STRENGTHS AND LIMITATIONS OF THIS STUDY

The study used data from a nationally representative survey of females aged 15–49 years in over 850 clusters to explore common determinants of low birth weight.

The findings of this study can be generalised to low birth weight from individual and cluster levels in Ghana with spatial model consideration.

Sampling weights were applied at the individual level to account for the non-proportional allocation of the sample to the different regions to obtain a representation of the survey results.

Potential bias from missing data could underestimate the prevalence of low birth weight for estimates, as health records from such babies are more likely to have died.

Introduction

Low birth weight (LBW) is an important indicator of newborn health and can have long-term implications for a child’s development. Infants with a birth weight of less than 2500 g experience increased mortality and morbidity, as indicated by previous studies.1–3 LBW remains a significant public health concern worldwide, as it is associated with increased risk of many non-communicable diseases in both childhood and adulthood4 5 with increased burden among several low-income and middle-income countries.6 Despite various efforts and projections to decrease its prevalence, Africa still harbours about a quarter of all LBW cases, with the majority born in Eastern and Western Africa.7–10 In Ghana, several studies have established a significant relationship between socioeconomic indicators and LBW.11–13 LBW is attributable to several community, biodemographic and socioeconomic factors. While governments continue to implement policies or programmes to mitigate LBW, critical evidence is needed to support such initiatives. Despite the established relationships, a few studies have explored the spatial distribution for zone and place of residence (common determinants) to understand the extent of the problem geographically.

Spatial exploratory analysis provides a toolkit to gain insight into inequalities by combining geographical and demographic data to understand underlying factors of public health concerns.14 Clusters within regions in Ghana may exhibit varying spatial distributions of birth weight, potentially influenced by a complex interplay of determinants. While birth weight data are unavailable for 40 million newborns, with more than half of them from Africa,15 16 the problem of data credibility used for monitoring and reporting also persists. Also, several challenges17 18 have been documented regarding geographical accessibility to health services due to the inadequacy of physicians, rural transport infrastructure and national health insurance in Ghana. Geographical differences in the distributions of LBW are essential for designing effective interventions and policies to improve maternal and child health in Ghana. Researchers have actively explored spatial models19 20 to generate valuable insights into existing inequalities and actively used this knowledge to develop targeted interventions aimed at improving neonatal health and outcomes related to LBW.21

In this research, we seek to explore both individual-level and cluster-level distributions for LBW using spatial components for zone and place of residence from nationally representative survey data. The objective of this study is to explore the association of zone and place of residence with LBW using spatial exploration methods from nationally representative data.

MethodsData source and data description

We conducted this study using data from the 2017 Ghana Maternal Health Survey (GMHS),22 following the first one conducted in 2007.23 The GMHS contains a repository of data collected at household-level and individual-level on maternal health and maternal mortality in Ghana. The birth data allow for linking detailed information on birth histories for women to their individual data. A shape file is also available that contains information on clusters and geographical positioning systems (GPS), where participants were interviewed along with their geo-referenced locations (longitudes and latitudes). It is feasible to establish links for all these datasets, thereby incorporating a comprehensive cluster information for a more thorough analysis.

The survey sample for the individual-level data contains two levels of stratification. For the first stratification factor, the 10 administrative regions of Ghana were grouped into rural and urban areas, resulting in 20 sampling strata. In the second stage, we employed probability proportional sampling to select a total of 900 enumeration areas (clusters) from all the regions, yielding 466 clusters from the urban areas and 434 clusters from rural. The individual-level data contain information on 25 062 women aged between 15 and 49 years. For the purpose of this analysis, we included only women with singleton babies, relying on health records to ascertain their newborns’ birth weights within the 5 years preceding the survey.

There were 71 271 babies born to 25 062 women from the survey. In contrast, we excluded self-reported birth weights and infants born to women with non-singleton pregnancies from the analysis. The final sample size included 4870 babies born to women who provided health records or birth record cards, recognising their higher accuracy and reliability over maternal recall (figure 1).

Figure 1

Data flow description.

Data were collapsed from the individual level to the aggregated cluster level for spatial analysis. At this level, we excluded coordinates missing in clusters for longitudes and latitudes. 36 missing clusters of such cases were excluded from birth weight data. To protect the confidentiality of respondents used for the clusters, the GPS location was geo-masked from their actual locations to up to 2 km for urban and 10 km for rural points, but this continued to exist within administrative boundaries. There were 864 complete clusters corresponding to coordinates of birth weight from women’s data presented in figure 2A. The unit of analysis used in this study was at individual and cluster levels using zone and place of residence as common determinants.

Figure 2

Cluster location distribution (A) and zone categorisation (B) for low birth weight.

