1 Introduction

1.1 Research Problem

Antenatal care (ANC) is a cornerstone of maternal and child health, proven to reduce maternal mortality, stillbirths, and low birth weight (WHO, 2016). The World Health Organization recommends a minimum of eight contacts during pregnancy, with at least four visits considered essential for basic interventions (WHO, 2016). In Kenya, despite progress in maternal health indicators, only 58% of pregnant women complete at least four ANC visits, and significant disparities persist across socioeconomic and geographic lines (KNBS & ICF, 2023). Beyond simple coverage, understanding the determinants of ANC visit frequency—a continuous measure of care intensity—is critical for designing targeted interventions to improve both the quantity and quality of antenatal contact.

1.2 Hypotheses

This study was guided by the primary hypothesis that ANC utilization in Kenya is determined by a combination of enabling resources (household wealth, maternal education) and predisposing factors (marital status, ethnicity, residence). Secondary hypotheses cited that demographic factors (maternal age) and geographic access would also influence visit frequency. The analysis tests these relationships while controlling for cultural and religious factors, with particular attention to potential wealth-ethnicity interactions given Kenya’s diverse socioeconomic landscape.

2 Methods

2.1 Study Design

A cross-sectional secondary analysis was conducted using data from the 2022 Kenya Demographic and Health Survey (KDHS). The KDHS employs a two-stage stratified sampling design to obtain nationally representative estimates. The complex survey design (clustering, stratification, and sampling weights) was accounted for in all analyses to ensure generalizability to the national population of women of reproductive age.

2.2 Study Population

The analysis included women aged 15–49 who had a live birth in the five years preceding the survey and provided complete data on ANC visits. Women reporting ANC visits as “don’t know” (code 98) or “missing” (code 99) were excluded from the analytic sample, as these codes represent non-numeric responses rather than visit counts. This resulted in an analytic sample of 10,529 women, representing a subset of women with a live birth in the five years preceding the survey.

2.3 Study Variables

Outcome Variable: The number of ANC visits during the last pregnancy, treated as a count variable (range: 0–20). Visual inspection revealed a right-skewed distribution with a mode at 4 visits, consistent with WHO’s minimum recommendation.
Explanatory Variables: Selected based on the Andersen-Newman Behavioral Model of Health Services Use (Andersen, 1995).
- Predisposing Factors: Maternal age group, marital status, ethnicity, religion.
- Enabling Resources: Household wealth quintile (reported as wealth index), maternal education level, current work status, type of residence (urban/rural).
- Need/Access Factor: Self-reported distance to the nearest health facility (in kilometers, as a proxy for geographic access).

2.4 Statistical Analysis

All analyses were performed in R (version 4.5.2) using the survey package to account for the complex sampling design. Descriptive statistics were calculated using survey-weighted proportions and means. Model diagnostics, including assessment of residual patterns, variance inflation factors (VIFs), and influential observations (Cook’s distance), were conducted to ensure model adequacy.

The primary analysis employed a survey-weighted quasi-Poisson regression model (svyglm with family = quasipoisson(link = "log")) to model the count of ANC visits. Diagnostics revealed underdispersion (dispersion parameter φ = 0.63), indicating less variability in visit counts than expected under a standard Poisson distribution; the quasi-Poisson family was chosen as it appropriately adjusts inference for under- or over-dispersed count data without changing coefficient estimates (Ver Hoef & Boveng, 2007). Results are presented as Incidence Rate Ratios (IRRs) with 95% Confidence Intervals (CIs). An IRR > 1 indicates a higher expected number of visits, while an IRR < 1 indicates a lower expected number, relative to the reference category. Model fit was assessed using the pseudo-R² statistic. Statistical significance was set at α = 0.05.

3 Results

The final analytic sample included 10,529 women. Comprehensive model diagnostics indicated good model specification: variance inflation factors for all predictors were below 1.5, suggesting negligible multicollinearity; Cook’s distance values were all below 0.004, indicating no unduly influential observations; and residual plots showed acceptable patterns. The model pseudo-R² was 0.114, indicating that approximately 11.4% of the variance in ANC visits was explained by the included predictors—a typical range for models of complex health behaviors in population-based studies, where unmeasured individual and community-level factors account for substantial variation.

