IHP 330 Module 3 Study Design Critique example

The Coffee Scare of 1981: A Study Design Critique of a Case-Control Study Linking Coffee to Pancreatic Cancer

[Student Name]

Southern New Hampshire University

IHP 330: Principles of Epidemiology

Module Three Short Paper

[Instructor Name]

[Date]

The Coffee Scare of 1981: A Study Design Critique of a Case-Control Study Linking Coffee to Pancreatic Cancer

The Study

In 1981, researchers at a major school of public health published a case-control study of pancreatic cancer. They interviewed 369 patients with confirmed pancreatic cancer and 644 control patients about their use of tobacco, alcohol, tea and coffee. The control patients were hospitalized at the same hospitals, under the care of the same physicians, for conditions other than cancer. The researchers found a weak association with cigarette smoking, no association with alcohol or tea, and a strong association with coffee: after adjustment for smoking, the relative risk for drinking up to two cups a day was 1.8, and for three or more cups it was 2.7, with a significant dose-response trend (MacMahon et al., 1981). The authors themselves urged that the association be evaluated with other data, but the finding drew wide public attention.

Why a Case-Control Design

Pancreatic cancer is rare and develops over many years, which makes a cohort study slow and expensive; a very large group would need to be followed for decades to observe enough cases. A case-control study starts with people who already have the disease and compares their past exposures with those of people who do not. It is efficient for rare diseases and can examine several exposures at once (Centers for Disease Control and Prevention [CDC], 2012). The choice of design was reasonable. Its validity, however, depends heavily on who is chosen as controls and how exposure is measured.

Selection Bias in the Controls

The controls should represent the exposure distribution of the population from which the cases arose. Here, many controls were patients of the same gastroenterologists, hospitalized for digestive conditions such as ulcers, gallbladder disease and inflammatory bowel disease. People with these conditions are often told to cut down on coffee, or cut down on their own because it aggravates symptoms. If the controls drank less coffee than healthy people in the community, cases would look like heavy coffee drinkers by comparison even if coffee had no effect at all. This bias would push the odds ratio upward, in the direction of the reported association, and it would also create an apparent dose-response trend. Selection bias of this kind cannot be fixed by statistical adjustment after the fact.

Information Bias

Coffee consumption was measured by interview after diagnosis. Patients with a serious illness may search their past for causes and report exposures differently from patients with less alarming conditions, a form of recall bias. Pancreatic cancer can also cause symptoms, such as abdominal discomfort and weight loss, for months before diagnosis, which may change a person's coffee habits and blur what counts as usual consumption. The direction of these errors is less certain than the selection problem, but they add uncertainty to the exposure measure. Interviewers who knew which patients had cancer could also have probed differently, a form of interviewer bias.

Confounding

Smoking is the obvious potential confounder, because coffee drinkers are more likely to smoke and smoking is a known risk factor for pancreatic cancer. The researchers adjusted for cigarette smoking, and the association persisted, which is a strength. Other possible confounders, such as diet, diabetes, obesity and socioeconomic factors, were not fully addressed. Adjustment only removes confounding from factors that are measured well; residual confounding from imperfectly measured smoking or from unmeasured factors could remain.

Summary of Threats

The table below summarizes the main threats to validity.

Table 1

Threats to Validity in the 1981 Coffee Study

How the Study Could Have Been Designed Better

Several design choices could have reduced the threats identified. Community controls, selected from the same neighborhoods as the cases through random telephone sampling or population registers, would better reflect coffee drinking in the source population. If hospital controls were used, excluding patients whose conditions are known to affect coffee intake, such as ulcers and other digestive diseases, would have reduced selection bias. Asking about coffee consumption at a defined time several years before diagnosis would limit the effect of early symptoms on reported habits. Blinding interviewers to case status, where possible, and using a standard questionnaire would reduce interviewer bias. Finally, collecting information on diet, diabetes and body weight would allow adjustment for other plausible confounders.

What Later Evidence Showed

The same research group later conducted a second case-control study and did not confirm a dose-related increase in pancreatic cancer risk with coffee drinking (Hsieh et al., 1986). Many subsequent studies, including cohort studies that measured coffee intake before any diagnosis, also failed to show a consistent link. In 2016, an international panel of the World Health Organization's cancer research agency reviewed the evidence and classified coffee drinking as not classifiable as to its carcinogenicity, noting no consistent evidence of an association with pancreatic cancer (Loomis et al., 2016). The 1981 association is now widely regarded as a product of study design, most likely the choice of controls.

Lessons for Reading Case-Control Studies

Three lessons follow. First, always ask where the controls came from and whether their exposure reflects the population that produced the cases. Hospital controls are convenient but risky when their illnesses are related to the exposure. Second, a dose-response relationship strengthens a causal argument only if bias could not have produced it, and here it could. Third, a single case-control study, however well-known its authors, is a signal to investigate rather than a basis for public advice. Replication with different designs is what separates a real association from an artifact.

Conclusion

The 1981 coffee study used an appropriate design for a rare cancer and adjusted for the most obvious confounder, but its choice of hospital controls with digestive diseases likely made cases appear to drink more coffee than they really did in comparison with the general population. Later studies did not confirm the finding. The case remains a valuable lesson in how selection bias can create a strong, dose-dependent association where none exists.

References

Centers for Disease Control and Prevention. (2012). Principles of epidemiology in public health practice: An introduction to applied epidemiology and biostatistics (3rd ed.). U.S. Department of Health and Human Services.

Hsieh, C. C., MacMahon, B., Yen, S., Trichopoulos, D., Warren, K., & Nardi, G. (1986). Coffee and pancreatic cancer (Chapter 2). New England Journal of Medicine, 315(9), 587-589.

Loomis, D., Guyton, K. Z., Grosse, Y., Lauby-Secretan, B., El Ghissassi, F., Bouvard, V., Benbrahim-Tallaa, L., Guha, N., Mattock, H., & Straif, K. (2016). Carcinogenicity of drinking coffee, mate, and very hot beverages. The Lancet Oncology, 17(7), 877-878. https://doi.org/10.1016/S1470-2045(16)30239-X

MacMahon, B., Yen, S., Trichopoulos, D., Warren, K., & Nardi, G. (1981). Coffee and cancer of the pancreas. New England Journal of Medicine, 304(11), 630-633. https://doi.org/10.1056/NEJM198103123041102