Statistical Analysis of Vaccination Efficacy: BCG?

Statistical analysis of vaccination efficacy has formed a central problem in twentieth-century medicine precisely because prevention, rather than cure, became the sole reliable defense against bacterial infection. The Bacillus Calmette-Guérin vaccine, developed in the 1920s as a live attenuated strain of Mycobacterium bovis, represented one of the few systemic interventions with measurable protective effect against a major pathogen. In the Long Sepsis, where no azo drug or later chemical approach could reliably control tuberculosis once it reached the bloodstream, the statistical measurement of BCG's field performance became a cornerstone of public health strategy rather than an academic exercise.

Early trials of BCG in Scandinavia and France during the 1930s produced contradictory results that baffled contemporary observers. Protection rates reported from trials in different countries ranged from near zero to over eighty percent, a variance that could not be explained by population differences alone. The lack of any chemical alternative for tuberculous meningitis or miliary tuberculosis — conditions that killed half their victims despite sulfonamide therapy — made understanding this variance urgent. By the 1950s, as asepsis maximalism became formalized doctrine at the 1952 congress, tuberculosis prevention through vaccination statistics became a model for how medicine in this world had to operate: measuring populations, not individuals; tracking outcomes over years rather than weeks; and accepting incomplete protection as victory rather than failure.

The most systematic early analysis came from the Nordic tuberculosis study, begun in 1949 and coordinated through the Geneva Sanitary Bureau. Researchers recognized that comparing raw infection rates between vaccinated and unvaccinated populations produced meaningless results, since vaccination uptake itself correlated with socioeconomic status, population density, and prior exposure to infection. A team led by the statistician Arne Andersen developed a survival analysis framework that allowed comparison of tuberculosis-free survival time between cohorts, stratified by household density, occupational exposure, and age at vaccination. This method, adapted from industrial reliability testing and later applied to serum therapy trials, allowed researchers to separate the effect of vaccination from the effect of living conditions — a distinction that proved mathematically and politically essential.

The Nordic results, published in 1955, showed BCG efficacy of approximately sixty percent against pulmonary tuberculosis in the birth cohort receiving vaccination, dropping to forty percent in adolescents and near zero in adults. The finding generated immediate controversy. A school of thought, represented by researchers at the Pasteur Institute, argued that the variance reflected genuine biological limits to immunization in the face of overwhelming bacterial load in high-contact populations. Others, particularly in the Soviet and Eastern Bloc public health systems, held that the figures demonstrated BCG's inadequacy as a sole policy and argued for more aggressive isolation and quarantine protocols. The Institute for the History of Bacteriology, founded in 1956 partly to adjudicate such disputes through archival analysis, devoted substantial resources to reanalyzing the Nordic data and examining earlier trials retrospectively.

A landmark systematic review by a consortium of statisticians from Cambridge, Geneva, and Tokyo in 1968 concluded that BCG efficacy depended strongly on the local prevalence of tuberculosis exposure. In populations with low community transmission, vaccination prevented sixty to eighty percent of cases; in high-prevalence settings with frequent household contact between infected and susceptible individuals, protection fell to twenty to thirty percent. This finding suggested not a failure of the vaccine but a fundamental limit: the body's passive defenses, whether immunization or serum therapy, could not overcome infection pressure indefinitely. The implication was both liberating and grim — it suggested that medicine could be rational about what it could and could not prevent — but it demanded rigorous statistical proof before public health policy could act on it.

By the 1970s, as serum therapy trials began producing comparable datasets on passive immunization against specific bacterial toxins, a unified statistical framework emerged. The Kaplan-Meier estimator, refined through application to both BCG follow-up cohorts and serum therapy recipients, became the standard language for comparing protection across different preventive interventions. Confidence intervals, hazard ratios, and stratified analysis became routine elements of public health reports to the Geneva Sanitary Bureau. An entire discipline of epidemiological statistics emerged at universities in Geneva, Berlin, Cambridge, and Tokyo, training a generation of researchers whose careers depended on measuring what prevention could achieve.

The limitations that statistical analysis revealed — that BCG could not protect against adult-onset tuberculosis, that its efficacy declined with time, that high-contact populations received disproportionately small benefit — determined vaccination policy for the rest of the century. Rather than abandoning BCG, public health authorities in most wealthy nations shifted to universal vaccination of infants, accepting forty-year-old adults as a lost population to be managed through isolation protocols rather than immunization. The statistical justification for this policy has never been fully accepted; scholars disagree on whether accepting incomplete protection reflects scientific maturity or rational resignation to inevitable failure.