	Results
Determinants of Antenatal Care Visits Among Women in Kenya
At the national population level, accounting for DHS sampling design, what variables are associated with antenatal care visits?
Variable	IRR	95% CI	p-value
(Intercept)	3.15	2.75, 3.61	<0.001
Respodent's Age Group
15-19	—	—
20-24¹	1.03¹	0.97, 1.10	0.3¹
25-29¹	1.03¹	0.97, 1.10	0.4¹
30-34¹	1.04¹	0.97, 1.12	0.2¹
35-39¹	1.00¹	0.93, 1.07	>0.9¹
40-44¹	0.96¹	0.87, 1.06	0.4¹
45-49¹	0.92¹	0.77, 1.09	0.3¹
Marital Status
Never in union	—	—
Married²	1.10²	1.05, 1.16	<0.001²
Living with partner²	1.09²	1.02, 1.17	0.012²
Widowed	0.90	0.77, 1.04	0.2
Divorced²	1.20²	1.01, 1.43	0.042²
Separated	1.07	0.98, 1.17	0.12
Respodent's Ethnicity
Other	—	—
Embu	0.91	0.82, 1.01	0.080
Kalenjin³	0.86³	0.81, 0.91	<0.001³
Kamba	1.03	0.95, 1.11	0.5
Kikuyu³	0.93³	0.87, 0.99	0.023³
Kisii³	0.91³	0.85, 0.98	0.010³
Luhya	0.99	0.93, 1.05	0.8
Luo	0.98	0.90, 1.06	0.6
Maasai	0.97	0.89, 1.05	0.5
Meru³	0.91³	0.83, 1.00	0.042³
Mijikenda/Swahili	1.05	0.98, 1.12	0.2
Somali³	0.77³	0.64, 0.92	0.004³
Taita/Taveta	0.94	0.80, 1.11	0.5
Respondent’s Religion
Other	—	—
Catholic	1.06	0.96, 1.16	0.2
Protestant	1.05	0.97, 1.15	0.2
Evangelical churches	1.04	0.95, 1.14	0.4
African instituted churches	1.07	0.97, 1.18	0.2
Orthodox	0.81	0.64, 1.01	0.064
Islam	1.03	0.93, 1.13	0.6
Hindu	0.97	0.81, 1.17	0.8
Traditionists	0.87	0.67, 1.13	0.3
No religion/atheists	0.97	0.84, 1.12	0.7
Wealth Index
Poorest	—	—
Poorer⁴	1.07⁴	1.02, 1.11	0.002⁴
Middle⁴	1.10⁴	1.06, 1.15	<0.001⁴
Richer⁴	1.14⁴	1.08, 1.20	<0.001⁴
Richest⁴	1.32⁴	1.22, 1.42	<0.001⁴
Place of Residence
Urban	—	—
Rural	1.04	1.00, 1.09	0.049
Highest Educational Level
No education	—	—
Primary	1.05	0.96, 1.14	0.3
Secondary	1.05	0.97, 1.14	0.2
Higher⁵	1.16⁵	1.06, 1.26	<0.001⁵
Currently Working
No	—	—
Yes	1.01	0.98, 1.03	0.7
Nearest Facility⁵	1.00⁵	1.00, 1.00	0.5⁵
¹ After adjustment, maternal age, work status, and minutes to nearest facility were not significant predictors, suggesting socioeconomic and cultural factors are more influential than geographic access.
² Marital status mattered: married women and those living with partners had more visits; divorced women showed the highest relative visit frequency.
³ Ethnic disparities persisted after adjustment, with Somali and Kalenjin women having fewer visits compared to other groups.
⁴ Household wealth demonstrated a strong positive gradient, with richest women having substantially higher ANC visit frequency.
⁵ Higher maternal education and rural residence were modestly associated with increased ANC visits.
Abbreviations: CI = Confidence Interval, IRR = Incidence Rate Ratio
Source: DHS Data Program 2022

Figure 1: Adjusted Associations Between Selected Characteristics and Antenatal Care Visits Among Women in Kenya (Survey-Weighted Quasi-Poisson Regression).

3.1 Key Findings

Household wealth demonstrated a strong, graded positive association. Compared to women in the poorest quintile, women in the richest quintile had a 32% higher expected number of ANC visits (IRR=1.32, p<0.001), with a clear monotonic increase across quintiles.
Marital status was significantly associated with ANC utilization. Married women and those living with a partner had approximately 10% more visits than never-married women. Notably, divorced women had the highest relative visit frequency (IRR=1.20, p=0.042).
Pronounced ethnic disparities were observed, independent of wealth. Women from Somali ethnicity had 23% fewer expected visits (IRR=0.77, p=0.004), and Kalenjin women had 14% fewer visits (IRR=0.86, p<0.001). These effects persisted after controlling for wealth, education, and residence.
Higher maternal education and rural residence were associated with modest increases in ANC visits (IRR=1.16 and 1.04, respectively). Primary and secondary education showed positive but non-significant associations.
Maternal age, current work status, and distance to the nearest health facility were not significantly associated with ANC visit frequency in the adjusted model, suggesting that socioeconomic and cultural factors outweigh pure geographic access in determining care intensity.

4 Discussion

4.1 Interpretation

The findings strongly support the primary hypothesis, highlighting economic capacity and social structure as fundamental determinants of ANC utilization in Kenya. The persistent wealth gradient aligns with studies across sub-Saharan Africa, where economic barriers such as indirect costs (transport, lost wages) and social marginalization limit healthcare access for the poorest women, even for nominally free services (Boateng et al., 2014). The graded nature of the association suggests that economic disadvantage operates along a continuum, not merely as a poverty threshold.

