A twenty-first century perspective on concepts of modern epidemiology in Ignaz Philipp Semmelweis' work on puerperal sepsis

Abstract

We aimed to review Semmelweis's complete work on puerperal sepsis mortality in maternity wards in relation to exposure to cadavers and chlorine handwashing and other factors from the perspective of modern epidemiological methods. We reviewed Semmelweis' complete work and data as published by von Györy 1905 according to current standards. We paid particular attention to Semmelweis's definition of mortality in and of itself, to concepts of modern epidemiology that were already recognizable in Semmelweis's work, and to bias sources. We did several quantitative bias analyses to address selection bias and information bias from outcome measurement error. Semmelweis addressed biases that have become known to modern epidemiology, such as confounding, selection bias and bias from outcome misclassification. Our bias analysis shows that differential loss to follow-up is an unlikely explanation for his results. Bias due to outcome misclassification would only be relevant if misclassification differed between time periods. Confounding by health status was likely but could not be quantitatively addressed. Semmelweis was aware that cause-specific mortality is a function of incidence and prognosis. He reasoned in potential outcome terms to estimate the reduced number of deaths from an intervention. He advanced a hypothesis of clinic overcrowding as a risk factor for puerperal sepsis mortality that turns out to be wrong. Semmelweis' data provide a great pool for illustrating the logic of scientific discovery by use of the numerical method. The explanatory power of his work was strong and Semmelweis was able to refute several previous causal explanations.

Keywords: Cross infection/history; Cross infection/prevention & control; Cross infection/transmission; Female; Hand disinfection; History; Humans; Nineteenth century; Obstetrics/history*; Pregnancy; Public health/history; Puerperal infection/history*; Puerperal infection/prevention & control.

Fig. 1

Recalculated monthly mortality ratios in Clinic 1, General Hospital Vienna (July 1840 through January 1853). Legend: first vertical reference line (June 1847): the first full month of chlorine washing of the hands; second vertical reference line (Feb 1849): the last full month of chlorine washing before Semmelweis leaves the Clinic; RD: difference of mortality ratios, 95%CI: 95% confidence interval; NNT: number needed to treat for the comparison of the later period with the earlier period; the average number of puerperae per months was 265

Fig. 2

Recalculated monthly mortality ratios (%) at Clinic 1, General Hospital, Vienna, January 1846 through July 1849. Legend: (1) December 1846 to March 1847: first assistant Breit and his students rarely undertake or join autopsies; (2) Semmelweis becomes first assistant again (March 20, 1847); (3) Mid of May 1847: start of chlorine washing of the hands; (4) September 1847: lack of compliance of chlorine hand washing by medical students and junior physicians, October 1847 treatment of a puerpera with a foully discharging medullary carcinoma of the uterus, November 1847 treatment of a puerpera with discharging carious left knee; (5) Semmelweis leaves Clinic 1 on March 20, 1849

Fig. 3

Recalculated mortality ratios of different time periods and exposure potential to “cadaverous particles”. Legend R: indicates mortality ratio; numbers in parenthesis indicate deaths and inpatient women of the corresponding time periods respectively

Fig. 4

Seasonality and mortality ratio at Clinic 1, General Hospital Vienna, July 1840 through November 1946. Legend: vertical whiskers indicate 95% confidence intervals; the blue line indicates the monthly observed mortality ratio and the red dotted line indicates the monthly estimated mortality ratios; estimated seasonal relative risk 2.0 (95%CI: 1.8–2.3) with an estimated time of peak on December 14 (344°)

Fig. 5

Association between monthly number of women in Clinic 1 and 2 and puerperal sepsis mortality ratio (%). Legend : Relative risk of puerperal sepsis death per increment of 50 inpatient women per month; Red points and whiskers indicate point estimates with 95% confidence intervals per quartile of inpatient women per month; monthly mortality ratios from July 1840 to April 1847 in Clinic 1 were potentially biased because of en masse transfer of ill women to other wards of the General Hospital; trend lines from log-binomial models with empirical sandwich errors