key: cord-281354-sa27k8o3 authors: Takahashi, Harutaka title: Role of Latent Tuberculosis Infections in Reduced COVID-19 Mortality: Evidence from an Instrumental Variable Method Analysis date: 2020-08-26 journal: Med Hypotheses DOI: 10.1016/j.mehy.2020.110214 sha: doc_id: 281354 cord_uid: sa27k8o3 Since the outbreak of the coronavirus disease 2019 (COVID-19) pandemic, there has been significant interest in the potential protective effect of the Bacillus Calmette-Guerin (BCG) vaccine against COVID-19 mortality. This effect has been attributed to innate immune responses induced by BCG vaccination. However, these studies ignore an important fact: according to World Health Organization estimates, about a quarter of the world's population may have latent tuberculosis infection (LTBI), a condition in which there is no evidence of clinically active tuberculosis but persistent immune responses are stimulated by Mycobacterium tuberculosis antigens. Thus, both LTBI and BCG induce lifelong immunity and may provide immunological protection against COVID-19. In this study, the relationship between LTBI and reduced COVID-19 mortality was analyzed using the instrumental variable method. The results showed with robust statistical support that LTBI was also associated with reduced COVID-19 mortality. Since the outbreak of the coronavirus disease 2019 (COVID-19) pandemic, significant attention has been focused on the relationship between Bacillus Calmette-Guerin (BCG) vaccination and COVID-19 mortality. In particular, there is interest in whether BCG vaccination is associated with a reduction in COVID-19-associated mortality. BCG is the most widespread vaccine against tuberculosis (TB) and also elicits non-specific effects and innate immune memory against non-mycobacterial diseases. A survey of key unpublished and published data regarding the association between BCG vaccination and COVID-19 mortality was conducted, and concluded that there was a lack of evidence to support a protective effect of BCG against COVID-19 [1] . However, such studies ignore the important fact that about one-quarter of the world's population may have latent TB infection (LTBI), a condition in which there is no evidence of clinically active TB but persistent immune responses are stimulated by Mycobacterium tuberculosis antigens. The regional data shown in Table 1 illustrate that the number of LTBIs far surpasses the number of active TB infections. The number of LBT infections clearly surpasses that of TB infections. LTBI also induces lifelong innate immune immunity [2] , [3] and may confer an immunological protective effect against COVID-19. Many countries with a relatively high incidence of TB infection, including Japan, require BCG vaccination during early childhood. Most citizens of these countries also have LTBI, which is highly immunoprotective because of elicited innate immune responses. In fact, TB infection leads to LTBI in 90%-95% of cases, while 5%-10% of individuals develop active TB disease [5] . Therefore, the number of TB infections per hundred thousand individuals can be used as a proxy for the number of LTBIs. Furthermore, M. tuberculosis infection via BCG vaccination can enhance innate immunity. Therefore, citizens of countries with high prevalence of TB infection (high TB burden countries) together with high BCG vaccination rates are considered to have enhanced innate immunity compared with the citizens of lower TB burden countries. This high level of natural immunity is thought to be responsible for the lower COVID-19 mortality rate. The aim of this study was to test the hypothesis that LTBI is associated with reduced COVID-19 mortality. The instrumental variable (IV) method was used to assess causality. All data used in the analysis are publicly available and are described in the appendix. Much discussion has centered around the strong correlation between BCG and COVID-19 mortality. However, correlation does not imply causation, and can sometimes instead reflect spurious relationships. Regression analysis, particularly the IV method, is a statistical method that addresses this problem to assess causality. Care must be taken in using COVID-19 mortality as a dependent variable [6] . This is because COVID-19 mortality is conditionally observed in potentially infected individuals, and can only be detected by testing of symptomatic or asymptomatic individuals. Therefore, the case fatality rate (CFR), defined as the ratio of the number of COVID-19 deaths per million people to the number of COVID-19 infections per million people, is typically used. As explained above, the logarithm of the number of TB infections per 100,000 individuals (lntb10) can be used as a proxy variable for LTBIs. For this regression analysis