2.1 Cost of illness (vaccine-preventable diseases)

Estimates of vaccine-preventable child deaths from a 6-month suspension of immunization (Abbas et al., Reference Abbas, Procter, van Zandvoort, Clark, Funk, Mengistu and Hogan2020) were used to estimate treatment costs, transportation costs, lost caregiver wages, and productivity loss due to death. Since deaths from vaccine-preventable diseases (VPDs) occur 1–5 years after missed vaccinations, costs were discounted from the year of death to the year 2020 at 3 % per year.

Direct costs include treatment and transportation costs from illness. Descriptive statistics of inputs for the approach are provided in Appendix 2 in the Supplementary Material. Treatment costs averted were estimated by multiplying the number of cases by syndrome-specific care-seeking rates (Demographic and Health Survey Program, n.d.). Although only fatal cases were modeled, demand and supply side barriers exacerbated by the COVID-19 pandemic, including fear of contracting the coronavirus, transportation restrictions, and health facility closures (World Health Organization and UNICEF, 2020), may prevent some severe and fatal cases from reaching health facilities. To account for this uncertainty, care-seeking rates were varied in sensitivity analysis. We assumed all care-seeking cases would be admitted to inpatient care. The number of care-seeking cases was multiplied by the syndrome-specific length of stay (Parashar et al., Reference Parashar, Hummelman, Bresee, Miller and Glass2003; Rudan et al., Reference Rudan, Tomaskovic, Boschi-Pinto and Campbell2004; Sinha et al., Reference Sinha, Constenla, Esteban, O’Loughlin, Gomez, de la Hoz, Valenzuela and de Quadros2008; Kim et al., Reference Kim, Lee and Goldie 2010a; Bishai et al., Reference Bishai, Johns, Nair, Nabyonga-Orem, Fiona-Makmot, Simons and Dabbagh2011; Lee et al., Reference Lee, Connor, Kitchen, Bacon, Shah, Brown, Bailey, Laosiritaworn, Burke and Derek2011) and the per diem inpatient health facility cost (Stenberg et al., Reference Stenberg, Lauer, Gkountouras, Fitzpatrick and Stanciole2018). Costs of medications and diagnostics were conservatively estimated to be 25–50 % of the facility fees (de Broucker et al., Reference de Broucker, Sim, Brenzel, Gross, Patenaude and Constenla2020). Transportation costs were estimated by multiplying the cost per round trip to a healthcare facility (Kim et al., Reference Kim, Sweet, Slichter and Goldie 2010b) by the estimated number of care-seeking cases.

Caregiver wages lost were estimated using minimum wage (U.S. Department of State, 2017; International Labor Organization, 2018) as a proxy for the opportunity cost of caring for a sick child, as some caregivers may not be formally employed or may have the lowest opportunity cost of time in the household. Lost caregiver wages were estimated by multiplying the wage by the number of care-seeking cases and length of stay (Parashar et al., Reference Parashar, Hummelman, Bresee, Miller and Glass2003; Rudan et al., Reference Rudan, Tomaskovic, Boschi-Pinto and Campbell2004; Sinha et al., Reference Sinha, Constenla, Esteban, O’Loughlin, Gomez, de la Hoz, Valenzuela and de Quadros2008; Kim et al., Reference Kim, Lee and Goldie 2010a; Bishai et al., Reference Bishai, Johns, Nair, Nabyonga-Orem, Fiona-Makmot, Simons and Dabbagh2011; Lee et al., Reference Lee, Connor, Kitchen, Bacon, Shah, Brown, Bailey, Laosiritaworn, Burke and Derek2011).

We used the human capital approach to estimate productivity loss due to death. Future productivity loss from vaccine-preventable deaths was estimated for years of life lost between ages 15 and 64. The number of productive years lost was multiplied by the gross domestic product (GDP) per capita in 2018 (The World Bank, 2019) to approximate an individual’s economic contribution to society. Since all deaths from VPDs occurred before reaching the working age, we adjusted the number of deaths by the probability of survival to 15 years and discounted to 2020 at 3 % per year (The United Nations, 2019). Productivity loss was multiplied by the country-specific labor force participation rate to account for individuals who would not have participated in the labor market (International Labor Organization, 2018).

