History of

J. Rosser Matthews University of Maryland, College Park, MD, USA

Correspondence to: J. Rosser Matthews Professional Writing Program English Department 1220 Tawes Hall University of Maryland College Park, MD, USA +1 (301) 405-3762 [email protected]

Abstract The history of biostatistics could be viewed as an ongoing dialectic between continuity and change. Although statistical methods are used in current clinical studies, there is still ambivalence towards its application when medical practitioners treat individual patients. This article illustrates this dialectic by highlighting selected historical episodes and methodological innovations – such as debates about inoculation and blood - letting, as well as how randomisation was introduced into design. These historical episodes are a catalyst to consider assistance of non-practitioners of such as and medical writers.

Methodologically, clinical trials and dialectic by discussing examples from different types of diseases.1 While basically epidemiological studies are united by a antiquity to the emergence of the clinical qualitative, the work is historically population-based focus; they privilege the trial in the mid-20th century. significant because it looked beyond the group (i.e., population) over the clinically individual to suggest a role for larger unique individual. Over time, this pop - Ancient sources: Hippocratic geographic and environmental factors. ulation-based thinking has remained writings and the bible Furthermore, it relied on naturalistic constant; however, the specific statistical Although the Hippocrates writers (active in explanations rather than invoking various techniques to measure and assess group the 5th century BCE) did not employ deities to account for illness and therefore characteristics have evolved. Consequently, statistical methods, one treatise does stand anticipated a modern scientific outlook. the history of biostatistics could be viewed out as a pioneering example of an environ - Another ancient forerunner of contemp- as an ongoing dialectic between continuity mental epidemiological study– the treatise orary clinical trials is discussed in the Bible’s and change. The continuity derives from On Airs, Waters, and Places (c. 400 BCE). 1 Book of Daniel. King Nebuchadnezzar of focusing on the group rather than the Relying on a view of disease as based on an Babylon wanted all of his subjects to eat a clinically distinct individual. The change imbalance in bodily fluids– known as diet of only meat and wine. However, Daniel derives from developments in statistical humours – the work emphasised how and some of the other Jewish children theory that have led to more sophisticated climatic changes throughout the seasons of wanted to eat a diet of legumes and water. analyses. In this article, I will illustrate this the year contributed to the spread of The King permitted them this diet for 10

8 | September 2016 Medical Writing | Volume 25 Number 3 Matthews – History of Biostatistics

