Penny S. Reynolds
The World Patient Safety Day on September 17 was established by the World Health Organization in 2019. This month, we celebrate a few of the many statisticians and physicians who pioneered the use of quantitative evidence and statistical methods for identifying preventable patient harms.
1840: James McCune Smith was the first African American to obtain a medical degree and was a founding member of the New York Statistical Society. In 1840, he published a series of articles in a leading medical journal demonstrating that silver nitrate treatments for sexually transmitted disease in women were ineffective and harmful. Later, he used statistical arguments to debunk bogus and pseudoscientific medical practices such as phenology and the “lay puffery” of homeopathy.
1860: At the fourth International Statistical Congress in London, long-term collaborators Florence Nightingale and William Farr presented a groundbreaking proposal for standardizing hospital statistics, establishing a foundational framework for modern hospital quality improvement and patient safety and identifying corrective actions for reducing harm to patients and healthcare providers. Their clarification and systematic accounting of competing risks of recovery and death were major conceptual breakthroughs.
1861: Handwashing was the most effective protection against hospital-acquired infections, which was still a significant patient safety concern. Using the “numerical method,” advocated by Pierre‑Charles Alexandre Louis, Ignaz Semmelweis published a quantitative comparison of mortality rates between two maternity wards at Vienna General Hospital. He showed that maternal deaths from puerperal (child bed) fever dropped dramatically from 18% to under 2% after stringent handwashing protocols were introduced. His attempts at reform did not go well: His findings were rejected by the medical establishment, and he lost his job and died in a mental institution. However, since then, handwashing has saved millions of lives.
1904: Karl Pearson introduced the first clinical meta-analysis of observational data. He compared rates of typhoid infection and mortality in vaccinated and unvaccinated soldiers. He provided a pooled estimate of effect in the last line of the tabulated results.
1935: Louis Isaac Dublin, chief statistician and vice president of the Metropolitan Life Insurance Company, went public with a shocking exposé of US maternal mortality rates. Arguing that maternal mortality was a preventable public health failure driven primarily by economic disparities, he advocated for practical reforms such as universal routine prenatal care and subsidized maternal clinics, providing affordable access to qualified care. In 1938, he announced “the tide has turned,” but almost a century later, US maternal mortality is still the highest among developed nations.
1954: Peter Armitage published the first comprehensive paper on the application of sequential analysis to clinical trials. Sequential analysis enabled statistically valid interim analyses of accumulating trial data to decide whether a trial should stop early for benefit, harm, or futility. This development minimized unnecessary patient exposure to ineffective or harmful treatments while maintaining the statistical integrity of the trial.
1969: The National Halothane Study was the first major collaboration between statisticians, computer scientists, and physicians—and one of the first to use mainframe computers to process big data. The study assessed over 865,000 patient records across 34 hospitals to evaluate the claim that the anesthetic agent halothane was associated with post-operative liver failure and patient deaths. Frederick Mosteller coordinated the distinguished team of statisticians: John Tukey, Lincoln E. Moses, Stephen E. Fienberg, Yvonne Bishop, Byron M Brown, Jr., and John P. Gilbert, along with computer scientists W. Morven Gentleman, Joseph Halpern, and Lawrence G. Tesler. Innovative and improved statistical applications developed by the team included multivariate log-linear models, hospital ranking and case-mix adjustment methods, robust estimation, and exploratory data analysis. Although the study did not find an association between halothane and liver injury, it did uncover an alarming 24-fold difference in death rates among different hospitals that could not be explained by differences in case mix or anesthetic agent.
1972: In his book Effectiveness and Efficiency, physician Archie Cochrane argued that medical practice should be grounded in evidence from well-designed, randomized controlled trials and regularly updated systematic reviews—rather than depending on individual studies, narrative reviews, tradition, or opinion. The Cochrane Collaboration was established in 1993 as a centralized repository for high-quality, systematically synthesized medical evidence.
1976: Gene V. Glass coined the term meta-analysis, defined as “the statistical analysis of a large collection of analysis results from individual studies for the purpose of integrating the findings.” Karl Pearson, Ronald A. Fisher, and William G. Cochran had already developed the basic methodology, with further advances in inference and effect-size estimation made by Ingram Olkin, Larry V. Hedges, and Sir Richard Peto. Clinical researchers, such as Thomas C. Chalmers and Iain Chalmers, demonstrated the importance of meta-analysis and high-quality systematic reviews for improving patient care and reducing avoidable patient deaths.
1978: An early version of a forest plot was published. It summarized results for 71 “negative” randomized controlled trials, using modified boxplots to show trial-specific effect sizes and respective confidence intervals. In 1982, the first modern forest plot appeared in a formal meta-analysis evaluating beta-blockers for reducing post-heart-attack mortality. In 1983, the signature forest plot “look” emerged, with point estimates represented by squares and confidence intervals by lines. In 1992, Thomas C. Chalmers and Fred Mosteller introduced the cumulative forest plot. Because it updated effect size as trial results were added, a cumulative meta-analysis revealed gaps between clinical guidelines and definitive evidence, capitalizing on accumulated evidence for either benefit or harm. A dramatic demonstration of the human cost of overlooked evidence was the 2005 retrospective meta-analysis of the association of infant sleeping position and risk of sudden infant death syndrome. Epidemiologist Ruth Gilbert and colleagues showed that robust evidence linking prone sleeping position to a threefold increased risk of SIDS already existed by 1970—24 years before the Back to Sleep campaign was launched. The failure to systematically synthesize available evidence contributed to more than 50,000 preventable infant deaths worldwide.
1998: The Bristol Royal Infirmary Public Inquiry launched to investigate unusually high mortality rates of infants undergoing complex cardiac surgery between 1984 and 1995. Sir David Spiegelhalter and colleagues used risk-adjusted mortality analysis, random-effects hierarchical modeling, case-mix adjustment, and sensitivity analyses to compare outcomes across 12 UK pediatric cardiac centers. Their analyses showed that Bristol was a true statistical outlier: Mortality rates were almost double those of comparable UK centers, leading to many avoidable deaths. The inquiry revealed system failures in outcomes monitoring, clinical governance, organizational culture, and accountability. It also determined that routine standardized statistical surveillance of clinical outcomes was essential for detecting unsafe practice.
Leave a Reply