
Statistical analysis and presentations were conducted using Stata IC 15.0 (StataCorp, College Station, USA) and MS Excel and a p-value of 0.05 was considered statistically signi cant.
The statistical analysis was followed by a plausibility argument in discussing the variations if any in the performances of the measured indicators across all the ten (10) regions of Ghana.

Page 5/19
Background characteristics of maternal and child health services and outcomes Table 1 presents background characteristics of the study population and intervention. Antenatal clinic (ANC) registrants increased by 47.5% post-intervention and a 59.7% increase in the number of pregnant women presenting for delivery. Deliveries conducted by TBAs post-intervention dwindled by 32.6% and also there was a reduction of 12.7% in stillbirths post-intervention. The maternal mortality rate decreased by 0.98% following the implementation of the intervention, but neonatal mortalities rate increased by 78.4% post-intervention.   Table 3 shows results of mean pre and post-intervention of all the outcome variables using unequal Welch t-test in the respective regions. Impact of MCHNP on the percentage of women who had four (4) ANC visits using interrupted time series analysis Table 4   Impact of MCHNP on the percentage of women who had skilled delivery using interrupted time series analysis Table 5 presents results from the impact analysis for the percentage of women who received skilled deliveries. As presented in the table, six of the ten regions in the country had positive impact on the variable been measured but none of these impacts were statistically  Signi cance level: *** p < 0.001, **p < 0.01, *p < 0.05

T-test comparing Pre and Post-intervention means of outcome variables by regions
Conversely the Greater Accra, Volta, Ashanti, and Brong Ahafo regions saw a reduced impact in skilled deliveries as shown in Table 2 of the supplementary material.

Impact of MCHNP on Maternal Mortality Rate using interrupted time series analysis
Following the introduction of the intervention, maternal mortality rates were anticipated to reduce across all the regions of the country. Table 6 displays the post trend change for regions that saw a positive impact in maternal mortality rate, however, these impacts recorded were not statistically signai cant. The Greater Accra, Volta and Western regions of the coastal belt recorded some reduction in maternal mortality rates with a post trend change of 0.001 (95% CI: -0.018, 0.017; p-value > 0.05), 0.010 (95% CI: -0.035, 0.016; p-value > 0.05) and 0.009 (95% CI: -0.031, 0.013; p-value > 0.05) respectively.  Fig. 3.
The Northern region was the only region in the savannah belt that recorded a reduction in maternal mortality rate with a post trend change of 0.024 (95% CI: -0.060, 0.011; p-value > 0.05).
Following the introduction of the intervention, maternal mortality rates, however, saw an increase in the Central, Ashanti, Upper East and Upper West regions shown in Table 3 Table 7. Signi cance level: *** p < 0.001, **p < 0.01, *p < 0.05 However, it appears that neonatal mortality rate rather increased signi cantly in the Western, Upper East and Upper West regions (p-value < 0.001). The Central, Greater Accra, Volta, Eastern, and Northern regions also saw an icrease in neonatal mortality rate following MCHNP implementatio as presented in Table 4 of the supplementary material