Study variables

We generated dependent and independent variables to fulfil our objective. Both zone and place of residence were used as independent variables. We formulated the zone classification based on regional locations representing the 10 administrative regions of Ghana. We, however, employed the zone classification for individual-level and cluster-level analysis, with the zones defined as Coastal (Western, Central, Greater Accra and Volta regions), Middle (Eastern, Ashanti and Brong Ahafo regions) and Northern: (Northern, Upper East and Upper West regions). Furthermore, we formulated the dependent variable as LBW<2500 g for individual and cluster-level analysis adhering to WHO definitions.8 In addition, we generated the LBW counts from clusters to formulate models used in the cluster-level analysis.

Patient and public involvement

None.

Statistical analysis

We performed data analysis by using Stata V.14.0 and R statistical software (R V.3.4.1 and RStudio V.1.3.959). At the individual level and cluster levels, we used boxplots to show the distribution of birth weight. We presented both weighted and unweighted frequencies for zone and place of residence at individual level. Also, a bivariate and multivariate logistic model was used to assess the association of zone and place of residence with LBW. Statistical significance was established at a level of 5%. Results were reported as OR and 95% CI.

At the cluster level, we summarised spatial exploration for the birth weight category, presenting summary tables for mean birth weight within cluster categories. A cluster categorisation will be provided for LBW (<2.5 kg) and disaggregated by region and zone. The cluster number can be mapped to the region and zone for the LBW categories. We conducted hypothesis testing to determine whether birth weights in clusters were randomly distributed using a Moran I statistic.24 25 To determine hotspots for LBW clusters, we performed the Gi* statistics.26

Additionally, we identified disaggregated information using counts of LBW in clusters on choropleth maps. We modelled the counts for LBW by clusters as our response variable, using a Poisson generalised linear mixed model with a spatial autocorrelation (fitme function in the spaMM package27 with a Matérn covariance matrix and maximum likelihood method). To test for the influence of zone and place of residence on LBW, we fitted a random effect model. Using spatial components, we conducted a likelihood ratio test based on the maximum likelihood of the full and null models.28 29 Furthermore, we tested the spatial correlation across clusters using the test for spatial autocorrelation from simulated residual in R.30 Finally, we conducted models for fixed and random effects with estimates, SEs, CIs and p values for two models.

ResultsLBW characteristics at the individual level

The mean birth weight for the individual-level analysis was 3.15 kg (95% CI 3.12 kg to 3.17 kg), as shown in boxplot distribution (online supplemental figure S1A). LBW prevalence was 7.2% (298/4127) for the individual-level analysis. There was also a higher prevalence of LBW in rural areas (3.9%) compared with urban areas (3.3%) (table 1). LBW was more prevalent in the middle zone (3.0%) compared with the coastal (2.4%) and northern zones (1.8%). The determinants zone and place of residence were significantly associated (p<0.05) with LBW in the bivariate analysis but not significant in the multivariate analysis (table 2).

Table 1

Baseline characteristics for individual-level common determinants

Table 2

Logistic model for LBW for individual level

LBW characteristics at the cluster level

The weighted mean for birth weight was summarised at cluster levels with northern (3.00 kg) and coastal (3.26 kg) zones, respectively, as presented in online supplemental figure S1A,B. Cluster locations of birth weight are presented in figure 2A. Mean birth weights were categorised at cluster level and revealed a prevalence of 1.9% (16/864) for LBW. Detailed information for the cluster, regional and zonal distribution for categorised LBW is presented in table 3 and figure 2B. There were 67.6% (584/864) clusters observed with no LBW counts while the remaining clusters had one or more LBW counts presented in online supplemental figure S2A. We used the Moran statistic to check for spatial correlation using weighted birth means within cluster levels. Hotspot analysis using Gi* in cluster counts revealed an indication of LBW in the middle and northern zones, as seen in online supplemental figure S2B. The log-likelihood from the null spatial model (−719.6) and full spatial model (−714.9) was significant with a likelihood ratio test of significance (p=0.002). Our final model II (table 4) provides results that help quantify the extent of geographical disparities in LBW using random effects of cluster locations in birthweigh counts and fixed effects of zone and place of residence. To assess the extent of geographical disparities in LBW, we assessed the cluster variance and Moran statistic for only model II. The correlation parameters, nu and rho, are also used to represent the strength and speed of decay in the spatial effect. Model I provides for the effect of a survey impact but without a spatial component. Model II considers the implications of spatial and survey components with fixed effects of common determinants and random effects for spatial error terms considering geographical variation (table 4). The final models were presented with estimates, SEs, CIs and p values. Accounting for spatial autocorrelation, our final model II, which incorporated spatial error terms together with the zone and place of residence determinants, yielded a Moran’s I statistic of 0.009 (p=0.044). Notably, clusters in urban areas demonstrated a significantly lower LBW (p<0.017) compared with rural areas. Furthermore, clusters in the northern zone exhibited a significantly higher LBW (p=0.018) compared with the coastal zones presented in (table 4).

