GEHIP interventions spanned the three organizational levels of Ghana’s district health system ( Figure 1) in which clinical service resources are concentrated at the level of District Hospitals and public health services are managed by “District Health Management Teams” ( Figure 1C). Interventions designed to develop district leadership and political engagement, and improve financial planning interventions were directed to level C. Basic primary health care services are available in Sub-district Health Centers that vary in staffing and level of service provision but in general provide basic preventive and curative services. Figure 1B supervisory leadership and referral systems interventions were targeted on the paramedical staff who manage clinical operations and supervise community-level workers. At the periphery of the system ( Figure 1A), community-based care is integrated into the health system under the CHPS initiative. CHPS frontline workers are based in health posts located in catchment areas termed “zones,” of which about 6,500 have been mapped in rural areas, each with a service population ranging between 3000 and 5000 residents. About two-thirds of all zones are regarded currently “functional,” a term connoting zones where at least one resident community nurse is assigned. These nurses, termed Community Health Officers (CHO), have two years of basic primary health care service delivery training which includes a six-month regimen of peer learning provided by experienced CHO. Detailed content of CHO care has been extensively documented elsewhere ²¹ . It includes the provision of WHO mandated immunization services, treatment of common ailments of children according to the WHO mandated regimen of “integrated management of childhood illness,” (IMCI) and the provision of contraceptive methods, including outreach provision of oral contraceptives, condoms, and the injectable contraceptive depo-medroxy-pregesterone acetate. Referral services for clinical conditions and contraceptives are also provided by CHOs. Nurses are instructed to provide community outreach and doorstep services, and most CHPS health posts have a motorbike or bicycles to facilitate outreach activities.

GEHIP was a two celled non-randomized plausibility trial for testing the fertility and mortality impact of a program of health systems strengthening strategies that aimed to accelerate CHPS coverage ¹⁷ . Conducted in the Upper East Region over the 2010 to 2016 period, the GEHIP initiative greatly improved the provision of community-based primary health care through CHPS. The mortality impact of GEHIP on neonates was pronounced among neonates ²⁹ . Mortality also declined significantly among children aged 1 to 59 months, but this trend was realized in both treatment and comparison areas owing to the successful improvement in IMCI care in all UER districts ³¹ .

Difference-in-difference methods are used to assess the hypothesis that this achievement of UHC, with its impact on survival, will also be associated with significant improvements in reproductive health and fertility reduction. Assessing the impact of GEHIP tests the proposition that the process of UHC development will impact on family planning and fertility; testing the proposition that access to functional CHPS affects fertility tests the hypothesis that achieving UHC in Ghana will address need for contraception and reduce fertility.

The GEHIP intervention was located in four districts of Ghana’s Upper East Region ( Figure 2). Seven neighboring districts have provided a basis for statistical comparison of fertility and mortality trends. Two Upper East Region districts, Kassena-Nankana West and Municipal Districts, are localities where extensive NHRC research activities had been ongoing for decades and conditions there would not be representative of the region. Since these districts were atypical of the health care environment elsewhere in the region, they were excluded from GEHIP. A program of interventions was launched in treatment districts to expand the range of CHPS services to include emergency referral care, improve the quality and range of primary care services, and develop leadership for sustaining community and political engagement in the CHPS implementation process ^{17, 32} .

Birth histories were compiled using interviewing instruments published by the Ghana Demographic and Health Survey of 2008 ³³ for interviews of all women resident in GEHIP sample households aged 15 to 49. Sample size and power calculations were designed to permit difference-in-difference evaluation of posited child survival effects of GEHIP exposure ³¹ . In the baseline survey (available as Extended data) ³⁴ , 66 clusters in GEHIP intervention and comparison districts were randomly selected based on the enumeration areas of the 2010 census (Ghana Statistical Service 2004). In the second stage, random household selection proceeded within each cluster proportional to enumeration area size by listing all households in sample clusters. The second stage sampling proceeded at a fraction designed to yield a target sample size of 6000 women of reproductive age. At the endline, the baseline survey clusters were reused to establish longitudinal records of GEHIP exposure. However, relisting and repeat stage two resampling was pursued. Therefore, the GEHIP baseline and endline merged data is a panel at the cluster level only. Interviews were conducted in the prevailing local language of sample households. Baseline survey interviews of 5511 women of reproductive age were conducted out of an estimated sample of 6000, yielding an achieved sample of 91.8 percent of the total sample. Correspondingly, 5914 out of a targeted sample of 7588 women were interviewed in the endline, yielding a 76% achieved endline sample. The baseline survey was paper-based and took place in early 2011 prior to the onset of GEHIP implementation. The endline took place in late 2014 and early 2015 utilizing the paperless “Open Data Kit” (ODK) technology to facilitate data editing and correction at the time of interviews ³⁵ .