Discussion
This paper adds to the other quasi-experimental studies that found mixed impact (see Tables: 4, 5 which asserted that the introduction of performance-based nancing (PBF) incentives increased the utilisation and coverage of maternal health services chief of it been increased ANC visits [20].
Early and ongoing monitoring during pregnancy is believed to be associated with favourable birth outcomes. Prenatal care is bene cial in reducing the frequency of preterm deliveries, low birth weight as well as maternal and perinatal mortalities. Despite these effects of having multiple ANC visits, the impact of MCHNP on the recommended four ANC visits seem insigni cant in the Greater Accra, Ashanti, Brong-Ahafo, and Northern regions following its implementation. Although the impact in these regions were in the negative direction it can not be seen as the intervention did not have a positive impact on the variable as the indicator could has worsen should the intervention not been asserted that, performance-based incentives increased the number of women who delivered in the health facility by skiiled birth attendants [21]. The study further indicated that the combination of nancial incentive packages had a larger effect compared to the only performance incentives [21]. Again results from the study are consistent with the study in Rwanda which reported a signi cant improvement in institutional deliveries following the implementation of a PBF intervention hence recommended that PBF interventions are the best approaches to improving institutional deliveries in LMICs [22]. It is however indicated that, particular caution be taken when initiating alteration such as the introduction of nancial incentive packages in a system as intricate as the health sector to improve coverage and or performance [23].
Post trend analysis revealed a reduction in the percentage of women who had facility deliveries in the Greater Accra, Volta, Ashanti and Brong Ahafo regions. This unexpected impacts could be as a result of uno cial payments, lack of equipment to delivery this essential services in certain health centres, myths surrounding facility deliveries in most of our indigenous settlements, as well as poor collaboration between clients and service providers.
It must however be indicated that, although the impact is in the negative direction for some regions, it does not necessarily equate to the inability of the intervention not producing positive impacts, but rather it must also be seen that the indicator might have seen worst outcomes if the intervention was not implemented.
Greater Accra, Volta, Western, Brong Ahafo, Eastern and Northern regions of Ghana recorded reduction in its maternal mortality rate following the implementation of the performance-based nancial incentive, MCHNP. The impact although positive, that is MMR reduced, none of the reduction was signi cant statistically. Findings of this study are consistent with an impact evaluation study in Rwanda which recorded a reduced maternal mortality rate in the intervention group compared to the control group [22]. According to Africa Progress Panel (2010), performance-based incentives like MCHNP are meant to increase the availability of logistics, provide well-motivated staff to administer essential maternal and child health services to residents in remote settlements to increase coverage and utilisation of available essential care in other to reduce maternal mortalities in the region. The report as stated by Africa Progress Panel also indicated that increased availability and accessibility of essential maternal health and obstetric services to women of reproductive age leads to a massive improvement in curbing the increased rate of maternal mortality.
In contrast, the Central, Ashanti, Upper East and Upper West regions however saw an increase in maternal mortality rates following the implementation of the intervention as shown in Table 3 of the supplementary material. The increased rate might not necessarily be that the intervention did not have a positive impact though the impact is in the the negative direction. This could be that maternal mortality rates might have have seen an astronomic increase if the intervention had not been implemented. The unexpected impact of the intervention on this outcome however could be due to reasons elaborated as follows: Poor design or implementation of the intervention, a one t all design, that is, adopting broad non-speci c ideas to ones' the local setting during the implementation of the nancial-based incentive in settings as complex as the health sector. Health system factors such as the poor nature of the Ghanaian referral system from the low facility to the advance facilities for specialised services could be attributed to the unexpected impacts recorded. Again on the health system factors is the inappropriate distribution of critical staff like midwives and obstericains by managers to provide obstetrics and gynaecological and other essential services across the country. Also the lack of consumable and non-conusmable commodities in our facilities such as monitors, foetoscope, ultrasound scans machines (USG), cardiotocography (CTG) machines, blood and blood products and other lifesaving equipment could be linked to the increased maternal mortality rate in the Central, Ashanti, Upper East and Upper West regions.   Table 4 of supplementary material). The reduction in the rate is in concordance with an evaluation study in Mexico, which reported an increased improvement in child health outcomes following the implementation of PBF where the intervention reduced signi cantly childhood morbidities and mortalities by declining the occurrence of arrested growth among nursery children indicating an improvement in child outcomes [24,25]. Conversely, these ndings are inconsistent with the impact evaluation conducted in the Philippines following the implementation of PBF which reported that the intervention group of the performance-based incentives reported more morbidities and mortalities than the control group [26].
The Central, Greater Accra, Volta, Eastern, and Northern regions also recorded higher rate of neonatal mortality rates following the implementation of the intervention and this can be seen in Table 4 of the supplementary material of this manuscript. Despite the increased neonatal mortality rates recorded, it does not necessarily mean the intervention did not work in reducing neonatal mortality rate.
The intervention might have work even though the impact is in the negative direction as mortality rates might have worsen should the intervention not been implemented.
Poor design of the technical component of the intervention, the use of administrative data, limited skills due to inadequate in-service training for staff in the newborn area on who to help the baby breathe, infection prevention, neonatal resuscitation among others coupled with limited equipment such as incubators, phototherapies, radiant warms among others could be reasons for the insigni cant impacts.
Among other factors could be the inadequate staff strength in newborn areas, socio-economic and sociodemographic characteristics of parents and the poor referral system.

Strength and Limitation
The strengths in the application of ITSA are the use of administratively collected data, population been studied acts as its own control, employs modern regression modeling tools and techniques [27]. Among the limitations of the use of ITSA is that, ITSA are sensitive to the choice of the model for evaluating the impact and trend of the intervention and it solely depends on aggregated data for population-level studies as well as diffusion effects in the situation were the lags of events are too long [27].

Conclusion
The impacts recorded following the implementation of the intervention was statistically signi cant. ANC four visits increased in the The study recommends review its technical components and adopt speci c designs to address the needs of ones' local setting, to be able to achieve highly signi cant impacts.  Trend of neonatal mortality rate in regions that had signi cant impact

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryMaterialMCHNP.docx