Association Between BCG Policy and COVID19 Infection Rates is Significantly Confounded by Age and is Unlikely to Alter Infection or Mortality Rates

Abstract

Recently a number of publications looked at the association between COVID-19 morbidity and mortality on one hand and countries' policies with respect to BCG vaccination on the other. This connection arises from differences in the rates of infection in countries where BCG vaccination is mandatory compared to countries where mandatory vaccination no longer exists or was never implemented in the first place. In at least 2 preprint publications the authors expressed the view that the "known immunological benefits" of BCG vaccination may be behind the biological mechanism of such observation.
One study accounted for different income levels in different groups. Another study did not attempted to do so, instead exploring the differences between countries where a booster shot is given vs others where no such practice exists (finding no connection).
Both of these studies did not explore other potential confounding factors. Meanwhile the press has focused on these headlines and pushed the narrative that BCG vaccination is causally linked to infection and mortality rates. This poses a serious challenge, demonstrated by the recently initiated clinical trials on BCG vaccination within the COVID19 context.
This study shows that population age is a very significant confounding factor that explains the rates of infections much better and has a solid biology mechanism which explains this correlation. It suggests that BCG vaccination may have little or no causal link to infection rates and advises that any follow up studies should control for several confounding factors, such as population age, ethnicity, rates of certain chronic diseases, time from community spread start date, major public policy decisions and income levels.

Full text

Association Between BCG Policy
and COVID19 Infection Rates is
Significantly Confounded by Age and
is Unlikely to Alter Infection or
Mortality Rates
Stefan Kirov, PhD
Bristol Myers Squibb
Abstract
Recently a number of publications looked at the association between COVID-19 morbidity and
mortality on one hand and countries’ policies with respect to BCG vaccination on the other. This
connection arises from differences in the rates of infection in countries where BCG vaccination is
mandatory compared to countries where mandatory vaccination no longer exists or was never
implemented in the first place. In at least 2 preprint publications the authors expressed the view that the
“known immunological benefits” of BCG vaccination may be behind the biological mechanism of such
observation.
One study accounted for different income levels in different groups. Another study did not attempted to
do so, instead exploring the differences between countries where a booster shot is given vs others
where no such practice exists (finding no connection).
Both of these studies did not explore other potential confounding factors. Meanwhile the press has
focused on these headlines and pushed the narrative that BCG vaccination is causally linked to
infection and mortality rates. This poses a serious challenge, demonstrated by the recently initiated
clinical trials on BCG vaccination within the COVID19 context.
This study shows that population age is a very significant confounding factor that explains the rates of
infections much better and has a solid biology mechanism which explains this correlation. It suggests
that BCG vaccination may have little or no causal link to infection rates and advises that any follow up
studies should control for several confounding factors, such as population age, ethnicity, rates of certain
chronic diseases, time from community spread start date, major public policy decisions and income
levels.

Introduction
BCG vaccine has been associated in multiple studies with effects beyond protection against
tuberculosis, which is the original target of the intervention(1). As such, BCG vaccination has been
shown to enhance the protection provided by H1N1 vaccine(2). Another study observed that prior BCG
vaccination attenuates yellow fever vaccine associated viremia(3). These findings prompted two
independent studies(4,5) in which a strong correlation between the presence of BCG vaccination and a
reduced rate of COVID19 infections and/or death rates was observed.
These observations were also communicated by the press (6–9). While some of these publications
authors contacted additional subject matter experts who cautioned that the studies are not yet peer
reviewed and urged patience(7), there is danger that the observations will be over interpreted and used
for policy decisions. There are currently at least 2 ongoing studies focusing on the effect of BCG
vaccination in the context of COVID19 infections(10) any decisions should be taken only after
conclusive evidence is presented after the end of the trials.
At the same time, some of the established risk factors for COVID19 hospitalization and death such as
age(11) and BMI(12) are easy to evaluate in the same context (per country). Given the extreme urgency
and the potential for serious consequences I decided to explore this matter in greater depth.
Materials and Methods
Data on average population age per country and number of infections was collected from Wikipedia.
BCG policy and income level data was obtained from one of the original studies. BCG and rubella
immunization rates were obtained from WHO website(13).
Country names were cleaned and data was merged in R (Rstudio 1.1.456 on Ubuntu 16 Linux, R version
3.2.3). All scripts and data files are available from https://github.com/kirovsa/covid19-bcg.
As with one of the original studies the countries with population of less than 1M were excluded(4). For
analysis of mortality rates countries with no deaths on record were excluded as these are likely to be in
the very initial phase of the epidemic and will introduce significant noise.
To determine the effect of different factors I used lm function from stats package. Evaluating factors
effects on infections in the presence of random effect was done with lme from nlme package. Log
likelihood was tested with lrtest from lmtest package.
BCG policy and income level were coded according to one of the studies that found the association
with COVID19 infection(4):
BCG Policy
1 = current universal policy
2= used to recommend, not anymore
3 = never had universal policy
2018 FY income level:
Low income (L) -1
Lower middle income (LM)-2
Upper middle income (UM)-3
High income (H)-4
Results
Based on a significant amount of accumulated data, age is a significant factor predicting hospitalization
of COVID19 patients and fatal outcome. Younger people also seem more likely to remain

