The Influence of Contextual Factors on Maternal Healthcare Utilization in sub-Saharan Africa: A Scoping Review of Multilevel Models

Inclusion criteria

Information sources

To avoid bias and to reflect the multidisciplinary nature of the subject of this review, literature searches were performed across a variety of databases including Science Direct, ProQuest, Google scholar, MEDLINE, EMBASE, PubMed, WHO Library Database (WHOLIS), African Health line, Wiley Online Library, (AIM), Web of Science, and Scopus

Literature search strategy

This review employed four sets of terms representing the outcome indicator variables, the contextual indicator variables, MLM filters and the filter for sub-Saharan African countries in the search strategy. Table 1 below reports an example of the search strategy adopted for this review. The search filter for sub-Saharan African countries was adapted from work by Pienaar, et al [35] who developed it for clinical studies conducted in an African environment. The filter comprised sub-Saharan African country names and truncated terms such as “Southern Africa”. North African countries and language filters were removed from the original list accordingly.

Data extraction and management

A search was conducted in databases using the strategy above to identify potential records for the review. All extracted entries were screened by examining titles and abstracts and all relevant records with the potential of meeting the criteria for inclusion were exported to EndNote X7 for data management. Full texts were extracted using the EndNote platform and those, which could not be found, were searched manually through the university library and other sources. Full texts were assessed by two independent researchers against the inclusion criteria and disagreements, where there were any, were resolved by consensus. Studies excluded from the review at this stage, along with reasons for exclusion, were recorded and retained by the researcher. The search also included hand searching reference lists of relevant articles to identify other articles of value.

Records fitting the inclusion criteria were then exported to excel to record the detailed general publication information (author(s), year of publication and type of journal), characteristics of studies (design, population, sample size and procedure, country, explanatory, and outcome variables of concern) and summary results (whether contextual factors significantly influenced maternal healthcare utilisation and recording the intra-class correlation ICC or variance partition coefficient VPC). The search was conducted on the 16^th of April 2018, and it was updated on the 14^th of June 2021.

Risks of bias and quality assessment

Poor quality assessment of studies has a considerable negative impact on the results of systematic reviews. As such, this review appraised all studies in accordance with the Quality Assessment Tool for Quantitative Studies developed by the Effective Public Health Practice Project [36]. Accordingly, the assessment considered selection, the appropriateness of study designs, whether there were stated confounders, whether there were variables which had been controlled for, whether blinding was applied, reliability and validity of data collection tools, the appropriateness of the units of analysis and the type of variables included. The way missing data were resolved was also part of the assessment procedures.

Just like the process of data extraction, two reviewers were involved in the quality assessment in the final processes of this review and in cases of disagreement, a third reviewer was enlisted to reach consensus. In cases of missing data, attempts were made to contact the primary authors of articles concerned. This was done in situations where, for example, study designs, explanatory or outcome variables were unclear or missing. When there was no success in obtaining the missing data, data were reported accordingly, and the implication thereof explained in the discussion section of the review.

Data analysis and presentation

Heterogeneity in the study population, selected countries, outcomes, and the nature of contextual variables was observed. As such, meta-analysis was not appropriate for this synthesis. Instead, narrative scoping review was employed as the method of reporting findings. A table was created with six columns and used to report the name of first author and reference in column 1. Contextual or structural factors that were examined after controlling for individual-level variables in column 2. Individual factors from adjusted models were reported in column 3. Column 4 reported the Intra-class correlation (ICC), which is an inferential statistic measuring variability within contextual factors influencing maternal healthcare utilisation. It is used here to determine the degree of variation in use of maternal health care attributed to contextual-level factors. Column 5 reports the outcome variable investigated in the research and the last column shows the direction of the relationship between significant contextual/structural factors and indicators of maternal healthcare under consideration.

Search results

The electronic data search process from the research platforms above yielded 1,654 potentially relevant records. Twenty-seven additional records were identified from the reference lists and citation checks of included studies. After searching for duplicates, 996 were discarded and 658 records were screened using titles and abstracts. Following this stage, 591 irrelevant manuscripts were excluded from the review and the remaining 67 records were fully assessed with respect to the illegibility criteria and among these, thirty-nine were selected. The main reasons for exclusion included studies not focusing wholly on sub-Saharan Africa, not using multilevel modelling techniques, using primarily qualitative methods, not being published in a peer-reviewed outlet, and having an outcome variable other than one of the indicators of maternal healthcare. After quality assessment for confounding and applying the measurement of contextual variables requirement in MLM studies, five were excluded and the remaining 34 records were eligible for inclusion in the study. The figure below is a flow diagram, which reports the process from article identification to inclusion.

