The study aimed to evaluate the South African COVID-19 vaccination campaign using a quantitative ecological time-series approach. The analysis examined rollout speed, geographic equity in vaccination coverage and temporal associations between vaccination coverage and mortality trends using the Logic Model and the RE-AIM implementation framework to interpret implementation performance across the country.^10,11

Research questions: How rapidly and equitably was the South African COVID-19 vaccination implemented, and what temporal association between vaccination coverage and mortality trends can be observed when using the RE-AIM and Logic Model implementation frameworks?

Secondary questions

By integrating national vaccination, geographic coverage and mortality surveillance with the Logic Model and the RE-AIM implementation framework, this study provides a framework-based evaluation of the South African COVID-19 vaccination campaign and contributes implementation evidence on how large-scale public interventions can be assessed in terms of speed, equity and health system performance in resource-constrained settings.

The study employed a retrospective ecological time-series design using aggregated national- and municipal-level data to evaluate the implementation of the COVID-19 vaccination campaign in South Africa. The analysis covered the period from February 2021 to February 2024 and examined the speed of the vaccination rollout, geographic equity in coverage and mortality trends. As an ecological study, all analyses were conducted at population level, without individual linkage, and findings are therefore interpreted as non-causal associations.

The study examined vaccination uptake over time, geographic distribution of vaccination coverage across the country and the mortality patterns that were observed during the pandemic period. The primary outcomes included: (1) vaccination uptake rates of the population over time, (2) equity of access as shown by the geographic coverage, (3) excess and all-cause mortality trends and (4) temporal associations between vaccination coverage and mortality.

Two complementary frameworks guided the implementation evaluation. The Logic Model was used to map the vaccination programme across inputs, activities, outputs, outcomes and impact, providing a structured representation of implementation processes and intended effects.¹⁰ The model has two sides: process and outcome. The process component described the resources and infrastructure that supported the vaccination campaign, the programme activities such as the vaccine delivery and the outputs, including vaccination coverage. The outcome component examined short and medium-term effects and shows the results the programme aims for, which can occur within a short period or over a longer time. Figure 1 illustrates the Logic Model’s relationship of activities throughout the vaccination campaign.

The RE-AIM framework was then applied to evaluate the evidence-based intervention.¹¹ To operationalise the RE-AIM framework in this study, measurable indicators were defined for each domain using the available national datasets, as summarised in Figure 2. Reach was assessed using national vaccination coverage. Adoption was evaluated through geographic comparisons of vaccination coverage. Implementation was assessed through vaccination throughput over time. Effectiveness was examined using national mortality surveillance data during the pandemic period. Maintenance was evaluated through sustained vaccination uptake over the 3-year study period. These indicators enabled the application of the RE-AIM framework to population-level outcomes using aggregated national data. For this research, the framework enabled a structured evaluation of how widely the COVID-19 vaccination campaign has reached the population, how effectively it was implemented and whether it was sustained over time.

The study population comprised all South African residents aged 12 years and older who were eligible for COVID-19 vaccination during the study period. Eligibility expanded during Phase II of the rollout (September 2021) to include individuals aged 12–17 years, resulting in a final eligible population of approximately 46 million.

For the purposes of this study, key terms were defined. Vaccination coverage refers to the proportion of the eligible population that received at least one dose of the COVID-19 vaccine. Vaccination uptake refers to the number of vaccines administered over time. Reach, as defined within the RE-AIM framework, refers to the extent to which the target population was exposed to or received the intervention, operationalised in this study using vaccination coverage across geographic areas. These terms are consistently used throughout the manuscript to distinguish between vaccination volume, population coverage and implementation reach.

The campaign data included vaccinations for the South African population, 12 years and over. Only vaccination data that were captured on the live EVDS were included. Vaccination data covered the period from 17 February 2021 to 29 February 2024, while mortality data covered the period from 2015 to 2023, allowing for the construction of the pre-pandemic baseline.

Data points not captured in the EVDS were excluded from the analysis, and analyses were restricted to recorded and validated entries in the national datasets. The dataset excludes unreported deaths. The study excludes individuals less than 12 years of age.

Secondary data were obtained from publicly available sources. Population estimates (2015–2023) were sourced from the World Bank Group and cross-verified by Worldometer.^12,13 Vaccination data were obtained from the National Department of Health (NDoH) of South Africa, covering the period from 17 February 2021 to 29 February 2024.¹⁴ Vaccination data were collated over 3 years, from the start of the vaccination campaign period, defined as 17 February 2021, to the end of February 2024, which corresponds to the national vaccine rollout following programme initiation. Mortality data were obtained from the South African Medical Research Council (SAMRC), including observed and expected deaths, from 2015 to 2023.¹⁵ Confirmed COVID-19 mortality data were obtained from Our World in Data to support contextual interpretation of mortality trends during the pandemic.¹⁶ The pre-vaccination baseline period was defined as 01 March 2020 – 16 February 2021, prior to the introduction of the COVID-19 vaccine in South Africa.

