Study setting
The antenatal care package in Kenya follows the WHO evidence-based guidelines for comprehensive care and offers services such as weight and blood pressure measurement, tetanus toxiod vaccination, iron supplementation, tests for sexually transmitted infections, urinary glucose or protein, and HIV/AIDS, emergency preparedness and family planning, tuberculosis screening and detection, intermittent presumptive treatment (IPT) of malaria, and prevention of mother-to-child transmission of HIV. Additionally, the health education component of the ANC package includes counselling on birth planning, nutrition, physical activity during pregnancy, personal hygiene, and breastfeeding [11]. Pregnant women at low-risk of complications are recommended to attend ANC clinics for four comprehensive visits, starting in the first trimester of pregnancy (<16 weeks gestation) [11].
This study employs data from a community-based cross-sectional baseline survey undertaken between April-May, 2011, prior to the implementation of an innovative intervention that integrates orange-fleshed sweet potato (OFSP) promotion and production with nutrition education at ANC services in rural Kenya [16]. The baseline survey is a key part of the project evaluation strategy aimed at assessing the impact of integrating OFSP on the health and nutrition status of mothers and children less than two years of age. The survey was conducted in the catchment area of eight healthcare facilities serving a population of 362,151 individuals in the larger Bungoma and Busia districts of Western Province, Kenya [17]. The health facilities were selected based on the population served and the number of service providers, to facilitate the operational strategy of the intervention and to prevent leakage during OFSP project implementation.
Data
A two-stage sampling strategy at the village and household level was used to select and recruit the target population of pregnant women and mothers with children 6–23 months old (mother-child pair). The sampling frame included all villages in the catchment areas of the selected health facilities with the number of households per village based on Kenya Census 2009 [17]. Out of 498 villages, 104 villages were selected according to a probability proportionate to size cluster sampling technique. In almost all villages (n = 97), a key informant interview was conducted with a village elder using a structured questionnaire. Key informants were asked about village characteristics, agricultural activities in their village, market access for produce, health services access, and the presence of village nutrition and health committees. On average there were 113 households per village. A within-village sampling frame was established through a door-to-door census of each village. Women who were pregnant or had an infant up to 23 months of age (or both) were invited to come to the survey site for informed consent and survey administration.
The survey questionnaires were administered by trained local enumerators. The survey team, composed of six enumerators and one supervisor, travelled to selected villages and established a survey site at a central location in the village. Eligible women were informed in advance of the survey location in the community, generally at a village elder’s homestead, and were interviewed by an enumerator upon arrival. A project car was provided for transport if women expressed difficulties in reaching the survey site. Prior to survey administration, participants provided informed consent in accordance with ethical clearance from the Kenya Medical Research Institute Research Ethics Board.
Each participant was asked a multi-dimensional questionnaire designed to collect data on several domains, including participant knowledge of nutrition, vitamin A and orange-fleshed sweet potato (OFSP), diet diversity and consumption of vitamin A rich foods, health seeking behaviours, child care practices, agricultural practices and socioeconomic factors. Field supervisors coordinated the survey and checked each questionnaire at the end of the field day to ensure optimal data quality. A total of 2761 women participated in the baseline survey, of which 1781 were mother-child pairs and 980 women were pregnant at the time of the survey. Only data from pregnant women were used for this analysis.
Variables
Outcome variables
Four dependent variables characterizing maternal KAP were constructed; (1) nutrition knowledge score, (2) health & healthcare knowledge score, (3) attitude score, and (4) dietary diversity score.
Nutrition knowledge score (NKS) and health & healthcare knowledge score (HKS) were derived from key variables using equally-weighted summative item scores (see Additional file 1 for a list of survey items used). Weights were not used to generate knowledge scores as the selected items were relatively homogenous and equally important, and therefore, would not benefit from the added complexity of weighting and would risk incorrect weight assignment to items [18]. The NKS ranged from zero to eleven points and HKS ranged from zero to twelve points. The attitude score (AS) was also a summative score derived from Likert scale responses for hypothetical scenarios on health and nutrition. Participants’ responses to each of the hypothetical scenarios could range from ‘strongly agree’, ‘agree’, ‘neither agree or disagree’, ‘disagree’, or ‘strongly disagree’. A greater value was assigned to the most ideal response: ‘2’ for responses that reflected agreement with scenarios that show-cased preventative health practice or disagreement with scenarios that showed negative health-practices, ‘1’ for neutral responses and a score of ‘0’ for non-ideal responses. For instance, the scenario “Emily feeds everyone in her household sweetpotato for breakfast because it is more nutritious than bread.” seeks to assess the participants’ understanding of the nutritive value of foods. Agreement with this scenario would constitute an ideal response (i.e. score of ‘2’) and disagreement was regarded as non-ideal (i.e. score of ‘0’). Ideal and non-ideal responses for each scenario were judged by the research team. The AS ranged from zero to ten points.
