Example country selection
We selected six countries representing upper-middle, lower-middle, and low-income contexts; they included priority countries for the Essential Programme on Immunization (EPI) and for the Global Polio Eradication Initiative (GPEI). A country was chosen from each of the six WHO regions. We compared countries across six indicators: percentage coverage of DTP3, number of nurses per thousand total population, percentage of population under 15 years old, human development index, percentage of population living in a rural setting, and percentage of population living in a slum or informal settlement. Selected countries represent different demographics, social structure, overall economic development, and health system strength. See additional file 1 for explanations of indicators, data sources, and 2019 values for other LMIC countries.
Transmission modeling
Burden forecasts were generated using EMOD, an individual-based disease modeling platform (13) that has been reviewed by the COVID-19 Multi-Model Comparison Collaboration (CMCC) (14). Additional details have been included in additional file 1. Simulations were intended to represent SARS-CoV-2 progression in the chosen contexts and used parameter values appropriate for the respective countries.
Each simulated person is assigned to an age cohort according to the demographics of the simulated country. Contact rates between simulated persons in the model are stratified across four routes (school, home, work, and community) and sixteen age groups (5-year age groups up to 75 years old, and one age group for those 75 + years old) using published model estimates (15), and by risk levels (low, medium, and high). Risk levels provide additional variance within age group without altering mean contact rates.
The baseline distancing scenario for each country assumes school closures, reduced work contacts, and restricted community gatherings (contact rates for school, work, and community are reduced to 0%, 50%, and 75%, respectively, from the values in [26]). For all distancing policies involving a reduction in work contacts or school contacts, twenty percent of the reduced contacts were redistributed to the home route to reflect extra time spent in the home. No community contacts were redistributed to the home route.
For each country, the model was used to fit a most-likely R0 value and case reporting rate to match reported case counts during the initial outbreak period; case count data were obtained from the COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University < https://github.com/CSSEGISandData/COVID-19> (16). Summary data are included in Table 1. Additional details on infectivity calibration have been included in additional file 1. The indicated value for R0 in Table 1 does not account for distancing policy and variable susceptibility with age. Contact fractions by route in Table 1 are input parameters from published model estimates (15); the R0 values were estimated in this study.
Table 1
Baseline scenario contact rate fraction by route and estimated base reproductive number (R0)
Country | Contact Rate Fraction by Route | R0 |
Home | School | Work | Community |
Angola | 0.188 | 0.264 | 0.071 | 0.477 | 3.6 |
Ecuador | 0.223 | 0.211 | 0.137 | 0.429 | 3.2 |
Lao PDR | 0.195 | 0.222 | 0.067 | 0.516 | 3.0 |
Nepal | 0.175 | 0.242 | 0.135 | 0.449 | 3.2 |
Pakistan | 0.188 | 0.251 | 0.043 | 0.518 | 3.6 |
Ukraine | 0.256 | 0.112 | 0.277 | 0.356 | 3.0 |
Reduced susceptibility among children is a significant unknown. Several publications (17–22) suggest that the under-15-year-old cohort acquires and transmits SARS-CoV-2 infections at a lower rate than the general population. This model incorporates a reduction in childhood acquisition of about 55% and childhood transmission of 15%, which has a substantial impact on transmission intensity, reducing the total burden and slowing the speed of the outbreak. Details on the effect of reduced childhood susceptibility have been included in additional file 1.
Connectivity and migration between city centers, peri-urban and rural communities is also poorly documented in many LMICs. In our spatial model, we assume a single large population center, with the sizes of the other population centers (when present) distributed exponentially. These other population centers represent more rural locations and have minimum population of 100 simulated persons (total simulation population is 1 million). The percentage of the population in the large population center is equal to the urban fraction of the country; urban fractions are calculated based on the rural fractions, assuming the urban and rural fractions sum to unity. A network of individual mobility between all population centers was based on the distance between and size of population centers. Modeling the outbreak of SARS-CoV-2 using this distributed community connectivity results in a slower growing and extended outbreak.
The ‘urban’ base case is representative of a single major population center without the network of rural locations, while the ‘urban-rural’ base case is representative of a major population center with surrounding rural locations. No simulations examine a rural-only setting. A rural-only setting would consist of a network of small populations without any single major center, and have outcomes dominated by the timing of disease introduction. Both types of base case depicted in Fig. 1 are used when presenting results for this study.
Daily infection trajectories per 100k population for the ‘urban’ base case and for the ‘urban-rural’ base case; in the ‘urban-rural’ base case, the urban fraction of the population is in the largest population center and the remainder of the population is distributed across smaller population centers. The x-axis describes the number of days post SARS-CoV-2 introduction to the community. Baseline scenarios for each country assumes school closures, reduced work contacts, and restricted community gatherings.
