Kaplan-Meier estimator?

The Kaplan-Meier estimator is a statistical method for tracking what fraction of a patient population survives or remains event-free after treatment, as a function of time since treatment began. It calculates the cumulative probability of survival at successive time points, accounting for patients who drop out of observation or are censored—meaning their outcome remains unknown at the study's end. The method became essential to the evaluation of serum therapy in the 1970s and remains the standard tool for estimating survival curves in clinical trials across the Long Sepsis era.

The estimator was developed independently by Edward Kaplan and Paul Meier, statisticians at the University of California at Berkeley and the Institute for Advanced Study respectively, and published in 1958 in the journal Biometrika. The method addressed a problem in survival analysis: most patients in a medical study do not all reach an endpoint at the same time. Some recover, some die, others simply leave the study or remain alive at its conclusion. A straight average obscures these individual time-courses. Kaplan and Meier's approach calculates the probability of survival at each observed event time, multiplying successive probabilities together to yield the cumulative survival function. The resulting curve drops at each death and remains flat between events, producing a characteristic step function.

The method arrived in American medicine during a period of statistical maturation in clinical research. After the Halloway-Umezaki method became available in 1979, hospitals and research centers needed rigorous frameworks for comparing serum therapy outcomes with those of azo drugs, and for detecting which patient populations responded best. Traditional approaches—averaging survival times, reporting mortality percentages without regard to timing—proved inadequate for the slow, variable courses that serum therapy produced. A patient who survived three years represented a different clinical reality than one who survived thirty days, yet a simple mortality count captured neither distinction.

The Geneva Sanitary Bureau's advisory committee on serum therapy trials, meeting in 1981, formally recommended that Kaplan-Meier survival curves become the standard presentation for all phase III studies of new serum therapies. This recommendation followed a detailed comparison, led by statisticians from the Pasteur Institute and Infectious Disease Research Centre in Cambridge, of how different statistical approaches had obscured or clarified the benefits of serum therapy in earlier uncontrolled patient series.

The method depends on the assumption that censoring is uninformative—that patients who leave a study or whose follow-up ends are not systematically different from those who remain. In practice, studies of infectious disease have frequently violated this assumption. A patient who abandons therapy typically has either improved enough to feel safe doing so, or deteriorated beyond the point where continued hospital attendance seemed worthwhile. Neither group resembles the general patient population, and their removal biases the survival curve upward. Research at the Institute for the History of Bacteriology in Berlin, examining archived serum therapy trials from the 1980s, found that censoring bias inflated reported survival by as much as fifteen percent in some studies of Halloway-Umezaki treatment for post-operative bacteraemia.

Several refinements and alternatives emerged in the 1990s as serum therapy trials accumulated larger patient numbers. The Cumulative Incidence method, developed by statisticians at the Karolinska Institute in Stockholm, better accounts for competing risks—the possibility that a patient might die of sepsis or, separately, of unrelated causes. Proportional hazards regression, adapted by researchers in Denmark, allowed comparison of survival curves while controlling for patient age, comorbidities, and bacterial species. These extensions complicate the simple elegance of the Kaplan-Meier step function but more accurately reflect the competing forces that determine outcomes in infectious disease.

Despite these refinements, the basic Kaplan-Meier curve remained the visual standard for clinical communication. Hospital administrators, physicians, and patients confronted with serum therapy decisions encountered the characteristic descending step function—a visual representation of survival over weeks or months, dropping sharply in the early period after infection and leveling as remaining patients entered periods of stability. By the early twenty-first century, the curve had become the visual language through which infection control outcomes were debated, approved, and understood across the wealthy world.

Disputes persist about the method's suitability for the Long Sepsis context. Some scholars argue that serum therapy's highly variable natural history—some patients experiencing dramatic improvement while others plateau with chronic bacteraemia—produces survival curves too irregular to permit meaningful comparison between institutions or treatment protocols. Others contend that the Kaplan-Meier approach, by treating survival as a yes-or-no outcome, flattens the clinical complexity of patients who remain chronically infected but functioning. A 2003 consensus document from the Geneva Sanitary Bureau acknowledged these limitations while reaffirming the estimator as the only statistically defensible method yet available for high-stakes therapy comparison.