Questions remain about how much evidence is needed to accept or at least to consider an observation as providing sufficiently strong evidence to justify a change in clinical practice (Glasziou et al. 2007; Guyatt et al. 2011), even when the key outcome measure is objective, definitive and clinically very important.
Maternal death is one such outcome, and Ignaz Semmelweis’ observation of the reduction in maternal deaths associated with the introduction of handwashing in a maternity hospital in Vienna provides an illustrative example (Semmelweis 1861; Loudon 2002). A plot of the rate of maternal deaths every month from January 1841 until May 1847 shows substantial variability and some evidence of a pattern in the changes, possibly related to the seasons (Carter 1983). After handwashing was introduced in June 1847, deaths were less frequent and showed less monthly variability in the period up to February 1849. How much analysis is needed to convince people that this is an effect of handwashing and not a chance association, or a reflection of bias (Guyatt et al. 2011)?
Eyeballing the plot shown in the Figure, many observers might consider the reduction in deaths following the introduction of handwashing to be real and significant. When considering the high risk of bias in studies using historical controls, Glasziou et al. (2007) suggested considering an effect as likely to be real if a risk ratio of between 5 and 10 has been observed. In this case, a comparison of the average monthly mortality rate before the introduction of handwashing (10.65%, blue points) and after it (1.98%, red points) yields a risk ratio of 5.38. We can explore these data in more depth by analyzing this interrupted time series with a SARIMA (Seasonal AutoRegressive Integrated Moving Average) model. The monthly mortality rate can be formulated as
MRt = B*H + AR*MRt-1 + MA*Et-12 + Et
Where MR = mortality rate, t = month, B = handwashing coefficient, H = handwashing; 0 before, 1 after the introduction of handwashing, AR = first order autoregressive coefficient, MA =monthly moving average coefficient, E = random error.
The estimations of the three coefficients of the model are presented in the table below.
1st order autoregressive
Monthly moving average
This yields an estimated absolute reduction in maternal mortality of 8.87%, which is highly statistically significant (p = 0.0042). The reduction is also of great human significance: for every 11 women examined after handwashing, 1 fewer died. The analysis could be extended to take into account the data from the two clinics concerned, and to cover an earlier period, during which post-mortem examinations were not done routinely (Review Article 2008). This discovery happened before the germ theory take off with the work of Louis Pasteur in the early 1860’s. The miasma theory was prevalent and no strong explanation was available to support this impressive observation.
These different levels of analysis strongly support the existence of a beneficial impact of handwashing on the risk of maternal mortality. This is not to suggest that data of this level of confidence are sufficient to introduce changes in clinical practice, which is influenced by medical tradition and cultural and other factors (Loudon 2002; Review article 2008). In Semmelweis’ case, the evidence of an important impact of handwashing on maternal mortality should at least have led traditional medical practices to have been questioned.
The Editor is grateful to Prof Paul Glasziou and Dr Irvine Loudon for helpful comments on an earlier draft of this article.
This James Lind Library commentary has been republished in the Journal of the Royal Society of Medicine 2013;106:459-460. Print PDF
Glasziou P, Chalmers I, McCulloch P, Rawlins M (2007). When are randomised trials unnecessary? Picking signal from noise. BMJ 334: 349-51.
Guyatt GH, Oxman AD, Sultan S, Glasziou P, Akl EA, Alonso-Coello P, Atkins D, Kunz R, Brozek J, Montori V, Jaeschke R, Rind D, Dahm P, Meerpohl J, Vist G, Berliner E, Norris S, Falck-Ytter Y, Murad MH, Schünemann HJ; The GRADE Working Group
http://www.jclinepi.com/article/S0895-4356(11)00184-3/fulltext (2011). GRADE guidelines: 9. Rating up the quality of evidence. Journal of Clinical Epidemiology 64:1311-1316.
Loudon I (2002). Ignaz Phillip Semmelweis’ studies of death in childbirth. JLL Bulletin: Commentaries on the history of treatment evaluation (https://www.jameslindlibrary.org/articles/ignaz-phillip-semmelweis-studies-of-death-in-childbirth/).
KC Carter (1983) Ignaz Semmelweis The Etiology, concepts and Prophylaxis of Childbed Fever. Madison, Wisconsin; University of Wisconsin Press, p 72 and pp 90-1.
Review Article (2008). Semmelweis and the aetiology of puerperal sepsis 160 years on: an historical review. Epidemiol. Infect. 136:1–9.
Semmelweis I (1861). Die Aetiologie, der Begriff und die Prophylaxis des Kindbettfiebers. [The etiology, concept, and prophylaxis of childbed fever]. Budapest and Vienna: Hartleben.