Study characteristics

Apart from a few entries [37][38][39][40][41], almost all reviewed studies used the Demographic and Health Surveys (DHS) data from a standard survey that is implemented in several developing countries. McTavish and others [40] is the only study using the World Health Survey (WHS) and controls for the national level socioeconomic characteristics when investigating the relationship between national female literacy and maternal healthcare use. Table 2 reports the distribution of selected studies across nine sub-Saharan African countries, including some cross-national studies, which used a combination of SSA countries as a unit of analysis.

Sample sizes for included studies have a wide range because of heterogeneity in these studies. Some studies, for example, only focus on a relatively small rural population in one country [38][39][40][42][43] while some others pool data in a cross-national design, making the total sample size relatively large [13][23][41][44]. However, all the selected studies have sample sizes adequate to represent populations under investigation.

Reviewed studies are mainly cross-sectional and utilized multilevel logistic regression analysis with a few making use of multinomial and structural equation regression analysis. This is plausible because of’ lack of robust data infrastructure in SSA which prohibits the use of more comprehensive and advanced analytical techniques and could be the reason why studies using such methods as longitudinal multilevel analysis are missing.

Table 3 reports the results of the systematic review in terms of the study references, factors influencing different indicators of maternal healthcare, including both contextual and individual-level factors controlled for. The ICC and the direction of the relationship between contextual-level factors and maternal healthcare are also reported. There is observed heterogeneity in the constitution of contextual factors in included studies. However, in most studies, these factors are constituted by aggregating individual-level data to represent community characteristics often defined as clusters or PSUs in the case of the demographic and health surveys (DHS) data [23][24][44][45][46][47]. Studies by Balew et al, [47] and Ngome and Odimegwu [48] define contextual factors as structural determinants and social contexts respectively and focus exclusively on their influence on maternal healthcare. Most of the other studies seek to investigate the effects of both individual and community-level factors and emphasize the importance of considering factors operating at both levels in policy strategies aimed at bolstering maternal healthcare use.

Elfstrom and Stephenson [44] and Masters [46] are examples of studies using a three-level modelling technique in contrast with most studies using two-level models. Both studies assume that individual actors are nested within households which, are in turn nested within communities. They are interested in the influence of contextual factors on contraceptive use at the community level of analysis and they studied demographics and fertility norms, gender norms and health knowledge. These studies included household and individual-level factors, which were used as control variables.

Outcome variables in included studies are equally varied, covering the broader spectrum of maternal healthcare continuum of contraceptive use [44,48,52,53,61,62,66], antenatal care [43,45,51,54,56,67] skilled or facility delivery [24,37,40,42,47,57,64], and postnatal care [55,60]. Some studies included a combination of two maternal healthcare indicators as outcome variables of antenatal care and facility/skilled delivery [23,46,50,63,68] or three indicators of antenatal care, skilled delivery and postnatal care [38,39,59,63,65].

Contextual factors influencing maternal healthcare utilization

Table 3 also reports several contextual factors that were found to be associated with maternal healthcare after controlling for individual level characteristics. These have been grouped into four categories, including health systems, relational factors, and community socioeconomic position and cultural. Health systems variables are those that relate to the location, capacity, quality and The most common health systems level factors at the community level include distance or access to health facilities [24][37][40][42][43] [46][48][50][58][60][61][65][67] and quality of healthcare offered [37][38[45] [64][67][68]. Quality of health care is broadly defined to include the ratio of healthcare professionals to population, perceptions of the quality of local health facilities, care readiness and service availability.

The overarching nature of the relationship between health systems and maternal healthcare is such that women who live in communities that are proximal to well-performing and high-quality health systems have higher odds of utilizing maternal healthcare services. It was unusual that a few studies either found a negative association between distance and maternal health care [61] or the relationship was non-significant [37][61]. Higher levels of facility delivery in these cases were explained by the quality of healthcare, which seems to supersede distance as an explanatory factor for facility delivery. Other health care-related factors that were found to increase the odds of utilizing maternal health care system included trust in the health care system [60] and the lack of payment requirements [37].