Table 3

Cluster, regional and zonal distribution for low birth weight category

Table 4

LBW model consideration for cluster level for common determinants

Discussion

In this research, we seek to fill an important gap in LBW using common determinants with spatial and survey components. A prevalence of LBW, 7.2% from individual-level estimates, was lower compared with earlier statistics in Ghana (10.2%)31 and sub-Saharan Africa 9.76%,7 respectively. Our estimate of prevalence was halved in comparison to the current 2023 UNICEF-WHO LBW estimates database using models32 and different data sources. These differences in the estimate of prevalence could be attributed to various methods used in our study and that reported by UNICEF-WHO. Despite these differences, Ghana is unlikely to achieve the target of reducing newborn and child mortality (Target 3.2) of the 2030 Sustainable Development Goals as the desired decline in the last decade has not been consistent.33

To further understand how individual cases of LBW contributed to cluster-level prevalence, spatial clusters indicated that 32.4% (280/864) had at least 1 LBW baby initially. There were significant disparities observed among mean categorised clusters for LBW. We further identified an estimated cluster geographical prevalence of 1.9% (16/864) for LBW categories. This study is the first to estimate prevalence at the cluster level for LBW in Ghana. We found evidence of zonal hotspots for LBW clusters in Ghana’s northern and middle zones. Interestingly, adding a spatial component was beneficial in our spatial model as several facility-based studies conducted in Ghana’s middle and northern zones show a higher prevalence of LBW than the national average34–37 reflecting zonal disparities documented in our research. Differences in healthcare infrastructure, socioeconomic status and access to maternal healthcare services across regions may account for these disparities.

Even more so, our results indicate that LBW was significantly higher in the northern zones compared with coastal zones. The particular reason for such an occurrence can be attributed to the existence of main farmer households in the northern zones that experience a significant degree of food insecurity spanning between 3 and 7 months consequently, there is a high risk of maternal under nutrition with a higher prevalence of malnutrition in these zones.38–40 As reported in our study, poor dietary practices during pregnancy could also be linked closely with LBW41 in the northern zone.

Comparatively, we observed a noteworthy significant association between place of residence (urban/rural) and LBW, with considerably lower levels in urban areas compared with rural areas. Existing studies conducted in Ghana consistently identified a heightened risk of LBW in rural settings marked by elevated concentrations of poverty as documented in previous research.30 37 38 This suggests a persistent pattern across various studies, reinforcing the understanding that socioeconomic factors and the urban–rural divide substantially influence LBW. Almost half of the population living in rural areas22 can be linked to the ongoing challenges in government efforts to improve access to critical services. Despite endeavours to enhance access, significant barriers persist in health services and programmes, primarily due to socioeconomic factors. Spatial analysis in other parts of Ethiopia42 highlights a significant challenge, there was evidence that health access to health services in the predominantly rural communities where the majority resides. Even though LBW meddles with a complex interplay of determinants, including maternal nutrition, access to maternal health services and socioeconomic barriers, there exists a major gap in the rural–urban continuum.

Despite these important revelations, there are limitations worth highlighting. The cross-sectional nature of the data from the survey design limits us from making a causal relationship between the outcomes and explanatory variables. Also, potential bias in reports from missing data could underestimate the prevalence of LBW for individual-level and cluster-level estimates, as health records from such babies are more likely to have died hence health records are unavailable.

Conclusion

In conclusion, our objective to explore LBW at individual and cluster levels using spatial techniques was critical, as demonstrated by hotspots choropleth maps with clear signs of LBW clusters in Ghana’s northern and middle zones. Our study defines critical implications to include spatial components to address LBW inequalities in Ghana. As an initial step, policy-makers must consider spatial dimensions to allocate limited resources to Ghana. There are disparities between the rural and urban continuum, requiring focused programmes to reach all mothers in these vulnerable zones and bridge the gap between healthcare systems in Ghana’s northern and middle zones.

Data availability statement

Data are available on reasonable request. This study is a secondary data analysis of the 2017 Ghana Maternal Health Survey. The dataset is not publicly available but can be requested from the DHS programme at a reasonable request.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study used data from the 2017 Ghana Maternal Health Survey, which involved human participants and received ethical approval from ICF in 2017 (approval number: IORG0001475). The current study did not require further ethical approval or participant consent as it was a secondary data analysis of publicly available anonymised survey data. Informed consent was obtained from participants for the 2017 Ghana Maternal Health Survey prior to their participation. A formal data request was submitted to the DHS Programme via the official website (https://www.dhsprogram.com/data), and permission to use the dataset was granted.

Acknowledgments

The authors wish to extend our gratitude to the DHS programme that provided access to the Ghana Maternal Health Survey database. We are also grateful to the survey participants.