Analytical dataset: The dataset comprised weekly vaccination counts, municipal vaccination coverage and national mortality data spanning February 2021 – February 2024. Mortality baseline estimates were derived from weekly mortality surveillance data for the period 2015–2019.

Data preparation: Datasets were reviewed for completeness, consistency and alignment across the time periods. Vaccination and mortality data were harmonised using epidemiological weeks to ensure temporal comparability. All variables were standardised to ensure consistency in units of measurement and population denominators across the datasets. Vaccination data were aggregated by epidemiological week and converted to cumulative vaccination coverage, allowing linkage of vaccination coverage and mortality trends over time. Mortality datasets were organised into weekly observed deaths, and expected deaths were calculated from the 2015 to 2019 baseline period.

Excess mortality was calculated as: Excess deaths = observed deaths – expected deaths.

These calculations enabled the identification of deviations from expected mortality during the pandemic period. No imputation of missing data was performed. Data points were assessed and retained only if consistent with overall reporting trends, as the datasets were derived from national surveillance systems. Data processing steps were conducted systematically to enable reproducibility of the analysis using publicly available datasets.

This article followed all ethical standards for research without direct contact with human or animal subjects.

The population estimates indicated that South Africa had approximately 61.5 million individuals in 2021, of whom 46 040 999 million were eligible for the COVID-19 vaccination once the 12–17 years age group was included in the campaign. These figures formed the denominator for evaluating the vaccination coverage (Figure 3).

The national vaccination campaign began on 17 February 2021, with a phased-in approach, prioritising healthcare workers and the high-risk population groups. During Phase I, approximately 39% of the targeted healthcare workforce were vaccinated within the designated time frame. During Phase II, the eligibility was expanded to older adults and essential workers, and the campaign achieved approximately 89% of its population target. Phase III further expanded eligibility to all adults. Approximately 16% of the phase target was achieved during the predetermined period (Figure 3).

By February 2022, approximately 43% of the eligible population had received at least one dose of the vaccine. By February 2024, cumulative coverage reached approximately 49.7% of the eligible population. Vaccination coverage varied across provinces, ranging from 42.7% in KwaZulu-Natal to 60.4% in the Free State. These values represent cumulative coverage at the end of the study period. Within the Logic Model framework, these outputs represent the direct implementation results of the vaccination campaign. Within the RE-AIM framework, they reflect the programme reach, defined as the proportion of the target (eligible) population receiving this intervention. Figure 3 summarises key demographic, implementation and vaccination indicators for the national COVID-19 vaccination programme. It includes population, targets across phases and temporal trends in vaccine coverage over 3 years, stratified by age, gender, province and dose.

An analysis of vaccination throughput shows it has declined substantially over time. During the first year of the programme, an average of 54 896 doses were administered per day (20 201 628 doses over 368 days). This decreased to 6599 doses per day in the second year and to 808 doses per day in the third year.

This decline in vaccination throughput within the Logic Model reflects changes to the programme’s activity levels, while within RE-AIM, it reflects variation in implementation fidelity and maintenance over time.

Linear regression analysis demonstrated a statistically significant decline in weekly vaccination uptake over time. The regression model estimated: Vaccinations_t = β₀ + β₁ (Time). The intercept (β₀) was 399 625.16 (p = 2.83 × 10^–26). The slope coefficient for time (β₁) was –3137.32 (95% CI: –3839 to –2507, p < 0.001), indicating a reduction of approximately 3137 vaccinations per epidemiological week following the initial rollout phase. As illustrated in Figure 4, weekly vaccination counts decreased following the initial peak of the rollout. This graphical trend is descriptive and reflects the temporal pattern of uptake. Inferential conclusions regarding the rate of decline are based on the regression analysis rather than the visual representation alone.

Equity is a multidimensional construct encompassing social, economic and structural determinants that were not fully captured in the analysis due to data constraints. Equity was operationalised as geographical equity in vaccination coverage, assessed by municipal classification as a proxy for urban-rural differences. This approach reflects spatial access to vaccination services but does not capture socioeconomic status or individual-level determinants.