to be statistically accurate, the explanatory variable X must first be correlated with the error term u (i.e., the covariance of X and u must be zero). This condition clearly does not hold in general: besides LTBI, many other co-occurring factors, such as cultural norms, mitigation efforts, health infrastructure, and urban concentration, may influence this relationship [8] . Therefore, it is possible that X is correlated with such factors excluded in the regression equation, and that X and the error term may be correlated. This would be an example of a "spurious regression". To overcome such a problem, the IV method can be used. An IV is a variable that is strongly correlated with the explanatory variable X but is not correlated or only weakly correlated with the error term. The IVs used here were as follows. Four diagnostic tests were performed to assess whether the estimates were statistically relevant. One test was concerned with the explanatory variables and the other three were concerned with the IVs. The Wu-Hausman test assesses the endogeneity of the explanatory variables. If the null hypothesis is rejected, one can simply use the standard ordinary least squares regression instead of using the IV. The first test of an instrument is Sargan's exogenous test, which assesses whether the right number of IVs are selected and confirms that they are sufficiently uncorrelated with the error term. Finally, it is necessary to perform a "weak IV test" to check if the selected IVs are strongly correlated with the explanatory variables. The instrumental variables used here were bcgindex, region and pop65. Two models were estimated using the IV method: one with three instruments and the other with two instrument (bcgindex and region). Most of the countries with low income levels (annual per capita income less than $825 USD) reported zero deaths attributed to COVID-19 [7] . To avoid underreporting bias in these countries, they were excluded. The total number of countries analyzed was thus 104. The results are shown in Table 2 . The estimates of the Generalized Moment Method (GMM), which is often used as an alternative to the IV method, are also reported. For the diagnostic tests of the two estimation models, Sargan's exogenous test indicated that the selected IVs met the exogenous property. The results of the weak IV test indicated that the IVs were sufficiently and strongly correlated with the explanatory variables. Therefore, all the estimation results presented here were statistically robust. All the coefficients of lntb10 were approximately -0.02, indicating a negative association between LTBI and COVID-19 mortality. Thus, these results lend statistical support to the hypothesis that LTBI can protect against COVID-19 mortality. Because these estimation models were linear-log type, the estimated coefficient of - Does BCG vaccination protect against acute respiratory infections and COVID-19? A rapid review of current evidence TB prevalence correlation to COVID-19 mortality Trained immunity from Mycobacterium spp. Exposure or BCG vaccination and COVID-19 outcomes Is Mycobacterium tuberculosis infection life long? The global burden of latent tuberculosis infection: A reestimation using mathematical modeling Is there evidence that BCG vaccination has non-specific protective effects for COVID-19 infections or is it an illusion created by lack of testing? doi Correlation between universal BCG vaccination policy and reduced morbidity and mortality for COVID-19: an epidemiological study Bauer (2020) Appearent difference in fatalities between Central Europe and East Asia due to SARS-COV-2 and COVID-19: Four hypotheses for possible explanation Data Appendix  Number of COVID-19 infections and deaths per million people as of  Number of TB infections per 100 The BCG Index variable was created to represent the total number of years since 1950 during which BCG vaccination was mandated in a country's immunization schedule. For example, Japan made BCG vaccination mandatory in 1942 and continues to do so today, so Japan's BCG Index is 1. France introduced BCG vaccination in 1950 and discontinued it in  Ratio of the population over 65 years old I would like to thank Masayuki Miyasaka at Osaka University for helpful suggestions on the issue and Edanz Group (https://en-author-services.edanzgroup.com/ac) for editing a draft of this manuscript. This work received no specific funding. The author declares no conflict of interest. I would like to thank Masayuki Miyasaka at Osaka University for helpful suggestions on the issue and Edanz Group (https://en-author-services.edanzgroup.com/ac) for editing a draft of this manuscript. This work received no specific funding. The author declares no conflict of interest.