NR 503 Week 5 Coffee Mortality Risk Factor Analysis Example

Coffee and Death in the NIH-AARP Cohort: A Hazard Ratio That Moved From 1.60 to 0.90 on Adjustment, the Covariate List Read as an Argument, and Diabetes as a Possible Mediator

Student Name

College of Nursing, Chamberlain University

NR 503: Population Health, Epidemiology and Statistical Principles

Instructor Name

Month Day, Year

Coffee and Death in the NIH-AARP Cohort: A Hazard Ratio That Moved From 1.60 to 0.90 on Adjustment, the Covariate List Read as an Argument, and Diabetes as a Possible Mediator

The Association Under Study

Patients ask nurses whether coffee is bad for them more often than almost any other diet question. The largest study on the question is Freedman et al. (2012), who followed 229,119 men and 173,141 women aged 50 to 71 in the National Institutes of Health-AARP Diet and Health Study from 1995 to 2008, excluding people with cancer, heart disease or stroke at baseline. Coffee consumption was measured once, by questionnaire, at the start. Over 5,148,760 person-years, 33,731 men and 18,784 women died.

This analysis keeps the association and any explanation of it apart throughout. The study reports an association between coffee and death. Whether coffee causes anything is a separate question, and the authors themselves say their data cannot answer it.

Crude Beside Adjusted

The study reports two sets of estimates, and the distance between them is the first thing to examine. Table 1 shows the hazard ratios for death from all causes among men, comparing each level of coffee drinking with none.

Table 1

All-Cause Mortality by Daily Coffee Consumption Among Men, Age-Adjusted and Multivariate-Adjusted

Note. From Freedman et al. (2012), Table 2. HR = hazard ratio; CI = confidence interval. The less-than-1-cup category is omitted for space.

What Moved the Number

For men drinking six or more cups a day, the hazard ratio went from 1.60 to 0.90: from a 60% higher risk of death to a 10% lower one. Something in the adjustment was doing a great deal of work, and whatever it was, the age-adjusted figure had been crediting it to coffee.

The baseline table shows the main candidate. Among men who drank no coffee, 4.8% were current smokers; among men drinking six or more cups, 34.7% were. For women the gap was wider, 8.1% against 48.1%. Heavy coffee drinkers were also more likely to drink more than three alcoholic drinks a day, ate more red meat and fewer fruits, and were less likely to be college graduates (Freedman et al., 2012). Smoking is strongly associated with both coffee drinking and death, which makes it a classic confounder. The authors identify smoking as the main factor behind the change.

The Covariate List as an Argument

The multivariate models adjusted for age, body-mass index, race or ethnic group, education, alcohol, the number of cigarettes smoked per day, pipe or cigar use and time since quitting, self-reported health status, diabetes, marital status, physical activity, total energy intake, fruit, vegetables, red meat, white meat, saturated fat and vitamin supplements, with hormone therapy added for women. Each item on that list is a claim that it could distort the coffee-death association and should be held constant.

Most of the list makes that case easily. Smoking, alcohol, education and diet plausibly came before a person's current coffee habit or alongside it, and each is linked to mortality. Adjusting for them is what confounding control means (Lash et al., 2021). The detail on smoking is a strength: the authors did not simply adjust for smoker versus non-smoker but for cigarettes per day and time since quitting, which reduces the chance that heavy smokers drive what remains.

Two items deserve a closer look, because they may not belong in the same category as the others.

Mediators Marked, Not Adjusted

Diabetes. A mediator is a variable on the causal path between exposure and outcome. Adjusting for a mediator removes part of the effect the study is trying to measure, a form of overadjustment (Schisterman et al., 2009). There is a body of research suggesting coffee drinking may be associated with a lower risk of type 2 diabetes. If that is true, then a lifetime coffee drinker's lower chance of having diabetes at baseline may be partly a result of coffee, and diabetes would lie between long-term coffee drinking and death. Adjusting for it would then take away part of any coffee benefit. The baseline table is consistent with this concern: women drinking no coffee reported diabetes at 7.4%, against 4.4% among those drinking six or more cups.

Health status. Self-reported health could work in either direction. People who feel unwell may cut back on coffee, making non-drinkers a sicker group, which is confounding. But if coffee affected health over years, baseline health could also sit partly on the causal path. The authors addressed the first possibility by excluding early deaths and by showing stronger associations in people reporting good health, which argues against sick people having simply stopped drinking coffee.

Neither concern overturns the finding, but a careful reading would mark both as possible mediators and ask what the estimate would be without them. The article does not report that model.

The Variable Nobody Had

Some variables could not be adjusted for because the study did not measure them. The authors name some themselves: coffee was measured once and may not reflect lifelong habits, the split between caffeinated and decaffeinated was based on what people drank more than half the time, and there were no data on how coffee was prepared, such as boiled, filtered or espresso (Freedman et al., 2012).

Income and wealth are the variable I would most want. Education is an imperfect proxy for socioeconomic position, and income affects access to health care, diet and stress. If income were related to coffee drinking in a way education does not capture, some confounding would remain. Because the direction of that relationship is unclear, I cannot say which way the estimate would move.

Where the Effect Differs

The association was not the same in every group. It was stronger among people who had never smoked or had quit than among current smokers, and stronger among those reporting good to excellent health than among those in poor to fair health, with significant tests for interaction in both sexes. The stronger association in never-smokers is reassuring, because it is the group least likely to be affected by leftover confounding from smoking.

For patient teaching, the analysis supports a modest statement: in a very large study of older adults, coffee drinkers did not have a higher risk of death once smoking and other factors were accounted for, and some had a slightly lower risk. It does not support telling a patient to start drinking coffee to live longer, and a current smoker who drinks a great deal of coffee should hear, first and clearly, about the smoking.

References

Freedman, N. D., Park, Y., Abnet, C. C., Hollenbeck, A. R., & Sinha, R. (2012). Association of coffee drinking with total and cause-specific mortality. New England Journal of Medicine, 366(20), 1891-1904. https://doi.org/10.1056/NEJMoa1112010

Lash, T. L., VanderWeele, T. J., Haneuse, S., & Rothman, K. J. (2021). Modern epidemiology (4th ed.). Wolters Kluwer.

Schisterman, E. F., Cole, S. R., & Platt, R. W. (2009). Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology, 20(4), 488-495. https://doi.org/10.1097/EDE.0b013e3181a819a1