The dietary diversity score (DDS) was the primary health practice of interest and was constructed from a 24-hour food recall, adding the number of different food groups out of twelve, which were consumed by the household within the last day [19]. Other health practices of interest were seeking treatment for malaria and intestinal worm (yes-no response for each) and qualitative assessment of amount of food intake during pregnancy as compared to before pregnancy (less than before, same as before, or greater than before).
Independent variables
The primary independent variable was attendance at ANC clinic (yes vs. no) at the time of the survey. We conceptualized ANC clinics as a key resource for pregnant women to attain knowledge regarding health and nutrition. Based on our review of the literature and evidence from a systematic review of factors affecting the utilization of antenatal care in developing countries [20], we further identified maternal, household, and village-level factors that would confound the association between ANC utilization and nutrition and health KAP (Figure 1). Increased uptake and coverage of ANC has been previously associated with higher maternal age, lower parity, higher maternal educational level, and higher socio-economic status [7, 14, 21–23].
We used the conceptual framework to guide the variable selection for the adjusted analysis in this study. Maternal demographic and socioeconomic characteristics from the survey, including age, gestational age, marital status, education level, agricultural activity and selling agricultural products, involvement in income generating activities, and radio use, as a proxy for media interaction, were identified as possible confounders in the association between ANC utilization and dependent KAP outcomes. At the household-level, household head education level, household size, and socio-economic status were identified as potential confounders of ANC use. The wealth index was used as a proxy for household socio-economic status, derived by constructing a scale from household-quality variables, such as type of housing and roofing, presence and type of toilet, the source of cooking fuel and the source of water during dry season, in addition to possession of household assets such as radio, television, telephone/mobile, solar panels, gas cooker, bicycle, water pump, motorcycle, car truck, tractor, and generator. The wealth index was categorized into quartiles for descriptive analysis [24]. Community-level variables such as village size, the presence of village health and nutrition committees or other health interventions in the areas, and distance to the nearest health facility were recognized as factors that influenced the ability to seek ANC services and maternal KAP.
Analysis
Data was entered using CSPro 4.0 software (US Census, 2011) and data cleaning was performed using SPSS version 19. Missing data and outliers were checked against hard-copy questionnaires to ensure accurate data entry. Data from 980 pregnant women who participated in the survey are included in these analyses. One case was excluded from further analyses due to a high proportion (>75%) of missing data. Descriptive statistics, including simple proportions, n (%), for categorical variables and mean with standard deviation for continuous variables, were noted for participant baseline characteristics.
Unadjusted analysis
In primary analysis, we hypothesized women who sought ANC services to demonstrate greater nutrition and health knowledge, positive attitudes towards preventive health practices, and better dietary diversity. Differences in knowledge (NKS and HKS), attitudes (AS) and dietary diversity (DDS) among women who had sought ANC services at least once at the time the survey versus those who had not, were assessed by Chi-square test and Student’s t-test, significant at two-sided alpha of less than 0.05.
Adjusted analysis
Due to the hierarchical nature of the data, multilevel modelling was initially employed to account for cluster sampling and to illustrate cross-village differences in the relationships between ANC attendance and maternal KAP [25, 26]. NKS and HKS demonstrated intra-class correlations (ICCs) below 5%, indicating that the between-village variance explained less than 5% of the total variance in the two knowledge scores (see Additional file 2 for ICCs for all four dependent variables). This suggests that the inclusion of contextual variables in adjusted analyses would not add value to the model as village characteristics explained little variance in maternal knowledge. The DDS and AS demonstrated greater clustering by village with ICC values of 6.8% and 4.9%, respectively, demonstrating small effects of village-level independent variables. However, the village-level characteristics measured in the survey were similar between women who had attended ANC clinics compared to those who had not, providing little evidence for confounding due to these variables (see Additional file 3 for table comparing village-level characteristics). As an additional check, multi-level modelling conducted for the dependent outcomes did not change the inferences (data not shown).
Hence, multiple linear regressions were employed to assess the impact of ANC attendance (none, <4, ≥4 visits) on NKS, HKS, DDS and AS, controlling for maternal and household-level confounders. A forward selection model-building approach was used, whereby independent variables were excluded from the model if they were insignificant above a two-sided p-value of 0.10 and did not substantially change the beta-coefficients of other variables when excluded (<10% change). In the final models, statistical significance for all variables was set at p < 0.05. Several interaction terms were tested in the models and included if they were statistically significant. Model fit was assessed by adjusted R-squared for linear regressions [25]. All statistical analyses were conducted using SAS version 9.2.