asymptomatic. Therefore I decided to evaluate a linear regression model that accounts for 3 factors-
BCG policy, income level and median age per country.
While the model as a whole explains very well the differences in infection rates across countries, the
most significant factor was income level, followed by median age. BCG policy was significant but
lagged behind the other factors (Figure 1). However, BCG immunization rates was not significant in
this model at alpha level at 0.05 (p=0.088). The likelihood test did find that the BCG policy had an
effect (p=0.0028) compared to the full model, however this was not true for BCG vaccination rates
(p=0.08). If there is a causal link between BCG vaccination and COVID19 infection rates one would
expect this association to hold or even get stronger, something I did not find evidence for.
The Pearson correlation between median age and infection rates was also much higher at R=0.774 than
the reported correlation between the BCG policy and the infection rates at R=0.521 or the reported
correlation between start date of BCG vaccination and infection rates (R=0.21).
The correlation between number of cases per million people with the median age in a country does not
change substantially between different policy categories (Figure 2A), though there was some separation
between categories 1 and 3. This can only be evaluated for countries with high rates of infection and
also higher median age. When the BCG immunization rates were used instead of the policy there was
no association (Figure 2B).
I also explored potential connection between countries with higher rubella immunization rates vs those
with lower rates (separated in categories by 50% threshold) and COVID19 infections. While this
variable on its own showed significant association (p<0.0001) with the observed infection rates per
country, it appeared that the effect is the opposite of what would be expected (Figure 2C) with
countries with low immunization rates scoring better in terms of infection rate. After the inclusion of
other factors such as median age and income level this association was not significant at alpha=0.05
(p=0.056).
Since income levels are unlikely to drive infection rates I decided to compare the performance of
median age and BCG policy. The data showed that median age explains the variance in the number of
COVID19 cases better than the BCG policy either with or without income level adjustment (Figure 3).
The median age explained 60% of the variability vs 30% for BCG policy. In a mixed model where
income levels are considered a random factor median age again appears to be more important than
BCG policy (Figure 4). BCG rates were again non-significant at p=0.0798.
Next, I looked at the median age distribution in different income levels and BCG policy categories
(Figure 5). There was a strong association between median age and BCG policy with or without
income level adjustment (p<0.0001). The same is true for median age and income level (p<0.0001).
I also explored associations with mortality rates (Figure 6). Again, there was demonstrably better
correlation between median age and mortality rates (R=0.653) compared to the correlation with start
date of BCG vaccination policy reported in one of the studies (R=0.54)(4).
BMI was another strong confounding factor in the context of mortality rates(Figure 7). Countries with
normal BMI were without exception in the policy category 1 (mandatory BCG vaccination). In
addition, death rates were substantially higher in countries with high BMI (p<0.0001).
Discussion
While observational studies are a valid and useful tool, there are also serious obstacles interpreting the
data correctly(14). In the specific case of the correlation between BCG vaccination policy and
COVID19 outcomes it is clear that important confounding factors may have been missed. An excellent
outline of these obstacles was given by Emily MacLean(15). In addition to missing hidden factors, the
critique in the blog goes further to challenge the biological plausibility of the BCG vaccine-COVID19
connection. It seems this is a very reasonable concern, given that the only established connection