2.2 Cost of illness (COVID-19)

We used a similar approach to estimate the COI associated with COVID-19 deaths attributable to routine immunization activities. Costs associated with COVID-19 deaths were not discounted, because they were assumed to occur in 2020. The care-seeking rate for childhood pneumonia (Demographic and Health Survey Program, n.d.) was applied as a proxy for the proportion of COVID-19 cases that would seek care and varied the rate in sensitivity analysis. All care-seeking cases were assumed to be admitted for inpatient treatment. Evidence measuring COVID-19 treatment costs in inpatient and intensive care unit (ICU) settings was available for Kenya (Barasa & Kairu, Reference Barasa and Kairu2020). These costs were converted to 2018 USD and adjusted to the remaining African countries using the ratio of modeled inpatient facility costs (Stenberg et al., Reference Stenberg, Lauer, Gkountouras, Fitzpatrick and Stanciole2018) in the target countries compared to Kenya. We assumed that all patients would initially receive treatment in a standard inpatient setting for 5 days and half of the patients would be transferred to an intensive care setting for an additional 7 days (Rees et al., Reference Rees, Nightingale, Jafari, Waterlow, Clifford, Carl, Jombert, Procter and Knight2020). In middle- and high-income countries, most COVID-19 deaths occur in an ICU setting (Wang et al., Reference Wang, Hu, Hu, Zhu, Liu, Zhang and Wang2020). However, availability and access to intensive care settings in African countries may be limited especially during the COVID-19 pandemic (Davies et al., Reference Davies, Sweeney, Torres-Rueda, Bozzani, Kitson, Barasa and Procter2020; Ma & Vervoort, Reference Ma and Vervoort2020). To account for this uncertainty, we varied the rate of intensive care transfers in sensitivity analysis (Appendix 3 in the Supplementary Material).

As for VPD cases, transportation costs were calculated by multiplying the number of care-seeking cases by the cost of one round trip to a health facility. Lost caregiver wages were estimated by multiplying the minimum wage by the length of stay.

Productivity loss due to death from COVID-19 was estimated by multiplying GDP per capita by remaining productive years at the age of death. To align with the epidemiologic study, we assumed that COVID-19 deaths may occur in individuals in four different age groups: the vaccinated child (age 1), a sibling of the vaccinated child (age 5), parents of the vaccinated child, and grandparents (Abbas et al., Reference Abbas, Procter, van Zandvoort, Clark, Funk, Mengistu and Hogan2020). To estimate the age at death among parents of the vaccinated child, we used the country-specific average childbearing age; for grandparents, we used twice the average childbearing age (The United Nations, 2017). The number of remaining productive years was calculated for each age group by taking the retirement age minus the age at death, or the retirement age minus age 15 for children. Additionally, for vaccinated children and their siblings, we adjusted the number of predicted COVID-19 deaths by the probability of survival to age 15 (The United Nations, 2017). For each group, remaining productive life years were multiplied by the GDP per capita and discounted to 2020 at 3 % per year. We calculated a weighted average of the productivity loss associated with COVID-19 deaths using the country-specific proportions of COVID-19 deaths at each age. Total productivity loss was estimated by multiplying the weighted average by the number of COVID-19 deaths and the labor force participation rate (International Labor Organization, 2018).

2.3 Value of statistical life year

Value of statistical life (VSL) is a risk assessment metric based on estimates of the average willingness to pay for a small reduction in mortality risk in a population (Robinson et al., Reference Robinson, Hammitt, Cecchini, Chalkidou, Claxton, Cropper and Eozenou 2019a). The marginal willingness to pay is typically multiplied across a population to estimate the value society places on saving one life (Robinson et al., Reference Robinson, Hammitt, Cecchini, Chalkidou, Claxton, Cropper and Eozenou 2019a). Using the standard approach, VSL does not vary based on the age of those whose lives are saved; thus, it does not provide additional insight beyond the benefit-risk ratios that compare the number of VPD deaths to that of COVID-19 deaths. Since our model only includes children, and COVID-19 disproportionately impacts older adults, we applied the VSLY approach. VSLY is derived from VSL and reflects the marginal rate of substitution between gains in life expectancy and other goods and services.

We first estimated VSL using methods outlined in the Reference Case Guidelines for Benefit-Cost Analysis in Global Health and Development (Robinson et al., Reference Robinson, Hammitt, Cecchini, Chalkidou, Claxton, Cropper and Eozenou 2019a). Since few direct estimates of VSL were available for the modeled countries, we applied a value-transfer approach to extrapolate VSL from the USA to African countries. The value-transfer approach adjusts for differences in income using a multiplier of the ratio of GDP per capita in the target country and the USA raised to the income elasticity for health. For this analysis, we assumed an income elasticity of 1.5 as estimated in a recent systematic review (Robinson et al., Reference Robinson, Hammitt and O’Keeffe 2019b).

