Results
It is estimated that one year of global BCG vaccination at usual levels of coverage (90 %) averts 117,132 (95 % UR: 5049–306,911) TB deaths during the first 15 years of life of that cohort (Table
2). Therefore, without BCG, the point estimate of annual HIV-negative paediatric TB mortality would be estimated to increase from 80,000 to approximately 197,100. The wide uncertainty ranges derive from the wide confidence intervals around the pooled estimate of the rate ratio of TB death in BCG-vaccinated versus BCG-unvaccinated neonates, as a limited number of studies with relatively small sample sizes from different settings contribute to meta-analyses of this parameter. In our main analysis, it was assumed that the rate ratio of TB death by vaccination status and the global estimate of paediatric TB deaths were uncorrelated and were thus sampled independently. As some correlation may exist, a post-hoc analysis assessing the sensitivity of the results to correlation assumptions was conducted. Even in the most extreme scenario of 100 % correlation between these two variables, uncertainty ranges around the annual number of TB deaths averted at 2013 vaccination levels narrowed only marginally to 5781–244,556.
Table 2
Estimated additional paediatric TB deaths in BCG supply shortfall scenarios
6.3 | 16.5 | 84 | 7433 | 320–19,477 | 9.3 |
10 | 26 | 81 | 11,713 | 505–30,691 | 14.6 |
20 | 52 | 72 | 23,426 | 1010–61,382 | 29.3 |
27.6 | 71.8 | 65 | 32,347 | 1394–84,755 | 40.4 |
100 | 260 | 0 | 117,132 | 5049–306,911 | 146.4 |
The hypothetical scenarios explored suggest approximately an additional 11,713 (95 % UR: 505–30,691) paediatric TB deaths in a 1-year cohort in the first 15 years of life per 10 %, or 26 million dose, BCG shortfall. This is a 14.6 % increase in the number of deaths.
A 16.5 million dose (6.3 %) shortfall as reported at the close of 2015, which is reflective of 84 % global coverage, would be estimated to be associated with 7433 (95 % UR: 320–19,477) excess TB deaths in the affected birth cohort in the first 15 years of life (Table
2). A 27.6 % shortfall on global demand was estimated to be associated with 32,347 (95 % UR: 1394–84,755) additional deaths. Therefore, measures to reduce the shortfall from the anticipated 27.6 % to 6.3 % are estimated to have avoided a possible 24,914 (UR: 1074–65,278) additional deaths.
The modelled estimate of the risk of TB death for an unprotected child in the first 15 years of life (u
TB
) was 0.00151 (95 % UR: 0.00064–0.00299), otherwise expressed as 151 (64–299) per 100,000 children in the first 15 years of life.
Discussion
Given historical security in BCG supply, the potential public health impact of shortfalls has not previously been quantified. The static mathematical cohort model presented here indicates that even relatively small BCG supply shortfalls could cause sizeable increases in paediatric mortality in the affected cohorts. It is estimated that current coverage of BCG prevents 117,132 (95 % UR: 5049–306,911) paediatric TB deaths in the first 15 years of life per birth cohort, suggesting that an additional 11,713 (95 % UR: 505–30,691) paediatric deaths could be expected per 10 %, or 26 million dose, annual shortfall in BCG supply. Therefore, in the worst case scenario of a 71.8 million dose shortfall anticipated by UNICEF in 2015, up to an additional 32,347 (95 % UR: 1394–84,755) paediatric TB deaths could have been expected in the affected cohort over the 15-year time horizon. However, measures minimising the 2015 shortfall to 16.5 million reduced the estimated number of additional deaths to 7433 (95 % UR: 320–19,477).
