Worldwide routine immunisation coverage regressed during the first year of the COVID-19 pandemic

Whilst COVID-19 vaccination strategies continue to receive considerable emphasis worldwide, the extent to which routine immunisation (RI) has been impacted during the first year of the pandemic remains unclear. Understanding the existence, extent, and variations in RI disruptions globally may help inform policy and resource prioritisation as the pandemic continues. We modelled historical, country-specific RI trends using publicly available vaccination coverage data for diphtheria, tetanus and pertussis-containing vaccine first-dose (DTP1) and third-dose (DTP3) from 2000 to 2019. We report a 2·9% (95 %CI: [2·2%; 3·6%]) global decline in DTP3 coverage from an expected 89·2% to a reported 86·3%; and a 2·2% decline in DTP1 coverage (95 %CI: [1·6%; 2·8%]). These declines translate to levels of coverage last observed in 2005, thus suggesting a potential 15-years setback in RI improvements. Further research is required to understand which factors – e.g., health seeking behaviours or non-pharmaceutical interventions – linked to the COVID-19 crisis impacted vaccination coverage.


Introduction
The COVID-19 pandemic has impacted society and public health infrastructures worldwide, influencing mobility [1], access to health services [2], livelihoods and poverty [3]. While COVID-19 vaccination strategies continue to receive considerable emphasis [4,5], the extent to which routine immunisation (RI) has been impacted during the first year of the pandemic remains unclear. Indeed, the World Health Organisation (WHO) pulse surveys reported disruptions in the first half of 2020 [2], and while some later studies suggested a potential recovery [6], recent observations again hinted at global coverage declines [7,8].
RI is estimated to prevent four to five million (M) deaths worldwide every year [9]. As such, there is an urgent need for assessing potential changes in RI coverage, as declines may result in considerable added morbidity and mortality.

Data collection
We investigated changes in RI coverage using two key indicators: diphtheria-tetanus-pertussis first-dose (DTP1) and thirddose (DTP3) vaccine coverage. DTP3 serves as a general marker for immunisation system performance, used by national and global immunisation stakeholders [10]. DTP1 is used as a proxy for inequity -quantifying Zero Dose (ZD) children, those that receive no childhood vaccinations [11]. We compiled vaccination coverage data from the WHO and United Nations Children's Fund (UNICEF) Estimates of National Immunisation Coverage (WUENIC) [12,13] for the last 20 years, using the latest (October 2021) WUENIC data release. Countries were excluded if (a) they did not have complete time series coverage estimates for 2000-2019 inclusive to enable expected coverage modelling (three countries); or (b) they had not yet reported 2020 coverage through WUENIC (16 countries).

Statistical analysis
We used AutoRegressive Integrated Moving Average (ARIMA) modelling [14]

Contents lists available at ScienceDirect
Vaccine j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / v a c c i n e in 2020 for each country and vaccine dose. Prior to investigating differences between expected and observed coverage, historic and predicted time series were assessed and countries were removed from analyses if they met one of three criteria -(1) large volatility in coverage estimates (over 10 percentage points) in the last decade since this may indicate high uncertainty in point estimates, (2) strong influence of most recent coverage estimates (i.e., 2018 or and 2019) contributing to model fitting, corroborated by WUENIC documentation indicating potential anomalous or rare events, and (3) ARIMA-models predicted coverage improvement greater than or equal to five percentage points from WUENICreported 2020 levels, since this may not be programmatically feasible. See Supplementary Text for details on removed countries and contexts by dose. We additionally conducted analyses with no exclusions as a sensitivity study.
After removing countries for which reliable temporal trends could not be assessed, changes in coverage were measured as the difference between the reported and expected coverage for 2020, expressed as percentage values, for the remaining 167 countries per vaccine dose. The significance of global changes was assessed using a t-test against the null hypothesis of the absence of change. Heterogeneities between groups of countries (UN regions or income groups) were tested using linear models with coverage change as a response variable and the corresponding ANOVA. Differences between individual countries were assessed by comparing the 95% confidence intervals derived from the linear models.
For additional validation we conducted the same analysis for measles-containing vaccine first-dose (MCV1) to compare whether similar trends were seen across other vaccine doses and immunisation touchpoints (MCV1 is typically administered at age 9-months compared to six-weeks for DTP1 and 10-weeks for DPT3).
We calculated missed immunisations by combining the estimated changes in coverage with surviving infant population estimates (medium variant births minus infant deaths) of the United Nations World Population Prospects (UNWPP) for 2020 [15].
All analyses were conducted using R [16] and can be reproduced using a publicly available reportfactory including all required data and scripts [17].

Global trends
After excluding countries for which reliable coverage predictions could not be obtained (see Supplementary Text for details), we were able to estimate differences between expected and observed coverage in 2020 for 167 countries for DTP1 and DTP3 -examples shown in Fig. 1.
The exact magnitude of coverage decline was often hard to assess for individual countries due to uncertainties in model predictions ( Fig. 2  We note that results hold for both DTP1 and DTP3 when no countries are excluded from analysis; and that similar trends were seen for MCV1 (an average global coverage decline of 2Á7%, 95 % CI : [2.0%; 3.4%], from an expected 88.3% to a reported 85.6% across 167 countries). See Supplementary Materials for more details on these sensitivity analyses.

