This study reveals profound and systematic inequities in COVID-19 vaccination coverage across Kenya, underscoring that a modest national average of 27% masks vastly divergent realities. Urban areas, including Nyeri, achieve substantially higher vaccination rates of up to 46%, while rural and arid regions such as Garissa barely reach about 6%. By merging geospatial clustering methods with machine learning and a composite DI, we show that these disparities are not incidental outcomes of personal decisions. Instead, they reflect entrenched structural disadvantages in healthcare infrastructure, financial inclusion, and environmental conditions. Notably, K-Means clustering identified seven distinct spatial groupings, revealing how an interconnected set of variables, ranging from the density of sleeping rooms in households to bank account ownership and exposure to seasonal floods, shapes the likelihood of receiving a COVID-19 vaccine. In many remote or arid parts of the country, coverage remains below 8%, pointing to significant logistical barriers and systemic neglect. At the same time, regions with higher nitrogen dioxide (NO₂) levels, typically a proxy for urban development, displayed better vaccination rates due to potentially stronger healthcare networks and better resource availability.

These findings resonate with analogous inequities in other LMICs, indicating that financial hardship, fragmented healthcare systems, and challenging environments converge to create what can be described as ’vaccine deserts.’ By making these inequalities visible, our study offers a fresh perspective on the underlying drivers of low vaccination coverage, reinforcing global health discourses that emphasize how place-based disadvantage, rather than mere hesitancy, drives unequal access to life-saving health interventions [15]. The broader implications extend to any resource-constrained setting where the interplay of poverty, environmental shocks, and uneven service provision hinders efforts to control infectious diseases.

Interpretation of our results underscores how socioeconomic constraints, environmental hazards, and uneven health systems jointly dictate vaccine access in Kenya. The Random Forest analysis confirms that financial inclusion, reflected by access to formal banking, is a decisive factor, likely because banked households are better positioned to handle indirect vaccination costs such as transport and lost wages. Meanwhile, high sleeping room density emerges as a compelling proxy for overcrowded living conditions, associating strongly with low uptake in regions already facing pronounced poverty. These structural vulnerabilities are magnified by environmental disruptions: seasonal floods and complex terrain limit the mobility of healthcare workers, disrupt cold-chain maintenance, and render certain communities virtually unreachable during rainy seasons.

Contrary to assumptions that air pollution invariably signals poor health outcomes [16, 17], our data show a more complex relationship, where higher nitrogen dioxide (NO₂) levels correlate with enhanced vaccination coverage, presumably reflecting urban infrastructure advantages that overshadow pollution’s detrimental effects. Similar patterns have been observed in countries like Nigeria [18] and Bangladesh [19], where major cities accumulate health services, specialized healthcare workers, and donor-funded projects, thus achieving superior vaccine outcomes compared with remote, environmentally challenging locales. Together, these insights call attention to the critical role of place-based development strategies that tackle the intersecting domains of infrastructure, climate resilience, and financial empowerment. In the Kenyan context, bridging the gulf between urban hubs and marginalized peripheries necessitates a coordinated response in which multiple sectors, ranging from transport and agriculture to digital finance, converge with public health policy. This integrated vision resonates with Africa’s Agenda 2063, which frames inclusive economic growth and robust health systems as mutually reinforcing pathways toward regional stability [20, 21]. Our study thus adds empirical weight to calls for policies that are no longer framed in single domains but rather align with a broader, equity-focused strategy for advancing public health security.

Translating the findings of our study into actionable policy requires addressing the social, infrastructural, and climatic barriers that perpetuate low vaccination rates in marginalized settings. Scaling up mobile vaccination programs is a vital step, especially for remote northern counties where traditional, facility-based approaches remain insufficient. Mobile teams, synchronized with post-flood recovery periods, can reduce both cost and travel time, as demonstrated in Uganda’s door-to-door campaigns that boosted rural vaccination coverage [22, 23]. However, mobile units alone cannot overcome the underlying issue of financial exclusion, which emerged as a principal predictor in our analyses. Collaborations with fintech platforms such as M-Pesa could allow governments to offer transport stipends or partial wage compensation to individuals missing work to get vaccinated, thereby removing a critical barrier for low-income households [24].