To evaluate geographic equity, vaccination coverage was compared between metropolitan municipalities (MM), which represented mostly urban settings, and district municipalities (DM), which represented mostly rural or mixed rural settings. Vaccination coverage varied across provinces: 42.7% (in KwaZulu-Natal) to 60.4% (in the Free State) of the eligible population. Provinces such as the Free State (60.4%), Limpopo (56.3%) and the Western Cape (54.5%) achieved higher coverage levels than other regions (Figure 3).

Using the national population as the denominator, vaccination coverage was significantly higher in metropolitan municipalities than in district municipalities across all years, with a combined urban mean of 0.0091 (95% CI: 0.0032 to 0.0150) compared to a rural mean of 0.0021 (95% CI: 0.0016 to 0.0026), with p < 0.05. Similar patterns were observed when using provincial population denominators.

However, when coverage was calculated using district-level population denominators, no statistically significant differences were observed between metropolitan and district municipalities in any year or in the combined analysis (p > 0.05) (Online Appendix 1).

Within the RE-AIM framework, these results represent differences in programme adoption across geographic settings. Within the Logic Model, they represent the disparities in programme outcomes related to service accessibility.

The SAMRC maintained ongoing surveillance of national mortality by providing weekly reports on excess natural deaths during the COVID-19 pandemic.¹⁵ Historical mortality data between 2015 and 2019 were approximately 528 548 deaths per year, equivalent to approximately 10 164 deaths per week under non-pandemic conditions. During the pandemic period, mortality levels deviated from this baseline. The largest increase in deaths occurred during 2021, with 705 354 deaths recorded. Table 1 presents national population estimates alongside all-cause mortality, predicted deaths, confirmed COVID-19 deaths and calculated excess mortality for the period 2015–2023.

The data illustrate that weekly excess mortality peaked in mid-2021, while vaccination coverage remained below 25% of the eligible adult population. Following the expansion of the vaccination coverage during late 2021 and early 2022, data show that excess mortality began to decline, and mortality levels moved closer to expected baseline values (Table 1, Figure 5). Within the RE-AIM framework, these mortality trends represent programme effectiveness at the population level.

Pearson correlation analysis was conducted to evaluate the association between vaccination coverage and excess mortality. The contemporaneous analysis demonstrated a moderately strong positive correlation between vaccination coverage and excess mortality, with r = 0.6819 (95% CI: 0.5884 to 0.7574) and p < 0.05. Figure 5 illustrates the co-occurring trends in variant waves and mortality over time. While the graphical representation suggests an inverse temporal pattern, it is descriptive in nature and should be interpreted alongside statistical analyses.

A lagged regression analysis examines whether changes in vaccination coverage are associated with subsequent changes in mortality. To account for potential delayed effects of vaccination and to further examine these temporal relationships, linear regression models were estimated using vaccination coverage variables. Lagged analysis showed a decreasing correlation strength with increasing lag duration. Table 2 presents Pearson correlation coefficients assessing the temporal associations between vaccination coverage and excess mortality under different lag conditions.^17,18

The vaccination campaign achieved rapid initial expansion but was not sustained over time. A statistically significant decline in vaccination activity was observed. Linear regression analysis showed that weekly vaccinations decreased by approximately 3173 doses per epidemiological week. This pattern indicates that the campaign’s momentum declined after the early rollout phase and suggests that even though implementation capacity enabled rapid deployment, sustaining the vaccination demand and programme intensity proved challenging.

By February 2024, the cumulative national vaccination coverage reached 49.7% of the eligible population. When interpreted relative to the national implementation targets and phased rollout milestones, including the benchmark of approximately 67% population coverage, the initial observed vaccination throughput during the first year (43.7%) indicates rapid early deployment of the vaccination programme. However, the pace was insufficient, as the subsequent declines in daily vaccination rates in the second (49.2%) and third (49.7%) years did not meet the sustained coverage targets within the projected timeframe. This demonstrates a partial reach of the intervention.

The observed inter-provincial variation in vaccination coverage, ranging from 42.7% to 60.4%, indicates heterogeneity in implementation rollout performance across regions, although the magnitude of variation was moderate. These findings suggest that early rollout success did not translate into sustained population-level coverage.

Geographic equity analysis showed higher vaccination coverage in metropolitan municipalities than in district municipalities when assessed using national and provincial denominators (p < 0.05). However, differences were not statistically significant when district-level population denominators were applied, and absolute differences were modest. This indicates that, while spatial disparities were present, their magnitude was limited, highlighting the importance of interpreting both statistical and practical significance in equity analyses. The use of municipal classification as a proxy for access further indicates that these findings reflect geographic rather than socioeconomic equity. The findings suggest that health system infrastructure, service accessibility and factors such as vaccine hesitancy or demand-related barriers may have influenced the distribution and uptake of vaccination services.