Heterogeneities
Patterns of RI coverage significantly varied across United Nations regions ( Fig. 3A As UN regions and income groups are highly correlated (nonparametric Chi-square test: X 2 = 115Á4, p < 10 À5 ), we also tested whether heterogeneities due to one variable (regions or income groups) remained after accounting for the effect of the other one. Interestingly, regional differences remained after accounting for differences in income groups (ANOVA: F = 5Á67, df = 159, p < 2Á7 Â 10 -4 ), but evidence for the converse was weak (ANOVA: F = 2Á67, df = 159, p = 0Á05).  Table 1 details results for the 10 countries with point estimate greatest additional missed DTP3 immunisations in 2020 (see also Fig. 1). Similar trends are seen for ZD children using DTP1 results. Detailed results for all analysed countries can be found in the Supplementary Tables S1 (DTP1), S2 (DTP3) and S3 (MCV1).

Discussion
While the estimated decline in coverage (DTP3: 2Á9% (95 % CI : [2Á2%; 3Á6%]) may seem small, this reduced level of coverage was last observed in the analysed countries in 2005, thus suggesting a potential 15-years setback in RI improvements. Note that global vaccination coverage has remained relatively stagnant over the last decade in many countries, so that the average decline in coverage observed here may often reflect changes between 2019 and 2020. The RI disruption observed in this study suggests there may be greater risk of vaccine-preventable disease outbreaks in the coming years, in the absence of multi-faceted catch-up and adaptation strategies, such as Supplementary Immunisation Activities (SIAs) to reach missed children or periodic intensification of routine immunisation [18].
The estimated changes in RI coverage reported in this study suggest a smaller global decline (approximately 1/3rd the magnitude) than previously found using alternative methodology and data [19]. We believe our findings may be more robust owing to a more comprehensive dataset including data from more countries (167 here vs. 94), plus increased data from the end of 2020 (annual here vs. majority of data from January-September 2020), and the use of WUENIC-reported data (less prone to data quality and completeness issues than administrative data). The observed discrepancies are compatible with a rebound of global RI coverage in late 2020 [6].
An important contribution of this work, beyond the global trends reported, is that it offers a replicable rationale for estimating and comparing the impact of COVID-19 on RI across countries and vaccine programmes, facilitating prioritisation of interventions and resources to those most-affected. Declines in DTP1 coverage indicate increases in the quantity of ZD children in some countriessuggesting that the most vulnerable populations have been strongly impacted by the reductions in RI observed in the first year of the pandemic and reinforced existing inequities in access to healthcare. ZD populations in key ZD ''hotspots" (e.g., India, Pakistan, and Indonesia) are estimated to have increased significantly in 2020, posing a genuine public health threat in the coming years. To alleviate such risks and reduce immunisation inequities, SIAs targeted specifically at these populations should be considered.
Additional research is needed to investigate heterogeneities in RI decline at finer scales and identify subpopulations which may have experienced even greater losses to RI coverage.
RI disruption may be worsened by the acceleration of COVID-19 vaccination campaigns, particularly in low-and middle-income countries where absorption capacity may be challenged [20], potentially competing with RI services. Careful monitoring of the interaction, trade-offs and synergies between RI and COVID-19 vaccinations is essential. Further studies are needed to understand which factors linked to the COVID-19 crisis impacted vaccination coverage, such as changes in health-seeking behaviours or nonpharmaceutical intervention policies, in order to successfully and efficiently address pandemic-associated losses to coverage.

Conclusions
As the COVID-19 pandemic continues to affect healthcare systems globally, maintaining the appropriate balance between access to routine immunisation and pandemic response will be essential to reduce both the direct and indirect mortality and morbidity associated with COVID-19. This research provides a transparent and replicable rationale for estimating gaps in RI coverage across countries, producing an objective measure for missed immunisations and coverage disruptions. As such, it can form a basis for identifying countries most affected by declines in RI coverage and prioritising efforts to alleviate the indirect impact of COVID-19.

Declaration of Competing Interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: It is noted that BE has been employed by the Clinton Health Access Initiative in the Global Vaccines team in the last three years; and is currently employed by Gavi. All research contained in this manuscript was conducted during a postgraduate qualification, outside and independent of employment. Neither facilities, data, nor any other forms of input from the Clinton Health Access Initiative or Gavi, were used in this study. The research and manuscript are independent of the Clinton Health Access Initiative and Gavi, and the findings have not been discussed, reviewed, or endorsed by the Clinton Health Access Initiative, the Gavi Secretariat, or any Alliance members. These funders had no role in the design and conduct of this study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. BE is an MSc student at the London School of Hygiene and Tropical Medicine and received no funding.

Funding
Funders had no role in the design and conduct of this study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data sharing
All analyses were conducted using free software R [16], and can be reproduced using a publicly available reportfactory including all required data and scripts [17] used to produce the results presented in this publication, and available on GitHub at: https:// github.com/bevans249/modelling_covid_impact_RI.