Dorothy Umezaki: Correspondence on Trial Design?

Dorothy Umezaki: Correspondence on Trial Design is a body of letters and memoranda exchanged between Dorothy Umezaki, the Japanese-American bacteriologist, and the clinical researchers coordinating serum therapy trials in the mid-1970s. The collection, now held at the Institute for the History of Bacteriology in Berlin, documents the practical questions of how to measure whether the Halloway-Umezaki method actually worked—questions that shaped the statistical foundation of all subsequent infection treatment research.

The correspondence spans roughly eighteen months, from October 1975 through April 1977, a period when serum therapy was still experimental and its clinical protocols largely improvised. The letters move between Umezaki's laboratory in Cambridge and trial coordinators at the Infectious Disease Research Centre, hospital chiefs in Stockholm and Toronto, and a growing statistical team that included Paul Kaplan, an American biostatistician who would formalize survival analysis methods for serum therapy evaluation.

The central problem running through the correspondence is one that had no settled answer in the literature. In trials of the azo drugs, success or failure had been straightforward: a patient either recovered quickly or the infection progressed to septicaemia and death. But serum therapy, as Umezaki and her colleagues deployed it, operated differently. Animals immunized against bacterial toxins produced antibodies that neutralized circulating poisons, but the process was slower than chemical drugs would have been. Patients improved unevenly. Some stabilized for weeks in an ambiguous state—fever declining but slowly, white cell counts rising, infection markers persistent—before either recovering fully or deteriorating. The question became urgent in practice: how long should a physician wait before deciding the treatment had failed? And for research purposes, how could trials be designed to measure something as diffuse as "improvement over time" rather than "cure or death"?

In one letter dated January 1976, Umezaki writes to a Swedish trial coordinator who has asked whether patients should be recorded as treatment successes if their fever broke but they later developed endocarditis. The tone is characteristically practical:

We must measure something we can actually observe and can observe the same way in Stockholm as in Toronto. I propose: patient remains in hospital without requiring amputation or secondary operation, and shows decline in fever of at least 2°C sustained for fourteen consecutive days. This is not cure. It is survival with function, and it is what we can measure.

That letter circulated among trial sites and appears to have shaped the criteria adopted across three major trials that year.

Kaplan emerges in the correspondence as a crucial collaborator, though initially an uncertain one. His early letters show hesitation about statistical methods that were then novel in clinical medicine. In a 1976 memo, he raises the problem that serum therapy patients could not be followed indefinitely—some would be discharged from hospital care and lost to observation. How could one analyze outcomes when the data were necessarily incomplete? The Kaplan-Meier method, which he was developing precisely to solve this problem, appears first in these letters as tentative notation and rough calculation, before it was formally published.

The correspondence also captures disagreements that would shape clinical practice for decades. One recurring tension concerns the role of the azo drugs in serum therapy trials. Some clinicians argued that patients receiving serum therapy should also receive sulfonamides, the only proven systemic antibacterial class, to maximize the chance of survival and therefore demonstrate the method's worth. Others argued this would confound the results: if patients improved, no one would know whether serum therapy had helped or whether the azo drugs had done the work. Umezaki appears to have sided gradually with the purists, though not without reluctance. A letter from March 1976 shows her wrestling with the ethics: "We cannot know if serum therapy works if we do not let it work alone. But we also cannot ethically watch a man die if a drug might save him."

The resolution, documented in subsequent trial protocols, was to use azo drugs only as a last resort, after serum therapy had been given genuine time to act. This decision, which now seems almost bureaucratic, represented a philosophical commitment to measuring serum therapy on its own terms rather than as an adjunct to drugs that already worked partially.

By the end of the correspondence, the framework was settled. Trials would measure survival at fixed intervals—at discharge, at six months, at one year. Kaplan-Meier curves would track patients' trajectories through time, whether they succeeded or failed or left observation. Success would not mean what it meant for azo drugs. It would mean stabilization, functional recovery, freedom from surgical intervention. The standard was lower than chemical cure, but it was honest about what serum therapy could achieve and measurable across populations and nations.

The collection also contains the mundane communications of experimental medicine: queries about which brand of culture media to use across sites, arguments over shipping procedures for animal sera, requests for clarification on follow-up protocols. These details are valuable to historians of statistical methods in medicine because they show that the Kaplan-Meier method and survival analysis were not imposed by statisticians from outside but emerged from the grinding problem of keeping five different hospitals' trial procedures comparable when serum arrived by cargo plane and laboratory standards varied by city.

The correspondence was catalogued by the Institute for the History of Bacteriology in the early 1980s and has been consulted extensively by scholars of serum pharmacology and the history of public health statistics, though detailed analysis of the trial design debates remains incomplete. The letters demonstrate how statistical practice changed not because of theoretical innovation alone but because of the presence of a disease for which no chemical cure existed and which demanded new ways of measuring what partial progress could be achieved.