days – after which it was determined that probability math ematics to contrast life 1872) (Figure 1). By collecting on they were indeed healthier. Consequently, expectancies for inoculated and non- patients admitted to hospitals, Louis argued they were allowed to continue on this diet.2 inoculated individuals; also, he calculated that the practice of bloodletting was actually Although not having the “apparatus” of a the benefits of inocu lation broken down by doing more harm than good. In his 1835 modern clinical trial (e.g., statistical tests to age. D’Alembert challenged Bernoulli’s treatise Recherches sur les effets de la saignée, determine p-values, confidence intervals assumptions and said that Bernoulli’s model Louis pointed out that 18 patients died out etc.), it does illustrate the use of a had not accurately captured the psychology of the 47 who had been bled (approximately comparison to test the efficacy of a dietary of human decision making – would an 3:7) whereas only nine died out of the 36 intervention. individual accept the risk of death now patients not bled – producing a lower (from inoculation) for an expected “pay-off” mortality rate of approximately 1:4.4 Eighteenth century of additional years of life when one was old Louis justified his approach by claiming developments and feeble?3 that the difference between numbers and In the 18th century, one prominent example While the debates about inoculation words (such as “more or less” and “rarely or of using statistical methods to resolve relied on mortality , the individual frequently”) is “the difference of truth and therapeutic debates centred on the practice that is more often credited with designing a error; of a thing clear and truly scientific on of inoculation. This involved controlled clinical trial (i.e., intentionally the one hand, and of something vague and inserting actual smallpox pustules under an dividing the participants into two or more worthless on the other.” Furthermore, Louis individual’s skin in the hope of creating a comparable groups to test hypotheses) is prophesied that, with the widespread mild (i.e., non-disfiguring) case of the James Lind (1716-1794). In 1757, Lind (a introduction of numerical reasoning, “we disease that would induce later immunity. ship’s ) had to deal with an outbreak shall hear no more of medical tact, of a kind Since this actually put patients at risk of of scurvy. He selected 12 of the sailors and of divining power of physicians.”4 In contracting a potentially fatal form of the divided them into six groups of twos. All were language that foreshadows contemporary disease, this became the subject of much given the same diet – except for a key discussions of “evidence-based medicine,” controversy. different ingredient for each of the distinct six Louis was basically saying that the key to Some argued against this procedure groups. For the two sailors who received transform medicine into a science was to based on the Hippocratic injunction “first, oranges and limes as supplement, there was rely on population-based thinking rather do not harm.” However, many writers one complete and one near recovery; none of than individual expertise. justified the procedure based on arguments the other five groups improved as much. Some of Louis’ contemporaries criticised that today would be called “risk-benefit Despite some obvious structural similarities his approach for failing to acknowledge that analysis.” For example, the London physician to the Biblical account, Lind is today the physician had to treat the John Arburthnot (1665-1735) published an regarded as the (modern) individual as a patient rather than anonymous pamphlet in 1722, in which he “father” of the controlled clinical a statistical construct. For examined the London Bills of Mortality from trial.2 instance, the phys ician Benigno earlier years and estimated that the chance Risueño d’Amador (1802- of dying from naturally-occurring smallpox Nineteenth century 1849) used an analogy to was 1:10. He then asserted (without developments maritime insurance. Although past evidence) that the chance of dying from In several areas of 19th century experience might tell you that 100 inoculation-induced smallpox was 1:100. scientific end eavour, statistical vessels would perish for each 1,000 This ten-fold reduction made him conclude reason ing was introduc ed – that embarked, these popul at ion- that inoculation made sense: “A Practice and the field of medi - based regularities could not which brings the Mortality of the Small Pox cine was no except - tell you which specific ships from one in ten to one in a hundred, if it ion. In the 1830s, obtain’d universally would save the City of one of the most London at least 1,500 People Yearly; and the pro m inent Figure 1. same Odds wou’d be a sufficient prudential advocates for Pierre-Charles- Motive to any private Person to proceed applying the Alexandre Louis upon.”3 In 1760, a more mathematically “numerical (1787-1872) was sophisticated version of this type of analysis method” to a pioneer of the took place in a debate between the Swiss medicine was “numerical mathematician Daniel Bernoulli (1700- the French clinician method” in 1782) and the French mathematician Jean Pierre-Charles- medicine. d’Alembert (1717-1783). Bernoulli drew on Alexandre Louis (1787-

www.emwa.org Volume 25 Number 3 | Medical Writing September 2016 | 9 History of Biostatistics – Matthews