UNICEF procures WHO-prequalified (PQ) BCG from five manufacturers. Adjusting quantities awarded to these manufacturers can meet small-scale changes in demand. However, this supply chain is relatively unresponsive to rapid, or large, changes in supply or demand. In recent years, insufficient supply in the non-UNICEF BCG market has inflated demand for UNICEF-procured PQ vaccine from usually self-procuring countries, such as Egypt, Philippines, Pakistan, and South Africa [
6,
25]. Thus, the UNICEF market is vulnerable to challenges faced by any manufacturer, not just those supplying PQ vaccine. During the same period, manufacturing shortfalls were also experienced by two PQ vaccine suppliers [
4,
6]. These concurrent changes in supply and demand for PQ vaccine far exceeded the responsiveness of the supply system, leading to unmet demand [
6].
Rapid and coordinated action from multiple stakeholders to increase supply and to hold buffer stocks at the global rather than national level to allow for prioritisation of limited stocks to high TB-burden countries reduced projected BCG shortages from potentially 71.8 million down to 16.5 million at the end of 2015 [
5,
6]. The laudable efforts by multiple stakeholders to minimise BCG shortfalls averted an estimated 24,914 (UR: 1074–65,278) additional paediatric deaths. However, due to the remaining shortfall at the close of 2015, it is estimated that a possible 7433 (95 % UR: 320–19,477) additional paediatric TB deaths could be expected in the first 15 years of life of the birth cohort affected in 2015 alone. Therefore, although rapid action in 2015 minimised shortfalls, avoiding a large number of potential additional deaths, the estimated public health impact of even such a relatively small shortfall highlights the importance of ensuring secure manufacturing capacity to meet global BCG demand.
In the estimates presented, the impact of only one year of shortfall was considered, yet recent shortfalls have already extended over several years. Such shortages affecting multiple birth cohorts would result in even greater public health impact. Resolution of the underlying causes takes time; therefore, strategies to pre-emptively avoid shortfalls are essential to avoid unnecessary childhood illness and deaths from TB. New BCG suppliers have been assessed for pre-qualification to increase availability, but future risks include an anticipated 30 % rise in the BCG weighted average price, the sale of one of the UNICEF suppliers making future production plans uncertain, and on-going supply shortages [
5,
6]. It appears unlikely that new manufacturers will enter the market, and there are no known plans for existing suppliers to expand their facilities; therefore, with fixed suppliers and production capacity, mass production of such biologicals will continue to suffer from the small, but very real, risk of production delays.
Reducing vaccine wastage from 20-dose vials could help maximise utilisation of available vaccines. Our model estimates 54.7 % dose wastage, aligning with the WHO/UNICEF-estimated wastage of approximately 50 % [
2]. Wastage is particularly prevalent in smaller health centres, where insufficient neonates present for vaccination during the 6-hour vial use window. Economic modelling has demonstrated that clinics vaccinating fewer than seven infants daily would likely benefit from a 10-dose format [
26]. It is also believed that availability of smaller vials would be associated with increased willingness to open vials to provide timely vaccination and increased coverage. Therefore, production of both 10- and 20-dose vials for infant vaccination could allow tailoring of supply by setting, reducing wastage, increasing coverage and potentially reducing the risk of shortages.
Another, longer-term solution to BCG supply issues could be implementation of a more time- and cost-efficient manufacturing process. Current BCG supplies are manufactured by the labour-intensive, classic surface growth method, but a more modern, cost-efficient, dispersed, liquid fermentation process currently in use for novel BCG replacement candidates could potentially provide a more consistent product if used in BCG production and could help ensure adequate global supplies (Personal communication, Ann Ginsberg, Aeras, January 20th 2016).
Mathematical models are, by definition, a simplification of reality. Some specific limitations of this approach are considered. As paediatric cases are generally minimally infectious [
27], a static cohort model assuming no transmission nor indirect vaccine effects was employed. This limitation may produce an underestimate of the impact of vaccine shortages on TB mortality. The WHO estimate of HIV-negative childhood TB mortality is a key input in this analysis, yet methodology for estimation of paediatric disease burden is evolving, and therefore the size of the estimates may change in coming years. However, such changes do not affect the proportional increase in the number of deaths expected in each scenario, and therefore the impact of shortfalls could be re-estimated in the future using the proportions reported in Table
2. If countries are contributing cases but are excluded from the denominator due to not having a policy of neonatal BCG vaccination, the model could marginally overestimate the effect of the intervention. However, this is expected to be minimal given paediatric TB is generally a sentinel of high transmission, and in such settings BCG is usually offered.