Deaths caused by the VPDs included in the analysis are estimated to occur within the next 5 years, so we estimated VSL using projected GDP per capita in the year of death (Global Burden of Disease Health Financing Collaborator Network, 2018). We applied the formula below, in which $ i $ indicates the target country, the USA serves as the reference country, $ j $ is the target year, and $ {e}_y $ is the income elasticity. The U.S. VSL was estimated as 160 times GDP per capita (The World Bank, 2019; Robinson et al., Reference Robinson, Hammitt, Cecchini, Chalkidou, Claxton, Cropper and Eozenou 2019a), and was approximately $10 million in 2018. For countries for which this method yields VSLs below the estimated value of future earnings, or 20 times GDP per capita, we imputed 20 times GDP per capita as a minimum value for the VSL (Robinson et al., Reference Robinson, Hammitt, Cecchini, Chalkidou, Claxton, Cropper and Eozenou 2019a).

$$ {\mathrm{VSL}}_{i,j}={\left(\frac{\mathrm{GDP}\hskip0.2em \mathrm{per}\ {\mathrm{capita}}_{i,j}}{\mathrm{GDP}\hskip0.2em \mathrm{per}\ {\mathrm{capita}}_{\mathrm{USA},j}}\right)}^{e_y}\times {\mathrm{VSL}}_{\mathrm{USA},j}. $$

We derived VSLY by dividing VSL by the country-specific average remaining life expectancy, which was approximated by taking half of each country’s undiscounted life expectancy at birth. We then multiplied the VSLY by the expected number of remaining life years at the age of death. For child deaths, this yields a higher value than the standard VSL approach; for older adults, it yields a lower value.

2.4 Benefit-cost ratios

To compute BCRs, we divided economic benefits of sustained immunization by costs of COVID-19 deaths attributable to immunization activities. BCRs using the COI and VSLY approaches were estimated for each country, vaccine, and EPI visit. The primary results are based on the calculation below:

$$ \mathrm{BCR} = \frac{\left(\mathrm{Benefits}\ \mathrm{of}\ \mathrm{lives}\ \mathrm{saved}\hskip0.4em \mathrm{by}\hskip0.4em \mathrm{routine}\ \mathrm{immunization}\right)}{\left(\mathrm{Costs}\ \mathrm{of}\ \mathrm{COVID}-19\hskip0.4em \mathrm{deaths}\ \mathrm{from}\ \mathrm{interaction}\ \mathrm{during}\ \mathrm{immunization}\ \mathrm{visits}\right)}. $$

2.5 Scenario analysis

The primary analysis compares the economic benefits of lives saved from routine immunization to the economic costs of additional COVID-19 deaths due to interaction during immunization visits. Because the full benefits of immunization would only be realized if investment in providing immunization is continued, we conducted scenario analysis to include the cost of routine immunization in the BCR. Routine immunization costs in African countries for the year 2021 were extracted from an analysis of the costs of reaching coverage goals outlined in the Immunization Agenda 2030 (Sriudomporn et al., Reference Sriudomporn, Watts, Sim, Hutubessy and Patenaude2021). Annual routine immunization costs were adjusted to 2020 using a ratio of the target population (United Nations, 2019) in 2020–2021. We divided these costs by 2 to approximate the cost of immunization during the 6-month suspension in the model. Furthermore, providing routine immunization during the COVID-19 pandemic may be more costly with mitigation measures in place. A recent study estimated that providing personal protective equipment, implementing social distancing, providing hazard pay, and implementing other mitigation measures may increase the cost per dose by 23–251 % depending on the setting and volume of the vaccination site (Banks et al., Reference Banks, Portnoy, Moi, Boonstoppel and Resch2021). We estimated BCRs that include the cost of providing routine immunization at 100, 123, and 351 % of the baseline costs in addition to the cost of COVID-19 deaths estimated. The BCR calculation for the scenario analyses is below:

$$ \mathrm{BCR} = \frac{\left(\mathrm{Benefits}\ \mathrm{of}\ \mathrm{lives}\ \mathrm{saved}\hskip0.5em \mathrm{by}\hskip0.3em \mathrm{routine}\ \mathrm{immunization}\right)}{\left(\mathrm{Costs}\ \mathrm{of}\ \mathrm{COVID}-19\hskip0.4em \mathrm{deaths}+\mathrm{Cost}\ \mathrm{of}\ \mathrm{routine}\ \mathrm{immunization}\right)}. $$

2.5. Sensitivity analysis

We selected five key parameters to vary in probabilistic sensitivity analysis: number of deaths from VPDs and COVID-19, care-seeking rates, ICU admittance rate for COVID-19 patients, facility costs, and GDP per capita (Appendix 4 in the Supplementary Material). Sensitivity analysis for VSLY was performed by varying GDP per capita and the number of deaths, as the other inputs do not impact VSLY calculations. We generated 10,000 Monte Carlo simulations to produce 95 % confidence intervals for all estimates.

We used one-way sensitivity analysis to assess the impact each parameter on the results by estimating benefits, costs, and BCRs within two standard deviations of the mean value for each of the five parameters listed above.