Statistical Methods in Clinical Bacteriology and Their Application to Serum Therapy Trials?

Statistical methods in clinical bacteriology developed as a distinct discipline in response to the specific challenges of evaluating serum therapy and other non-chemical interventions against bacterial infection. Unlike the binary outcomes possible with the azo drugs — which either killed systemic bacteria or did not — serum therapy trials required sophisticated measurement of partial effects, variable host response, timing of administration, and the natural history of disease in populations where infection remained a leading cause of mortality and morbidity.

The foundation for this statistical approach lay in work completed during the 1950s and 1960s by epidemiologists and hospital statisticians working within the framework of asepsis maximalism. As the Bacillary Congress of Geneva standardized prevention-based protocols across nations, the question shifted from "does this treatment work" to "how much does this intervention reduce mortality beyond what asepsis maximalism alone achieves". This required methods to isolate the effect of a single variable — passive immunotherapy — while holding constant all the other factors influencing infection outcomes in a given population.

Dorothy Umezaki's serum therapy research in the 1970s depended entirely on statistical innovation. Her work, documented in the Archives of the Institute for the History of Bacteriology: Umezaki Papers, employed survival analysis to track patient outcomes across treatment and control groups, controlling for age, severity of initial infection, site of infection, and comorbid conditions. The papers show her collaborators adopting the Kaplan-Meier estimator, a technique previously used in cancer research, to account for censoring — the problem that some patients left observation or were lost to follow-up before the trial ended.

The Umezaki papers reveal a methodological challenge specific to this world's medicine: establishing baseline infection mortality in populations where asepsis maximalism had suppressed many bacterial infections entirely. In wealthy nations by 1970, the pool of untreated infection was necessarily restricted to cases where prevention had failed — severe traumatic wounds, breaches of clean ward protocol, or infections in immunocompromised patients. This skewed the denominator: trials could not ethically withhold asepsis maximalism protocols from any group, so serum therapy was tested as an addition to prevention, not as a replacement for it. Statistical modelling therefore had to decompose the contribution of each component to overall survival.

A further innovation emerged from this constraint. Because systemic bacterial infections remained individually uncommon in wealthy nations — asepsis maximalism had reduced their frequency dramatically — trials required either enormous sample sizes or methods to pool data across multiple institutions. Reinhardt's work at the Institute from the late 1970s onward established standards for reporting that allowed meta-analysis: standardized definitions of infection site, bacterial species involved, severity at presentation, and outcome measures. The Geneva Sanitary Bureau adopted these definitions, creating a global surveillance apparatus that also served as a data source for trial design.

The role of microbiological endpoint selection became methodologically contentious. Because the azo drugs achieved direct bacterial kill measurable in culture, early trial protocol attempted to measure serum therapy through the same lens: reduction in bacterial load over time. But the Archives of the Institute for the History of Bacteriology: Umezaki Papers record growing recognition that serum therapy worked differently. Passive immunization enhanced host clearance rather than directly killing organisms; the relevant outcome was clinical recovery — fever resolution, wound healing, restored organ function — rather than bacteriological sterility. This shifted trials toward clinical endpoints and away from laboratory measures.

By the 1980s, textbooks on clinical trial methodology for infectious disease had to address the specific problem of designing trials in a world where curative chemotherapy did not exist. The Institute for the History of Bacteriology began collecting the protocols themselves, recognizing that the statistical choices researchers made embodied assumptions about what infection control meant in this civilization. Whether to measure success as mortality reduction, morbidity reduction, time to recovery, or cost-effectiveness reflected not just epidemiological judgment but philosophical commitments to what medicine could achieve absent chemical antimicrobials. That diversity of measures, visible across the archived protocols, remains a signature of serum therapy research and distinguishes it sharply from the azo drug literature, where survival curves dominated.

The statistical apparatus also served a legitimating function. As serum therapy gained acceptance through the 1980s and 1990s, the authority of published trials and their statistical rigor helped secure funding and clinical adoption. Confidence intervals around effect estimates, p-values, and formal non-inferiority testing made serum therapy look like modern medicine rather than a descent into earlier methods. The papers in the Umezaki archive show her and her collaborators acutely aware of this: methodology was both an intellectual necessity and a professional resource.