between BCG vaccination and protection against viral infections seems to be within the scope of actual
anti-viral vaccines(2,3). On the other hand, the biological rational for causal link between age and
COVID19 morbidity and mortality seems a lot more straightforward; if we follow the Occam’s razor
we should prioritize this link over BCG vaccination. I also want to emphasize that the association
observed in this work between infection rates and rubella immunization are almost certainly spurious.
The arguments so far is that early childhood vaccinations might be protective, which is the opposite of
our observation. Prior preclinical research(16) that was done during one of the previous coronavirus
crisis shows clearly that childhood vaccinations are unlikely to drive different outcomes of COVID19
infections.
This is further enhanced by the data presented in this study. I need to emphasize that I have not
included a number of other potentially confounding factors such as blood pressure, public policy
(mandatory travel restrictions, use of masks, etc.) or time from first infection (start of community
spread). Finally, any conclusive study will need to address the disagreement between BCG policy and
actual BCG vaccination rates with the first still contributing to the regression model, whereas the
second did not.
References
1. Dockrell HM, Smith SG. What Have We Learnt about BCG Vaccination in the Last 20 Years? Front
Immunol [Internet]. 2017 Sep 13 [cited 2020 Apr 3];8. Available from:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5601272/
2. Leentjens J, Kox M, Stokman R, Gerretsen J, Diavatopoulos DA, van Crevel R, et al. BCG
Vaccination Enhances the Immunogenicity of Subsequent Influenza Vaccination in Healthy
Volunteers: A Randomized, Placebo-Controlled Pilot Study. J Infect Dis. 2015 Dec
15;212(12):1930–8.
3. Arts RJW, Moorlag SJCFM, Novakovic B, Li Y, Wang S-Y, Oosting M, et al. BCG Vaccination
Protects against Experimental Viral Infection in Humans through the Induction of Cytokines
Associated with Trained Immunity. Cell Host Microbe. 2018 Jan;23(1):89-100.e5.
4. (2) (PDF) Correlation between universal BCG vaccination policy and reduced morbidity and
mortality for COVID-19: an epidemiological study [Internet]. ResearchGate. [cited 2020 Apr 3].
Available from:
https://www.researchgate.net/publication/340263333_Correlation_between_universal_BCG_vaccin
ation_policy_and_reduced_morbidity_and_mortality_for_COVID-19_an_epidemiological_study
5. (PDF) BCG vaccination may be protective against Covid-19 [Internet]. ResearchGate. [cited 2020
Apr 3]. Available from:
https://www.researchgate.net/publication/340224580_BCG_vaccination_may_be_protective_again
st_Covid-19
6. Explainer: How an old tuberculosis vaccine might help fight the new coronavirus - Reuters
[Internet]. [cited 2020 Apr 2]. Available from: https://www.reuters.com/article/us-health-
coronavirus-tbvaccine-explaine/explainer-how-an-old-tuberculosis-vaccine-might-help-fight-the-
new-coronavirus-idUSKBN21K372

7. Coronavirus deaths are fewer in countries that mandate TB vaccine | Fortune [Internet]. [cited 2020
Apr 2]. Available from: https://fortune.com/2020/04/02/coronavirus-vaccine-tb-deaths/
8. BCG vaccine: US scientists link BCG vaccination with fewer Covid-19 cases, Indian scientists
hopeful but cautious - The Economic Times [Internet]. [cited 2020 Apr 2]. Available from:
https://economictimes.indiatimes.com/news/science/us-scientists-link-bcg-vaccination-with-fewer-
covid-19-cases-indian-scientists-hopeful-but-cautious/articleshow/74931591.cms
9. Rabin RC. Can an Old Vaccine Stop the New Coronavirus? The New York Times [Internet]. 2020
Apr 3 [cited 2020 Apr 6]; Available from:
https://www.nytimes.com/2020/04/03/health/coronavirus-bcg-vaccine.html
10. Australian researchers to trial BCG vaccine for Covid-19 [Internet]. [cited 2020 Apr 2]. Available
from: https://www.clinicaltrialsarena.com/news/australia-bcg-vaccine-trial-covid-19/
11. Zhou F, Yu T, Du R, Fan G, Liu Y, Liu Z, et al. Clinical course and risk factors for mortality of
adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. The Lancet. 2020
Mar 28;395(10229):1054–62.
12. Zeng L, Li J, Liao M, Hua R, Huang P, Zhang M, et al. Risk assessment of progression to severe
conditions for patients with COVID-19 pneumonia: a single-center retrospective study. medRxiv.
2020 Mar 30;2020.03.25.20043166.
13. Vaccination and Immunization Statistics [Internet]. UNICEF DATA. [cited 2020 Apr 4]. Available
from: https://data.unicef.org/topic/child-health/immunization/
14. Nørgaard M, Ehrenstein V, Vandenbroucke JP. Confounding in observational studies based on large
health care databases: problems and potential solutions – a primer for the clinician. Clin Epidemiol.
2017 Mar 28;9:185–93.
15. Community NRM. Universal BCG vaccination and protection against COVID-19: critique of an
ecological study [Internet]. Nature Research Microbiology Community. 2020 [cited 2020 Apr 3].
Available from: http://naturemicrobiologycommunity.nature.com/users/36050-emily-
maclean/posts/64892-universal-bcg-vaccination-and-protection-against-covid-19-critique-of-an-
ecological-study
16. Yu Y, Jin H, Chen Z, Yu QL, Ma YJ, Sun XL, et al. Children’s vaccines do not induce cross
reactivity against SARS‐CoV. J Clin Pathol. 2007 Feb;60(2):208–11.

