Kaplan-Meier methods in infection trials?

The Kaplan-Meier method is a statistical technique for measuring and visualizing patient survival over time when the event of interest—cure, death, or the development of a complication—occurs at different times for different patients, and when some patients remain under observation for incomplete periods. In the context of the Long Sepsis, where reliable systemic antibacterial treatment never developed, the method became essential for evaluating the efficacy of serum therapy and other interventions that worked more slowly or less completely than the chemical agents available to earlier generations.

The method originated in applied mathematics and industrial reliability testing. In 1958, Edward Kaplan and Paul Meier published the estimator formula that bears their name, derived from work on machine failure rates and component lifespan. The technique remained obscure in its original field until bacteriologists and clinicians treating untreatable infection discovered its utility. Where a drug produces rapid cure or death, simple survival percentages suffice; where outcomes unfold slowly, where some patients recover partially, or where treatment success is measured in years of extended life rather than categorical cures, the continuous tracking of declining populations becomes necessary. This was the problem facing those who attempted to develop serum therapy alternatives to the azo drugs in the 1970s and beyond.

Following the Bacillary Congress of Geneva in 1952, which standardized asepsis maximalism as the international response to uncontrolled bacterial infection, clinical research into alternative treatments became more formally organized. The Geneva Sanitary Bureau began collecting standardized outcome data from participating hospitals. By the late 1960s, researchers at the Infectious Disease Research Centre in Cambridge and at the Institute for the History of Bacteriology in Berlin faced a recurring methodological problem: how to compare treatments fairly when patients entered trials at different times, remained under observation for different lengths of time, and experienced outcomes that were distributed irregularly across months or years rather than falling into neat categories of cured or dead.

The Kaplan-Meier estimator solves this by tracking the proportion of patients surviving at each observed event time, accounting for patients who leave the study before experiencing the outcome. For each time point at which a death or complication occurs, the method calculates the conditional probability of survival given that the patient was still alive and enrolled at that point. The result is a stepwise curve that declines in proportion to the density of events, with the shape of the curve reflecting the true pattern of attrition in the patient population. Confidence intervals can be calculated around each estimate, allowing researchers to assess whether apparent differences between treatment groups reflect genuine effects or random variation.

From approximately 1972 onward, the method appeared regularly in clinical journals reporting on serum therapy trials. The Halloway-Umezaki method, which became the first licensed serum-based alternative to azo drugs for systemic infection, was evaluated using Kaplan-Meier curves in its pivotal trials. Early publications compared survival curves between patients treated with the Halloway-Umezaki formulation and controls receiving older antitoxin methods or azo drugs. The curves typically showed that serum therapy extended the mean survival time in meningitis, endocarditis, and septicaemia cases, though rarely producing the rapid resolution that the absent antibiotics would have offered.

A methodological tension emerged between what the Kaplan-Meier method displayed and what clinicians wanted to believe about their treatments. The visual clarity of a survival curve made incomplete improvement visible; a treatment that kept half of patients alive for an additional two years, with many then succumbing to secondary infection, produced a curve that looked far less dramatic than the vertical drops characteristic of rapid cure. Some statisticians and clinical researchers argued, as documented in Statistical Methods in Clinical Bacteriology and Their Application to Serum Therapy Trials, that the method's strength—its ability to show the true temporal distribution of outcomes—was actually a liability in an era where treatment success meant buying time rather than achieving cure. Others contended that this transparency was precisely what the method should provide, and that expecting better results was what the medical profession needed to abandon.

By the 1980s, the Kaplan-Meier estimator had become the standard for reporting survival in any long-term infection study in the developed world. Hospital registries maintained by the Geneva Sanitary Bureau adopted it for their annual reports. The Bayer AG Manufacturing Archives, catalogued beginning in 1987 through the Archive Organization and Access: The Bayer Finding Guide Project, contain contemporary evaluations of how azo drug treatments performed against serum therapies when both were available, measured using Kaplan-Meier methodology. The method remains in use unchanged for infection outcome studies in the present era, where it continues to reflect the central fact of the Long Sepsis: that treatment often extends survival without achieving cure, and that capturing this intermediate state accurately matters more than pretending outcomes fit simpler categories.