The advantage associated with being married or in a union is consistent with the literature on social support and shared resources facilitating health-seeking behavior (Story & Burgard, 2012). The particularly high utilization among divorced women is a novel finding that warrants qualitative exploration, potentially relating to greater autonomy in health decisions, specific support networks, or a heightened perception of vulnerability. The lack of association with maternal age contradicts some prior studies but may reflect successful targeting of adolescent health programs in Kenya.

The significant ethnic disparities, especially for Somali women, point to deep-seated cultural and potentially systemic barriers. These may include language differences, cultural norms around pregnancy, historical marginalization, discriminatory experiences within the health system, or co-location with regions experiencing security challenges (Mohamed et al., 2022). Crucially, these disparities were independent of wealth, suggesting they stem from factors beyond material deprivation.

The lack of association with minutes to a nearest facility is notable and mirrors findings from other DHS analyses, suggesting that when physical infrastructure is present, social and economic factors become the primary constraints (Kitui et al., 2013). The modest rural advantage is intriguing and may reflect stronger community health worker networks or different social norms regarding preventive care in rural areas.

The observed underdispersion (φ = 0.63) is methodologically and substantively significant. It indicates less variation in visit counts than expected under a standard Poisson model, likely reflecting the success of standardized ANC scheduling and messaging in Kenya, which creates a “ceiling effect” around the recommended 4–8 visits. This reduced variability suggests a degree of programmatic success in normalizing a minimum standard of care.

4.2 Limitations

This study has limitations inherent to cross-sectional survey data, including the inability to establish causal relationships. ANC visits were self-reported and subject to recall bias. Important unmeasured confounders, such as pregnancy intention, quality of care, male partner involvement, and detailed measures of cultural beliefs, were not available. The exclusion of women who could not recall their visit number (“don’t know”) may have biased the sample toward women with more regular care schedules. Finally, while the model diagnostics were favorable, the quasi-Poisson model accounts for but does not explain the source of underdispersion.

5 Conclusion

5.1 Conclusion

This nationally representative analysis, validated by robust model diagnostics, confirms that ANC utilization in Kenya is inequitably distributed along economic, social, and ethnic lines. Household wealth, marital status, and ethnicity are key independent determinants of visit frequency, overshadowing the role of geographic proximity and individual demographics. The economic gradient is continuous, and ethnic disparities persist even after accounting for socioeconomic status. This indicates that achieving universal ANC coverage requires moving beyond a focus on facility placement to address underlying social and economic determinants of health and to confront specific cultural and systemic barriers faced by marginalized ethnic groups.

5.2 Recommendations

Programmatic: Maternal health programs should integrate poverty-sensitive strategies, such as conditional cash transfers or transportation vouchers, targeted at the poorest women. These should be complemented by culturally competent outreach and services tailored to the specific needs of Somali, Kalenjin, and other underserved ethnic communities. The role of community health workers in bridging these gaps should be strengthened.
Policy: National policies should strengthen intersectoral collaboration between health, social protection, and gender affairs ministries to mitigate the economic and social vulnerabilities that limit ANC access. Policies should also explicitly address ethnic disparities in health, potentially through targeted funding, language services, and anti-discrimination training for health workers.
Research: Future studies should employ mixed-methods approaches to explore the qualitative mechanisms behind the quantitative associations found here, particularly the high utilization among divorced women and the specific barriers faced by Somali communities. Research should also investigate the potential interaction between wealth and ethnicity. Implementation research is needed to test the effectiveness of equity-targeted interventions, and longitudinal studies are required to establish causal pathways.

By systematically adhering to this research framework and rigorously testing model assumptions, the findings provide credible, evidence-based insights to guide more effective and equitable maternal health strategies in Kenya.

5.0.0.1 References

1. Andersen, R. M. (1995). Revisiting the behavioral model and access to medical care: does it matter? Journal of Health and Social Behavior, 36(1), 1–10.

2. Boateng, D., Awunyor-Vitor, D., & Kumi-Kyereme, A. (2014). Barriers to Antenatal Care Use in Sub-Saharan Africa: A Meta-Synthesis. Journal of Health Care for the Poor and Underserved, 25(2), 464–476.

3. Kenya National Bureau of Statistics (KNBS) and ICF. (2023). Kenya Demographic and Health Survey 2022. KNBS and ICF.

4. Kitui, J., Lewis, S., & Davey, G. (2013). Factors influencing place of delivery for women in Kenya: an analysis of the Kenya demographic and health survey, 2008/2009. BMC Pregnancy and Childbirth, 13, 40.

5. Mohamed, M. A., Mwangi, M., & Kiptoo, S. (2022). Determinants of maternal health service utilization among Somali women in Kenya: a qualitative study. BMJ Open, 12(3), e058412.

6. Story, W. T., & Burgard, S. A. (2012). Couples’ reports of household decision-making and the utilization of maternal health services in Bangladesh. Social Science & Medicine, 75(12), 2403–2411.

7. Ver Hoef, J. M., & Boveng, P. L. (2007). Quasi-Poisson vs. negative binomial regression: how should we model overdispersed count data? Ecology, 88(11), 2766-2772.

8. World Health Organization. (2016). WHO recommendations on antenatal care for a positive pregnancy experience. WHO.