Baseline and endline household survey data were merged and used to estimate separate Heckman “difference-in-differences” (DiD) models that compared the change in contraception, unmet need, and fertility measures over time in the treatment area with corresponding changes estimated over time for the comparison area ³⁶ . Average treatment effects (ATE) were estimated as follows:

A T E = E ( Y M t ' – Y M t ) – E ( Y C t ' – Y C t ) ( 1 )

In this model, Y describes a health outcome such as a birth event, the subscript t refers to measurements of health outcomes at baseline, t’ refers to measurements of health outcomes at the end of the point of observation, M indexes GEHIP exposed sample cluster areas and C indexes comparison sample cluster areas ³⁶ .

While the Heckman formula is widely applied, it has limitations for health systems research Heckman and others have recommended multivariate extensions of (1) for offsetting confounding effects that are unaddressed in (1) ³⁷ . To refine the model, adjusting for the potentially confounding effects of social and demographic characteristics of households, regression methods refine the estimation of the net effect of GEHIP, and permit estimation of the conditional effect of GEHIP on segments of maternal age that may represent contrasting responses to UHC access. Moreover, utilization of GEHIP results for interpreting systems effects requires a systems approach to the Heckman procedure ^{38, 39} . To address the need for a systems approach, monitoring operations during GEHIP assessed the location, timing, and content of community-based primary health care. Census enumeration areas are used for defining areal units of exposure to GEHIP and CHPS services in the statistical analysis.

Discrete time hazard regression analysis was applied to merged baseline and endline birth history data, with provision for linking exposure to CHPS services at exact ages of mothers as GEHIP implementation and maternal parity progressed. For the analysis of GEHIP impact on fertility, units of observation are maternal months registered in GEHIP baseline and endline survey pregnancy histories. The model for estimation is a logistic regression model for the age conditional birth event function defining the TFR, given by:

In [ p m i 1 − p m i ] = α 0 + ∑ t = 2 7 α t   a g e t m i + δ 1 t r e a t m i + δ 2 p e r i o d m i + δ 3 ( t r e a t m i * p e r i o d m i ) + ∑ t = 2 7 γ t ( a g e ( t ) m i * t r e a t m i ) + ∑ t = 7 γ t ( a g e ( t ) m i * p e r i o d m i ) + ∑ t = 2 7 γ t ( a g e t m i * t r e a t m i * p e r i o d m i ) + ∑ j = 1 4 β j X j m i ( 2 )

where,

By setting covariates at sample grand means and summing predicted probabilities of (2) for all five-year age groups from 15 to 49 and the 19 implied interaction terms, this numerical integration of linear combination estimates defines the total fertility rate (TFR). Therefore, linear combinations implied by the parameters of (2) express conditional TFR and treatment or programmatic exposure differences between predicted TFR define expect births averted associated with exposure to program activities. The treatment by time interaction parameters δ and γ thus specify regression adjusted estimates simulating Heckman’s Model 1 “Average Treatment Effect.” ^{36, 37} .

GEHIP represents exposure to a process of system improvement rather than an end product of UHC development. Model (2) therefore tests the impact of household exposure to a process of systems strengthening of CHPS that unfolded during the GEHIP implementation period. Model (2) thus tests the proposition that pursuing the process of UHC development at the district level has reproductive health and fertility effects. Assessing direct UHC fertility effects can be represented by estimating the impact of residing in a residence located in a service area where CHPS services are functional, as given by:

l o g i t   f j = θ 0 + ∑ j = 2 7 θ j a g e i j + λ 1 C H P S i + λ 2 p Y e a r i + ∑ k = 1 K ξ k z i k + ∑ j = 2 7 θ j ( a g e i j * C H P S i ) ( 3 )

where, θ 0 + ∑ j = 2 7 θ j a g e i j defines the contribution of individual i to the TFR, as in Model (2) and

Since the age parameters of equation 3 define adjusted age specific fertility effects if CHPS is fully operational, parameter estimates for CHPS exposure represent an estimation of the fertility effects of achieving UHC in Ghana.

Ethical safeguards for the GEHIP project and its data collection processes were instituted and approved by the ethical review procedures of the Ghana Health Service and Columbia University.