would be destroyed. Analogously, Risueño followers committed to Gavarret’s specific interest in developing statistics derived from d’Amador argued that the calculus of the mathematical approach emerged. As a a desire to make explicit the statistical math em aticians “cannot be used to forecast result, the meaning of statistical evidence implications of Darwin’s theory of natural a determined event, but only to establish the remained contentious throughout the 19th selection, he also advocated the extension of probability of a certain numer ical prop - century. For example, the famous surgeon, these methods into medicine. To that end, ortion between two classes of possible (1827-1912), argued for his he often contributed to the British Medical events. But it is precisely this fact which particular method of Journal, , and The Royal Society of makes it completely useless in medicine.”5 based on statistical studies; however, his Medicine as attempts to “educate” the Drawing a different analogy, the physician critics had alternative theories of how to medical profession on the proper methods François Double (1776-1842) claimed that make surgery safer, citing other statistical of statistical reasoning.10 relying on the numerical method would studies that claimed to establish the One physician who would actively reduce the physician to “a shoemaker who superiority of their alternative theoretical embrace Pearson’s recommendations was after having measured the feet of a thousand approaches.8 Major Greenwood (1880-1949). Green - persisted in fitting every one on the basis of wood studied under Pearson in 1904-1905 the imaginary model.”6 The creation of the modern at the same time that he received his licence These types of criticisms discussed stat- clinical trial to practice medicine. At the beginning of istical reasoning in the context of medical The move to standardise and “mathematise” 1910, Greenwood would be awarded a full- ethics: should the physician be concerned statistics came with the creation, at Univer - time position as a medical at the primarily with advancing scientific knowledge sity College London, of the Biometric Lister Institute of Preventive Medicine. Like (through collecting empirical data), or with School in 1893 and the Biometric Pearson, Greenwood would proselytise for treating the individual in need of medical Laboratory a decade later.9 Heading the statistical methods by debating with care? At the same time, however, a more School and Laboratory was the pioneering physicians. One of his most noteworthy mathematically sophisticated critique of statistician Karl Pearson (1857-1936) encounters involved an exchange in the Louis’ work was developed by the French (Figure 3) who developed many modern Lancet in 1912-1913 with the bacteriologist physician Jules Gavarret (1809-1890) statistical techniques to study biological Sir Almroth Wright (1861-1947) over (Figure 2), who had been trained as an variation – such as curve-fitting and Wright’s use of vaccines to combat pneumo - engineer before becom ing a physician and goodness-of-fit tests, as well as methods for coccal among South African mine therefore understood probability mathem- measuring correlation.9 While Pearson’s workers. Centring on the issue of the atics. Gavarret published a treatise in 1840 accuracy of a blood test, the debate evolved entitled Principes généraux de statistique into a more generalised discussion over médicale in which he pointed out that Louis’s which forms of scientific evidence were averages could vary between what he called more credible.11 “limits of oscillation” if multiple samples By forging a career in academic science, were taken from the same population. For Greenwood would help lay the foundations instance, Louis had observ ed 140 cases of for a mathematically-informed understanding typhoid fever with 52 deaths and 88 of . In 1927, he would become recoveries, or a mortality of 37%. Relying on the first professor of epidemiology and vital probabilistic considerations, however, statistics at the London School of Gavarret posited that the results could vary and Tropical Medicine (LSHTM);12 in by 11.55%, or between 26% and 49% in 1928, he would be elected to the Royal every 140 cases observed.7 In modern day Society. Also, Greenwood would parlance, Gavarret was reporting the train many students, of which the “confidence interval” associated with Louis’ most prominent would be Sir result. Austin Bradford Hill (1897- To modern eyes, Gavarret seems 1991). remarkably prescient; however, there was no receptive audience for this marrying of statistics to probability mathematics in Figure 2. mid-19th century medicine. While his Jules Gavarret (1809-1890) used treatise was commented on throughout the probability mathematics by applying 19th century (with varying degrees of the concept of the confidence interval to mathematical sophistication), no “school” of .

10 | September 2016 Medical Writing | Volume 25 Number 3 Matthews – History of Biostatistics