Outcomes depicted in Fig. 1 (and throughout) are trajectories of mean behavior based on ensembles of 1000 simulations. Timeseries are depicted with respect to ‘days-post-introduction’; SARS-CoV-2 introduction to the community occurs at day-zero on this axis. No adaptive distancing policy is included in these scenarios. For instance, an outbreak as acute as depicted for the Ukraine setting would be expected to result in significant self-modification of behavior, which was not included or examined in this study.
These scenarios are illustrative of a wide range of potential outcomes, principally depending on the level of urbanization and shape of the population pyramid. Countries were selected as archetype contexts; the range of this variation is depicted in Fig. 2 for all LMICs in the six WHO regions.
Variation in population fraction younger than 15yrs and rural population fraction in low- and middle-income countries for all six WHO regions. Representative country contexts examined in this study are annotated.
Delivery scenarios
Base case scenarios do not attempt to quantify pre-pandemic levels of routine immunization services or reductions in those services that occurred due to social policy responses at the start of the pandemic. All the scenarios described below are incremental to the base case and the impact of each is calculated as the net difference between the delivery scenario described and this base case value.
Fixed-post (FP) SIAs were reflected in the model by adjusting the contact rates among different age cohorts for seven days, to reflect the community coming together to a central location and having some level of social interaction as well as travel. This was represented by a 50% increase in contacts within the under-5 cohort (children), a 50% increase in contacts among individuals in the 20-35-year-old cohort (their caregivers), and a 200% increase in contacts between the two groups. This scenario approximates a fixed-post SIA with a single-antigen delivered to children, during which vaccination is provided by a health worker who has brief interactions with each child and caregiver. In these scenarios, the increase in community contacts among and between the target population and their caregivers is the primary cause of additional SARS-CoV-2 infections.
House-to-house (H2H) SIAs were reflected in the model by adjusting the interaction rates between health workers and the general population, to reflect the process by which a vaccinator moves from one house to the next administering vaccinations. No changes were made to general community contacts rates with each other. This implies that the children and their caregivers continue to abide by general distancing practices and no additional travel would be required.
To reflect house-to-house SIAs in mixed urban-rural scenarios, health workers were moved from the urban center to smaller communities. For these scenarios, vaccination outreach incorporates many more occurrences of long-distance travel than are present in the baseline mobility structure. In these house-to-house SIAs in rural locations, the potential for health workers to unintentionally introduce SARS-CoV-2 into communities not currently undergoing transmission is the primary cause of additional infections, which then cascade to further infections in the community.
In a typical measles SIA, a vaccinator is expected to deliver between 100–150 vaccinations per day in urban settings and 75–100 per day in more rural areas (23). All scenario results report SARS-CoV-2 infections using a per-population basis (e.g. per-100k). Outcomes for fixed-post SIAs incorporate a fractional increase in the number of contacts among the target population and care givers, which accounts for the difference in target population sizes between the contexts. Outcomes for house-to-house SIAs account for the difference in target population sizes by scaling the number of health workers used by the size of the target population. SIA durations and frequency were not varied based on context; however, variations in timing independent of context were examined for sensitivity purposes.
Routine outreach was implemented similarly to a fixed-post SIA, with both children and adults experiencing a 20% increase in intra-community contact rates. However, in the case of outreach, the health worker was expected to interact with both adults and children, since outreach events are intended to serve a wider population. They are also held periodically and consistently; for these purposes we assumed three days per month for a period of three months, with a cumulative 15% of the target population receiving health services.
Sensitivity analyses
Timing of the vaccination delivery, including scenarios where the delivery occurred prior to peak, near the peak, or after the peak of the SARS-CoV-2 outbreak, was examined as part of the sensitivity analysis.
Impact of infections within health worker populations were varied by simultaneously adjusting the acquisition and transmission rates of health workers; these variations were intended to represent the application of infection prevention and control (IPC) measures. Modifications affected both 1) the acquisition of SARS-CoV2 by the health worker if susceptible and 2) the transmission of SARS-CoV-2 by the health worker if infectious. Levels examined include relative acquisition and transmission rates (RATRs) of 20x, 15x, 10x, 5x, and 1x; levels are for health workers with respect to non-health worker individuals of similar ages. It is likely HW acquisition and transmission is asymmetric; onward transmission may be more strongly moderated than acquisition based on factors both controllable by health workers (e.g., mask wearing) and structural (e.g., occupational risk). Symmetry has been assumed in this model for simplicity.
The health worker cohort persists for the entire duration of the simulation, the RATRs in this cohort do not change during the vaccine delivery scenarios. Health worker contact patterns did not follow the age structured matrix used for other groups, see additional file 1 for details. Relative acquisition and transmission rates should be interpreted as an input that controls the overall attack rate of the HW cohort; they capture the aggregate effect of IPC measures and do not correspond to or attempt to quantify the effect of any specific implementation of IPC (e.g., mask wearing or distancing).