Mortality trends showed that excess mortality peaked during the early epidemic waves of 2021, when vaccination coverage remained quite low and declined following the expansion of vaccination coverage. However, these trends represent population-level temporal patterns and cannot be attributed solely to vaccination. Multiple concurrent factors likely influenced mortality dynamics, including epidemic waves, variant evolution, prior infection and hybrid immunity, improvements in clinical management, and public health non-pharmaceutical interventions such as lockdowns and mask mandates. These factors likely interacted dynamically during the pandemic and may have influenced mortality patterns independently or synergistically.

Statistical analysis identified moderate temporal associations between vaccination coverage and excess mortality trends. Lagged correlation analyses provided additional insight into the temporal relationship between vaccination coverage and excess mortality. Pearson correlation analysis showed a contemporaneous correlation coefficient of r = 0.6819, with decreasing correlation strength observed in lagged analyses. The strongest correlation was observed when vaccination coverage and excess mortality were analysed contemporaneously, while the strength of the association declined when vaccination coverage was shifted forward by 3 weeks and 6 weeks. The reported CIs provide an estimate of the precision of the observed associations and should be considered when interpreting the strength and reliability of the findings.

These findings indicate that the relationship between vaccination coverage and mortality varied over time and was not temporally synchronised in a simple linear manner. Given the ecological design and a lack of adjustment for time-varying confounders, this regression model quantifies statistical associations with the observed time-series data and should be interpreted as non-causal. In addition, these observed findings are based on aggregated national population-level data without individual-level linkage between vaccination coverage, vaccination status and mortality outcomes, limiting causal interpretation. The findings should be interpreted as population-level temporal associations rather than evidence of causal effects.

Collectively, the results demonstrate that the vaccination campaign expanded rapidly but did not sustain sufficient uptake to achieve national coverage targets. At the same time, geographic disparities in coverage and the observed temporal association with mortality trends provide insight into how implementation performance varied and evolved across the health system.

For an intervention to be effective, planning and execution are vital. The Logic Model provided a structured lens to interpret the implementation pathway and assessed the outcomes and impact of the campaign, as it focused on equity, access, health system coordination and improving the public knowledge and their acceptance (Figure 1).

At the level of inputs and activities, South Africa mobilised substantial infrastructure, including public–private partnerships, mass vaccination sites and the EVDS.¹⁹ Operational strategies such as phased prioritisation, mobile vaccination units, outreach programmes and extended service hours facilitated early access and rapid scale-up.^20,21 These measures reflect the resources and operational strategies deployed to support vaccine delivery. However, implementation challenges, including coordination constraints, workforce pressures and limited rural integration, likely influenced the programme performance.

The outputs of the intervention were reflected in vaccination coverage and programme reach. Although the campaign achieved early expansion and vaccinated more than 20 million individuals within the first year, overall national coverage plateaued below 50%. The plateau observed in Phase III suggests that the operational capacity alone was insufficient to sustain demand once the high-risk populations had been vaccinated.

At the level of short- and medium-term outcomes, improvements in geographic access were observed through the integration with primary health care services and outreach programmes. However, spatial variation in access persisted.⁹ Changes in mortality patterns coincided temporally with increased vaccination coverage, but the ecological design of the study limits the ability to attribute these changes directly to vaccination.

The RE-AIM framework was operationalised using measurable indicators aligned with each domain, as summarised in Figure 2. Metrics were used to evaluate the vaccination campaign across reach, effectiveness, adoption and implementation, and it enabled a multidimensional evaluation of the implementation performance (Figure 2):

The observed decline in vaccination uptake over time is consistent with international evidence demonstrating reduced demand for COVID-19 vaccination following the initial rollout phases.^23,24 This pattern reflects the saturation of high-risk populations and the increasing reliance on voluntary uptake among lower-risk groups.

Geographic disparities in vaccination coverage have also been widely reported across countries. The observed urban-rural disparities are consistent with the literature from both high-income and middle-income settings, which demonstrate that urban populations typically achieve higher vaccination coverage than rural populations, reflecting differences in healthcare infrastructure, geographic access and service delivery.²⁵ These structural determinants of healthcare access are particularly relevant in South Africa, where spatial inequities in health system resources have been well documented.²⁶

The mortality trends observed in this study align with broader global evidence, demonstrating that COVID-19 vaccination programmes contributed to reductions in hospitalisation and mortality at the population level.²⁷ However, several additional factors may also have influenced mortality patterns during the pandemic, including improvements in clinical management, changes in variant virulence, prior infection-induced immunity and non-pharmaceutical interventions.²⁸

The findings of this study provide important insights for strengthening the implementation of large-scale public health interventions, particularly in resource-constrained and inequitable health system contexts. The observed trajectory of the COVID-19 vaccination campaign in South Africa highlights the distinction between rapid programme initiation and sustained implementation performance, with implications for future emergency preparedness and health system resilience.