Baseline results portrayed in Table 1 assess the statistical balance of the survey data (available as Underlying data) ³⁴ . In the endline, women in both intervention and comparison areas were younger, less likely to have given birth before, and less likely to be married. In terms of religious affiliation, the majority of women were Christian, followed by Muslim and then traditional. The number of women who professed traditional religion declined between the baseline and the end line in both arms of the experiment while those born to women of the Christian faith increased. Both literacy and poverty levels decreased between the baseline and endline periods. The proportion of women living in a functional CHPS zone increased over time in both intervention and comparison areas, though the increase in the treatment area was much greater probably due to the effect of GEHIP on CHPS scale-up.

All surveyed women were asked to provide information regarding their pregnancy and childbearing experiences, following interviewing procedures routinely conducted by Demographic and Health Surveys. Women were asked to provide information on all live births they have had in their lifetime including the age, sex, and survival status of all live births reported. Based on the birth history data, we computed basic fertility indicators such as the age-specific fertility rates and total fertility rates using children born within the last 12 months for all districts combined and separately for the intervention and non-intervention districts. Based on the birth history data, age-specific fertility rates and the total fertility rates using children born within the 12 months prior to interviews could be assessed, leading to the estimation of Model 2. The GDHS estimated TFR in the Upper East Region was 4.1, roughly one half of one birth lower than the GEHIP baseline TFR in both treatment and comparison areas. This is most probably due to the fact that the urban parts of the region were excluded in the GEHIP activities and districts covered by the Navrongo project were also excluded from GEHIP research.

The difference-in-differences estimate of the net effect of GEHIP on fertility presented in Table 2, Model 2A, indicates that GEHIP had no overall impact on total fertility. Model 2B introduces an age interaction into the difference-in-differences model, with predicted values that estimate regression adjusted age specific fertility rates and the TFR ( Figure 3). The regression-adjusted TFR was slightly higher in comparison areas (4.7), relative to the treatment areas (4.5). Each area experienced a small estimated decline in the TFR, ranging from 0.3 to 0.4 births, yielding a null overall fertility impact of GEHIP. GEHIP was associated with reduced fertility at the peak ages for childbearing, ages 25-29, relative to the comparison area (OR = 0.83, p<=0.025, 95% CI 0.71, 0.98). Figure 3 illustrates that the net fertility reduction attributable to GEHIP among women 25–29 is due to a small decrease in fertility for this age group served by GEHIP, in contrast to a larger concomitant increase in fertility among women 25–29 in the comparison area.

The fertility effects associated with exposure to CHPS as estimated by Model 3 are reported in Table 3. Columns labeled Model 3A present results for main effects of CHPS only; whereas Model 3B includes parameter estimates for assessing the possibility that CHPS effects are conditional on age. Results show that exposure to CHPS is associated with an overall fertility reduction of approximately 5%. The Model 3B results is therefore consistent with the proposition that CHPS exposure reduces fertility. While the effect is not pronounced, representing impact of less than 0.3 births off the TFR, CHPS exposure has significant fertility reducing effects. As Figure 4 shows, this impact arises from the significant age conditional effect of CHPS exposure among women under age 20 (OR = 0.86, p=0.022, 95% CI 0.76,0.98). and among women aged 35 to 39 (OR = 0.87, p=0.015, 95% CI 0.78,0.97). There is no evidence of CHPS effects among other age categories.

Difference-in-differences analyses of GEHIP net effects on contraceptive prevalence and unmet need provide useful reproductive behavioral contextual information for understanding the fertility effects of CHPS and GEHIP. The increase in contraceptive prevalence between baseline and endline was significantly higher in the GEHIP study area, with a net difference-in-differences increase in the GEHIP treatment area, relative to the comparison area, of nearly 80% ( Figure 5). The adjusted estimate of the net increase in modern contraceptive prevalence due to GEHIP was estimated to be an odds ratio of 1.80 (95% CI 1.32 - 2.44). While results thus show that GEHIP implementation was associated with increased contraceptive use, a concomitant increase in unmet need was ongoing in both treatment and comparison areas ( Figure 5).

Data Archiving and Networked Services: GEHIP PLAUSIBILITY TRIAL _FERTILITY. https://doi.org/10.17026/dans-xph-vte9 ³⁴ .

This project contains the following underlying data:

Data Archiving and Networked Services: GEHIP PLAUSIBILITY TRIAL _FERTILITY. https://doi.org/10.17026/dans-xph-vte9 ³⁴ .

This project contains the following extended data:

Please note that free, unrestricted registration with Data Archiving and Networked Services is required prior to data access.

Data are available under the terms of the Creative Commons Zero “No rights reserved” data waiver (CC0 1.0 Public domain dedication).