A host of socioeconomic factors were found to be associated with maternal healthcare utilization at the contextual level, and these included community education levels[24][54][55][56][57][60][65], community wealth levels or poverty [53][57][60] [65], population density [45][63][65], community media exposure or exposure to family planning messages [24][47][52][54][64], place of residence [24][45][47][52][54][57] [58][59][62][63][65][67], and gender norms [23][44]. Most of these factors have a consisted and predictable relationship with maternal healthcare. For example, it is well-known that women who live in communities or countries with higher levels of education or economic status have increased odds of utilizing maternal healthcare.

On the other hand, it is also established that women who live in rural areas have a lower propensity of using maternal healthcare services. The same applies to community media exposure, which has mostly been found to have a positive influence on maternal health care. Problematic gender norms such as tolerance towards violence hinder women from using maternal healthcare.

Relational factors are also found to be important predictors of maternal healthcare utilization at the contextual level. These factors mainly result from everyday relationships or interactions women have with men. These relations may significantly influence women’s decisions to either use maternal healthcare or not. Community women empowerment [24][62], women autonomy [62], PSU household size [61], number of children per PSU [48] and small family norms [52][59] are among the most consistent predictors of maternal healthcare utilization.

The findings show that in sub-Saharan Africa, residence in communities where most women are empowered is advantageous in terms of use and access to maternal healthcare services. It is also evident that women who live in areas where their individual freedoms are inhibited have less likelihood of using maternal healthcare. The results also indicate that contextual small family norms are positively associated with contraceptive use.

To measure the independent variations in maternal healthcare attributable to contextual factors, we observed the value of the Intra-class Correlation Coefficients (ICC) for the contextual level factors in MLMs of reviewed studies. Included studies have not been consistent in reporting ICC measurements, which indicate the relative importance of contextual factors to individual-level factors in explaining variations in the use of maternal healthcare services. We extracted the ICC values from null models (i.e. before individual and contextual level variables are included) in all included studies. Where the ICC was not reported at all, it was computed using available variance partition information.

The ICC results reported in Figure 2 show that they are averaging 35%, which indicates that relative to individual-level factors, contextual level factors contribute less to maternal healthcare utilization but have substantial independent influence. There are a few studies which are above the ICC of 50% [24][43][54][57][58][64] and this show more impact attributed to contextual factors than individual-level characteristics. These studies are addressing antenatal care and skilled facility delivery using several contextual factors, including place of residence, women empowerment, community education, level of fertility, distance to health facility, poverty, and community media exposure, among others.

• Download figure
• Open in new tab

Figure 1. Flow of information through different phases of systematic review

Source: Moher et al, [30].

• Download figure
• Open in new tab

Figure 2. ICC values by study

Theoretical pathways linking contextual factors and maternal healthcare utilization

We sought to identify theoretical underpinnings behind observed relationships between contextual factors and maternal healthcare utilisation in SSA. Reviewed studies have mostly been sound in identifying the joint effects of individual and contextual level factors associated with maternal health care [24][38][39][54][57][64]. However, there are still challenges in articulating the underlying mechanisms behind these relationships. What, for example, are the social mechanisms making women living in underprivileged, less educated, or rural communities have inadequate maternal healthcare utilization? There are only a few studies that offer plausible explanations to unpack such findings. Adu and others [65] for example, attempt to explain the influence of educated communities in maternal health care utilization by illuminating the potential shift in community power structures that may result in more autonomy for women and subsequently better use of health services.

Kruk and others have equally attempted to do so by demonstrating the role played by contextual social interactions as a mechanism through which women make decisions to utilize maternal health care services [37]. They argue that, as women interact within communities, they tend to share information about the quality of health care systems and these collective perceptions shape women’s decision-making processes about use of health services. Furthermore, a few other studies that have used contextual social norms in explaining variations in health service utilization [23][42] help to illuminate the role of the social structure and how it is distinctive of individual characteristics in influencing health behaviour. This is an insightful approach as it gives a broader picture through which the observed relationships are shaped by prevailing social and cultural factors in a particular society.

Most reviewed papers do not offer strong theoretical frameworks on the influence of contextual factors on maternal health care utilization. This may be partly because of inability to articulate the presence of emergent contextual properties that are distinctive from individual-level constructs [69]. Lack of clear-cut distinction between contextual and individual-level constructs can result in atomistic fallacy, whereby conclusions about contextual-level variables are made from individual-level data [70]. There are many instances among reviewed studies where researchers are discussing contextual effects with individual-level data. This is problematic, and it poses a challenge in grounding such studies in theoretical constructs.