May 2012 Critical Care Journal Club

May Journal club focused on Clinical Epidemiology and Research Design using a classic paper that demonstrated the relationship between smoking and lung cancer:  Doll R and Hill AB.  A study of the aetiology of carcinoma of the lung.  Br Med J 1952;2:1271-86. (Click here for a full text version of the article)

A little background is needed to truly appreciate the imagination, innovation and perseverance that went into this research.   At the time the study was performed, just after World War II over 80% of men in Great Britain smoked, and it was almost inconceivable that tobacco might cause cancer.  Doll and Hill themselves smoked, and Doll originally thought that car exhaust, or perhaps fumes rising from the tarred surface of streets were more likely to be implicated in the rising incidence of lung cancer observed after the war. 

The case control study design that Doll and Hill employed is now considered “classic", but at the time it was entirely novel, with no established standard methodology.  Doll and Hill thoughtfully and meticulously designed the study from scratch to eliminate all three threats to internal validity: bias, confounding and chance. 

At its core, the study compared 1357 men with lung cancer admitted to hospitals in London, Bristol, Cambridge, Leeds and Newcastle-upon-Tyne, to 1357 control patients who were matched by age, hospital of admission, area of residence and social status.  Matching was carefully performed to eliminate any potential bias related to exposure to occupational or environmental toxins.  “Almoners” (essentially British hospital social workers) administered a standardized three and a half page "questionary" that detailed the smoking history in terms of quantity, duration and method of smoking.  The questionary was designed to illuminate potential confounding variables, such as whether cigarette holders and petrol lighters were used, because it was just possible that lighter fumes rather than cigarette smoke might be carcinogenic.  Other potential confounders including occupation, residence near a Gasworks, exposure to different types of home heating, and previous respiratory illnesses were all closely examined and shown to be independently unrelated to lung cancer.   

Doll and Hill considered the concept of blinding.  They originally sought to keep the almoners uninformed as to each patient’s diagnosis at the time of the questionary, but when they specifically assessed this blind, they found the almoners often had discerned the diagnosis.  Later, they were able to compare data from the subset of patients originally thought to have had cancer who eventually ruled out, and show that their smoking history was identical to that of patients without cancer - thereby demonstrating that the almoners were not biased in their history-taking. 

Doll himself visited each hospital in order to confirm the diagnosis of lung cancer for each patient – this diagnostic process typically required 2-3 weeks.  Doll noted that non-smoking patients with suspected lung cancer almost always ruled out.  Although the association between smoking and lung cancer became apparent to the researchers early in the study – enough so to convince Doll to quit smoking - they pushed ahead with their work in order to achieve a convincing statistical result.  Of course, computers were not available at the time to perform statistical tests such as the Fisher exact test, but a Chi-squared statistic showed that the relationship between smoking and lung cancer had a p value of < 0.000001 - less than one in a million to have occurred by chance alone.  Despite the apparent methodological and statistical validity of their study, cancer researchers advising the Ministry of Health reportedly remained skeptical and initially advised against publishing the results "for fear of scaring people" unnecessarily. 

Doll and Hill went on to follow this study up with another classic - the British Doctor’s Study, a large prospective cohort study which confirmed their initial findings and also demonstrated the previously unsuspected relationships between smoking and chronic bronchitis and ischemic heart disease.  Despite the profound statistical significance of many of their studies, Hill would later argue against overemphasis on statistical significance in medical research – noting that systematic design errors were much more of a threat to validity than random error – this lesson is fundamental to critical appraisal of the medical literature.  In the course of their research, Doll and Hill developed much of what we now consider standard research design in Clinical Epidemiology.  Ultimately, our standard approach to consideration of causative associations in medicine was derived from these studies and codified as Sir Austin Bradford Hill's Criteria for Causation - which we will examine at a future date.  

Robert A. Raschke, MD

Associate Editor, Critical Care Journal Club

Reference as: Raschke RA. May 2012 critical care journal club. Southwest J Pulm Crit Care 2012;4:178-9. (Click here for a PDF version of the journal club)