Figures and tables
Figure 1. Linear regression model exploring policy, age and income level effect on infection rates
Call:
lm(formula = log(CasesPerM) ~ Median + Policy + IncomeLevel,
data = covid.stage2)
Residuals:
Min 1Q Median 3Q Max
-3.3978 -0.7759 -0.0137 0.9219 2.9194
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -1.16279 0.48224 -2.411 0.01738 *
Median 0.07980 0.01968 4.054 8.86e-05 ***
Policy2 1.06411 0.38036 2.798 0.00598 **
Policy3 1.48048 0.58576 2.527 0.01276 *
IncomeLevel2 0.56466 0.39583 1.427 0.15625
IncomeLevel3 2.25019 0.43483 5.175 8.99e-07 ***
IncomeLevel4 3.14738 0.54213 5.806 5.12e-08 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.222 on 123 degrees of freedom
Multiple R-squared: 0.7463, Adjusted R-squared: 0.734
F-statistic: 60.32 on 6 and 123 DF, p-value: < 2.2e-16
Call:
lm(formula = log(CasesPerM) ~ Median + BCG + IncomeLevel, data = covid.stage2)
Residuals:
Min 1Q Median 3Q Max
-3.4935 -0.7091 0.0879 0.9235 2.9354
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.788849 0.594703 -1.326 0.1871
Median 0.083537 0.020396 4.096 7.54e-05 ***
BCG -0.005930 0.003453 -1.717 0.0884 .
IncomeLevel2 0.609350 0.411892 1.479 0.1416
IncomeLevel3 2.334206 0.455878 5.120 1.13e-06 ***
IncomeLevel4 3.365031 0.551903 6.097 1.26e-08 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.258 on 124 degrees of freedom
Multiple R-squared: 0.7288, Adjusted R-squared: 0.7179
F-statistic: 66.64 on 5 and 124 DF, p-value: < 2.2e-16

Figure 2. Relationship between median age per country and infection rates in the context of BCG
policy
A
B

C

Figure 3. Effect of BCG policy or median age and infection rates in the presence and absence of
income level effect
> anova(lm(log(CasesPerM) ~ Policy + IncomeLevel, data= covid.stage2))
Analysis of Variance Table
Response: log(CasesPerM)
Df Sum Sq Mean Sq F value Pr(>F)
Policy 2 219.53 109.763 65.376 < 2.2e-16 ***
IncomeLevel 3 296.29 98.763 58.825 < 2.2e-16 ***
Residuals 124 208.19 1.679
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
> anova(lm(log(CasesPerM) ~ Median + IncomeLevel, data= covid.stage2))
Analysis of Variance Table
Response: log(CasesPerM)
Df Sum Sq Mean Sq F value Pr(>F)
Median 1 433.67 433.67 269.662 < 2.2e-16 ***
IncomeLevel 3 89.31 29.77 18.511 5.329e-10 ***
Residuals 125 201.03 1.61
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
> anova(lm(log(CasesPerM) ~ Policy , data= covid.stage2))
Analysis of Variance Table
Response: log(CasesPerM)
Df Sum Sq Mean Sq F value Pr(>F)
Policy 2 219.53 109.763 27.632 1.089e-10 ***
Residuals 127 504.48 3.972
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
> anova(lm(log(CasesPerM) ~ Median, data= covid.stage2))
Analysis of Variance Table
Response: log(CasesPerM)
Df Sum Sq Mean Sq F value Pr(>F)
Median 1 433.67 433.67 191.19 < 2.2e-16 ***
Residuals 128 290.33 2.27
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Figure 4. Mixed models used to evaluate associations between infection rates, median age, BCG policy
and BCG immunization rates
Fixed effects: log(CasesPerM) ~ Median + Policy
Value Std.Error DF t-value p-value
(Intercept) 0.1470049 0.8923356 123 0.164742 0.8694
Median 0.0864087 0.0193181 123 4.472947 0.0000
Policy2 1.0936249 0.3796067 123 2.880942 0.0047
Policy3 1.5033315 0.5856559 123 2.566920 0.0115
Fixed effects: log(CasesPerM) ~ Median + BCG
Value Std.Error DF t-value p-value
(Intercept) 0.6196404 1.0028649 124 0.617870 0.5378
Median 0.0902190 0.0200332 124 4.503473 0.0000
BCG -0.0060820 0.0034438 124 -1.766091 0.0798

Figure 5. Relationship between median age and BCG policy in the context of income levels

Figure 6.

Figure 7. Association between normalized deaths from COVID19 and BMI (high>25) in the context of
BCG policy
Funding: None
Note: The opinions expressed in this paper are personal and do not represent in any way Bristol Myers
Squibb. No Bristol Myers Squibb resources were used to generate results or prepare this publication.
Acknowledgment: I would like to thank Max Lau, PhD for critically reading this work and Kamen
Kirov for editing the text.