criteria can be used to determine la Faculté de Médecine de Paris; 1840. whether an empirically observed assoc - 8. Tröhler U. Statistics and the British iation (e.g., between cigarette smoking controversy about the effects of Joseph and cancer) might be suggestive of an Lister’s system of antisepsis for surgery, underlying causal relationship. 1867-1890. JLL Bulletin: Today, the clinical trial is held as the Commentaries on the history of for certain knowledge, treatment evaluation. 2016 [cited 2016 and statistically-based epidemiological June 30]. Available from: studies are widely reported in the news. http://www.jameslindlibrary.org/ However, as this brief historical sketch articles/statistics-and-the-british- has illustrated, the current ascendency controversy-about-the-effects-of- of these population-based thinking joseph-listers-system-of-antisepsis-for- masks a larger ambivalence towards surgery-1867-1890/. statistical methods within the medical 9. Hardy A, Magnello ME. Statistical profession. Even as statistical methods methods in epidemiology: Karl have been used to justify notable Pearson, Ronald Ross, Major Figure 3. therapeutic breakthroughs, the population- Greenwood and Austin Bradford Hill, Karl Pearson (1857-1936) developed based thinking on which they are predicated 1900-1945. Soziologie- curve-fitting methods and measures still runs counter to the individualistic focus Präventivmedizin. 2002; 47: 80–9. of correlation. of clinical practice. Perhaps, this historical 10. Magnello E. The introduction of legacy is one of the reasons that clinical trials into medical Bradford Hill was the third son of the often require the services of “outside” research: The roles of Karl Pearson, physiologist Sir Leonard Hill (1866-1952) experts – such as statisticians and Major Greenwood, and Austin and had planned on following his father’s professional medical writers. Bradford Hill. In: Magnello E, Hardy medical profession. However, he contracted A, editors. The road to medical tuberculosis during World War I, and References statistics. Amsterdam-New York: eventually earned an economics degree by 1. The Internet Classics Archive. On Airs, Rodopi; 2002. p. 95-123. correspondence. Hill gravitated towards Waters, and Places by Hippocrates. 11. Matthews JR. Almroth Wright, vaccine statistics attending Pearson’s lectures. In 2016. [cited 2016 June 30]. Available therapy, and British biometrics: 1933, he would be appointed to a Reader - from: http://classics.mit.edu/ Disciplinary expertise versus statistical ship at the LSHTM; upon Greenwood’s Hippocrates/airwatpl.html. objectivity, In: Magnello E, Hardy A, retirement in 1945, Hill would succeed him 2. Collier R. Legumes, lemons and strep - editors. The road to medical statistics. as the head of the Statistics and Epidemi - to mycin: A short history of the clinical Amsterdam-New York: Rodopi; 2002. ology Unit.9 Like Pearson and Greenwood, trial. CMAJ. 2009;180(1): 23-4. p. 125-147. Hill sought to educate the medical prof - 3. Rusnock AA. Vital accounts: 12. Farewell V, Johnson T. Major ession on the proper use of statistics. In Quantifying health and population in Greenwood (1880–1949): 1937, he wrote a series of articles explaining eighteenth-century and A biographical and bibliographical statistical methods for the Lancet; sub- France. Cambridge: Cambridge study. Stat Med. 2016;35:645–70. sequently, they would be republished as University Press; 2002. Principles of Medical Statistics and go through 4. Louis PCA. Researchers on the effects multiple editions and translations. In 1946, of bloodletting in some inflammatory Hill would design a famous clinical trial to diseases, and on the Influence of Author information test the efficacy of streptomycin in treating tartarized antimony and vesication in J. Rosser Matthews received his tuberculosis – a methodologically note- pneumonitis. Putnam CG, translator. undergraduate degrees in mathematics worthy trial because it used a series of Boston: Hilliard, Gray; 1836. and philosophy, and then earned a Ph.D. random numbers to assign patients 5. Risueño d’Amador B. Memoire sur le in the and Medicine. to the control (bed rest) or experimental calcul des probabilites applique a la He is the author of the 1995 book (streptomycin) group. This trial is often medicine. Montpellier: L.F. Gaste; 1838. Quantification and the Quest for Medical characterised as the first clinical trial to use 6. Bulletin de l’Académie Royale de Certainty, which traced the history of a randomisation scheme effectively. In 1965, Médecine. 1836/37; 1: 702. statistical reasoning in 19th and 20th Hill would articulate what have come to be 7. Gavarret J. Principes généraux de century medicine. known as the “Bradford Hill Criteria.” These statistique médicale. Paris: Libraires de

www.emwa.org Volume 25 Number 3 | Medical Writing September 2016 | 11