A central implication relates to the sustainability of intervention uptake. While the vaccination campaign achieved rapid early expansion, the subsequent decline in vaccination activity highlights that implementation capacity alone is insufficient to maintain programme momentum. Sustained uptake requires continuous demand-generation strategies that address behavioural, structural and contextual determinants of health service utilisation. Targeted risk communication, community engagement and context-specific interventions to address vaccine hesitancy and perceived risk are essential to maintaining programme effectiveness beyond the initial rollout phase.

Geographic equity remains a critical consideration for public health implementation. The observed differences between metropolitan and district municipalities suggest that structural factors, such as health system infrastructure, service accessibility, and resource distribution, continue to influence the reach of national interventions. Therefore, strengthening equitable access requires targeted investment in non-metropolitan areas, including expansion of mobile outreach services, reinforcement of community-based healthcare delivery and improved integration of services at primary health care level. These strategies are particularly important in settings characterised by persistent spatial and socioeconomic inequalities.

The COVID-19 vaccine was not immediately available at primary health care facilities during the initial rollout. The integration of vaccination services into routine primary health care represents a key implementation strategy for enhancing both equity and sustainability. Embedding preventive interventions within existing care platforms, such as chronic disease management, maternal and child health services, and community-based care, can reduce access barriers, improve continuity of care and support more resilient health system responses during public health emergencies. This approach aligns with broader principles of health system strengthening, emphasising integration, accessibility and continuity of care.

From a systems perspective, the findings highlight the critical role of data-driven decision-making in public health implementation. The EVDS enabled real-time monitoring of vaccination coverage and programme performance. However, the absence of integrated linkage between vaccination data, mortality surveillance and clinical datasets limits the ability to fully evaluate intervention outcomes. Strengthening integrated health information systems to enable interoperability across data sources would support more precise, timely and context-specific policy responses.²⁹ Such systems would allow for the identification of underserved populations, monitoring of geographic disparities and evaluation of intervention effectiveness in real time, thereby enhancing both responsiveness and accountability.

More broadly, the application of the Logic Model and RE-AIM frameworks demonstrates the value of structured implementation science approaches in evaluating complex public health interventions. The integration of these frameworks enables a link between programme inputs, processes and measurable outcomes, facilitating a more comprehensive understanding of implementation performance. Embedding such frameworks within routine programme evaluation can strengthen the translation of empirical findings into actionable policy and improve the design, monitoring and adaptation of future interventions.

Collectively, these findings emphasise that effective public health interventions require not only rapid deployment but also sustained engagement, equitable access, integrated service delivery and robust data systems. Strengthening these dimensions will be critical for improving the implementation of future vaccination programmes and enhancing preparedness for emerging public health threats.

The study has several important limitations. Firstly, the ecological study design, based on aggregated population-level data, precludes causal inference and introduces the potential for ecological fallacy. Associations observed at the population level may not reflect individual-level relationships. Secondly, vaccination coverage and mortality data were not linked at the individual level, limiting the ability to directly assess vaccine effectiveness. Thirdly, the analysis did not adjust for key time-varying confounding factors, including variant dynamics, prior infection and hybrid immunity, changes in clinical management and non-pharmaceutical interventions, all of which may have influenced mortality trends independently of vaccination coverage. Fourthly, the equity analysis was limited to geographic comparisons using municipal classifications as a proxy for urban-rural differences. It also did not include socioeconomic stratification or formal equity indices, thereby providing only a partial assessment of equity. Fifthly, the use of secondary administrative datasets may be subject to reporting delays, data completeness issues and potential inconsistencies across sources. Sixthly, behavioural factors, including vaccine hesitancy, risk perception and public perception, were not measured in this study, yet may have played a significant role in shaping vaccination uptake and observed coverage patterns. Finally, while time-series regression analyses were conducted, more advanced modelling approaches, such as fully specified interrupted time-series analyses or adjustment for autocorrelation, were not applied. The findings should therefore be interpreted as exploratory population-level associations rather than definitive estimates of effect. Despite these limitations, the study provides valuable population-level insights into vaccination rollout dynamics within a real-world public health context.