Another immediate opportunity lies in integrating COVID-19 vaccination with Kenya’s well-established childhood immunization programs that already reach around 85% coverage nationwide [3]. By capitalizing on existing cold-chain networks and community trust, health authorities could expedite vaccine delivery to the hardest-hit zones, mirroring Rwanda’s successful alignment of COVID-19 campaigns with child immunization outreach [25, 26]. Beyond these immediate measures, our study underscores the necessity of climate-resilient health infrastructure, particularly in flood-prone and arid counties. Investments in durable roads, drone-based vaccine deliveries, and decentralized storage facilities will not only improve emergency preparedness but also enhance routine health services in historically neglected communities. While these interventions may appear costly, evidence suggests that their long-term returns in saved medical expenditures and improved health outcomes are substantial. Indeed, each percentage increase in coverage may avert significant treatment costs downstream and mitigate the economic fallout of future pandemics. Taken together, these approaches endorse a multisectoral strategy that prioritizes equity, anticipating a broader transformation in how public health systems operate under conditions of climate and financial stress.

A key implementation of our research lies in its multifaceted methodology, integrating spatial autocorrelation measures, K-Means clustering, and Random Forest regression to capture intricate dynamics that traditional analyses might overlook. By linking socioeconomic indicators, environmental hazards, and health infrastructure metrics, we uncovered how overlapping deficits can entrench low vaccination coverage. The DI further synthesizes these domains into a single, interpretable metric, enabling policymakers to identify emergent ’hotspots’ of vulnerability. Real-time monitoring systems that combine routine health data with climate forecasting would offer a dynamic lens on evolving coverage gaps, by demonstrating that a composite DI and spatial clustering techniques can pinpoint pockets of vulnerability, our approach offers a universal toolkit. Health systems worldwide can use similar data-driven, geospatial methods to assess local vaccination challenges and allocate resources more equitably. Furthermore, developing real-time spatial monitoring systems is a key priority for improving vaccine delivery in crisis settings. GIS-based dashboards that link health facility data with vaccine stock levels and coverage metrics could enable public health authorities to monitor coverage gaps and respond proactively to emerging challenges, such as vaccine stockouts or climatic disruptions that impede access. Finally, expanding the DI framework to other LMICs would test its adaptability in regions grappling with similarly complex constraints, thereby creating a global repository of comparative insights. Taken together, these paths underscore the imperative of continuous methodological refinement and cross-sectoral data sharing, which could ultimately strengthen pandemic preparedness beyond the immediate challenges posed by COVID-19.

While our study provides valuable insights into spatial vaccination disparities, several important limitations warrant discussion. First, our reliance on cross-sectional data from the KDHS means we cannot capture temporal changes in vaccination coverage or assess causal relationships. DHS data, while nationally representative, has inherent limitations: surveys are typically conducted every 3-5 years, potentially missing rapid changes in health behaviors during crisis periods like the COVID-19 pandemic. Furthermore, the cluster-based sampling design of DHS may underrepresent highly mobile populations or those living in sparsely populated areas, particularly relevant in Kenya’s arid and semi-arid regions where nomadic communities reside. Another important limitation is our dependence on self-reported vaccination status, which may introduce recall bias and social desirability bias. Respondents might overreport vaccination to appear compliant with public health recommendations or underreport due to stigma or political beliefs.

Without verification against immunization records, we cannot assess the accuracy of these self-reported measures. Similarly, other key variables such as household wealth indicators, and health behaviors rely on self-reporting, potentially affecting the reliability of our socioeconomic analyses.

Likewise, the spatial resolution of our analysis was constrained by DHS confidentiality protocols, which randomly displace cluster coordinates to protect participant privacy. This displacement (up to 2 kilometers for urban clusters and 5 kilometers for rural clusters) may affect the precision of our spatial statistics and environmental exposure estimates. Additionally, we lacked data on several potentially important factors, vaccine supply chains, cold storage capacity, healthcare workforce distribution, and community attitudes toward vaccination. The environmental data used in our analysis, while comprehensive, represents long-term averages rather than real-time conditions during the vaccination campaign. We also lack direct measures of vaccine hesitancy or misinformation, which can significantly impede coverage in certain communities. Furthermore, while remote sensing data on nitrogen dioxide (NO₂) levels and precipitation provided invaluable environmental insights, measurement imprecision may arise from spatial and temporal resolution gaps. K-Means clustering, for its part, hinges on predefined cluster counts, potentially oversimplifying subtler spatial patterns. These gaps point to the need for ongoing inquiry, particularly longitudinal studies that track how coverage responds to policy shifts, supply chain disruptions, or evolving public sentiment. Future research could also incorporate behavioral metrics, such as trust in health institutions, perceived vaccine efficacy, and the influence of social networks, to enrich our understanding of demand-side barriers.