Introduction

In 1935, a research group in Evanston, Illinois, began measuring the alcohol content in urine from drivers who were hospitalized after a traffic crash. Over 3 years, they found that 47% of 268 injured drivers had been drinking, and 25% had measurements consistent with a blood alcohol level of 21.7 μmol/L or greater. The researchers reasoned that if alcohol use was a cause of traffic crashes, then the prevalence of alcohol in the blood of injured drivers should be greater than that in drivers who had not crashed. To test this hypothesis, they developed a machine which could measure alcohol in expired air. In April of 1938, they towed this machine to several sites in Evanston and measured alcohol levels in 1726 drivers at all hours of the day and night. They found that only 12% of drivers had alcohol in their breath, and only 2% had values consistent with a blood level of 21.7 μmol/L or more. When Holcomb (1938) published these findings, he made no mention of the case–control design, as this term had not yet been coined.

The case–control study design was more formally developed in the 1950s with a series of papers regarding the causes of cancer. The first statistical methods for analyzing these studies, still in use today, were formulated during the same era (Cornfield, 1951; Mantel and Haenszel, 1959). Had the Evanston researchers known these methods, theycould have estimated that the crude relative risk of hospitalization due to a crash was 22 (95% conidence interval, 14–34) for drivers with blood alcohol levels of 21.7 μmol/L or greater compared with drivers with no alcohol in their blood.