Uncertainty ranges around the modelled estimates are wide due to large confidence intervals around the vaccine efficacy parameter derived from the Abubakar et al. [
16] meta-analysis. Factors contributing to this uncertainty include heterogeneity between study settings and the small number of studies included in the meta-analysis (
n = 5), with relatively low numbers of person years accrued and very few mortality endpoints. All of the studies were conducted between 1933 and 1960. New research could help improve the body of evidence for this parameter, and therefore improve modelled estimates of BCG impact. Placebo controlled trials of BCG are no longer ethical due to lack of equipoise, and retrospective observational designs often suffer from substantial biases, but prospective observational studies with larger sample sizes could potentially contribute to the evidence base for this parameter.
It is possible that some HIV-positive TB mortality may be BCG-preventable, as some healthy HIV-positive neonates with unknown HIV status may be vaccinated, and some children become HIV-infected after receipt of BCG. BCG efficacy against mortality is unknown in these groups, but if BCG were to provide partial protection in these children, the reported impact of BCG shortages would be an underestimate.
Although data are limited, it is possible that BCG protection could wane or the duration of protection could be shorter than the modelled period, and therefore the assumption of 15 years of full protection could produce an overestimate of the impact of shortfalls. Conversely, protection is also likely to continue to prevent deaths beyond 15 years of age, but such longer-term impact was outside the scope of this research question.
There is an increasing body of evidence suggestive of a potentially important non-specific beneficial effect of BCG vaccination against all-cause mortality [
28]. Controlled trials and observational evidence from the USA, UK and West African settings have reported point estimates of 25–60 % reduction in all-cause mortality, with impact particularly pronounced in the neonatal period [
28,
29]. Therefore, the number of all-cause deaths may also increase with BCG shortfalls, but such non-specific vaccine effects are not accounted for in the model.
It is assumed in the model that neonates missed due to 1-year BCG shortfalls would not receive catch-up vaccination, as WHO does not recommend vaccination above 12 months of age and catch-up campaigns may be difficult to implement [
15]. However, for infants under 12 months when supply is restored, catch-up doses may be given to some infants concomitant with other routine Expanded Program on Immunization vaccines, potentially reducing the impact of shortfalls.
UNICEF-reported shortfalls account for demand due to basic country needs and likely some over-procurement to replenish buffer stocks [
6]. The scale of over-procurement is unknown and therefore could not be accounted for, but may cause some overestimate of impact in the 2015 scenarios. There are also insufficient data available to quantify the possible mitigating effects of existing buffer stocks in the model. However, in 2014, 28 countries reported BCG stockouts of at least a month at the national level, of which 86 % reported interruption of BCG vaccination services [
30]; therefore, country reports clearly support the model assumption that global BCG shortages lead to programmatic vaccine shortages.
This global-level model using a single pooled estimate of vaccine efficacy cannot account for likely heterogeneity in the location of stockouts and possible latitudinal variation in vaccine efficacy [
16,
18]. A regional- or country-level model would have been desirable, but estimates of country- or regional-level burden of paediatric TB mortality have not yet been published. When supply shortfalls occur, available UNICEF-procured vaccines are prioritised to high TB burden countries [
25,
31]. This could not be accounted for in the model due to the lack of the required mortality burden estimates to develop a country-level model, and there are also no data available on how BCG was distributed. Differential distribution of stock could therefore not be accounted for in this global-level model, and thus may overestimate the impact of shortfalls if high burden countries experience fewer shortfalls. However, this mitigation mechanism does not guarantee supply, especially when the global demand gap is large, as demonstrated in 2014 when six of the 22 WHO high-burden countries reported an interruption of vaccination services [
30]. This limitation also highlights the importance of UNICEF-led BCG procurement, as left to market forces alone it is likely that many high-burden countries would struggle to procure sufficient vaccine.