Random Error, Confounding, Effect Modification
Although researchers do their best to reduce error within every study, there will always be error. It is important to identify and report any possible error within the research study in order to accurately interpret the research study’s findings. In epidemiologic research, the focus is on assessing confounding and effect modification along with normal statistical measures (p-values, confidence intervals, etc.).
For this Application Assignment, you will calculate and interpret the effects of confounding, random error, and effect modification within epidemiologic research. Read each of the following questions and answer them appropriately:
· Consider each of the following scenarios and explain whether the variable in question is a confounder:
a. A study of the relationship between exercise and heart attacks that is conducted among women who do not smoke. Explain whether gender is a confounder.
> In this case, gender is not a confounder since it is specified as the population under investigation. On the other hand, the quantity and duration of smoking is a confounder and needs to be defined to a certain criteria.
b. A case-control study of the relationship between liver cirrhosis and alcohol use. In this study, smoking is associated with drinking alcohol and is a risk factor for liver cirrhosis among both non-alcoholics and alcoholics. Explain whether smoking is a confounder.
> Smoking is a confounder since it affects the outcome between the dependent and independent variables. If smoking is not measured, it could be the source of the correlation between liver cirrhosis and alcohol use. Therefore, it needs to be accounted for.
· Interpret the results of the following studies
a. An odds ratio of 1.2 (95% confidence interval: 0.8-1.5) is found for the association of low socioeconomic status and occurrence of obesity.
>This explains that the occurrence of obesity is not likely for those population in low socioeconomic status.
b. A relative risk of 3.0 is reported for the association between consumption of red meat and the occurrence of colon cancer. The p-value of the association is 0.15.
>There is a low risk for people who consume red meat to have colon cancer.
c. An odds ratio of 7 (95% confidence interval: 3.0 – 11.4) is found for the association of smoking and lung cancer.
>It is most likely for the population of smokers to have lung cancer.
· The relationship between cigarette smoking and lung cancer was conducted in a case-control study with 700 cases and 425 controls. Using the results below, calculate the crude odds ratio and explain what the ratio means:
· Heavy Smoking—Cases: 450; Controls: 200
· Not Heavy Smoking—Cases: 250; Controls: 225
(450/200) / (250/225) = 1.11
> The crude odds ration is 1.11. Since the value is greater than 1, we can conclude that heavy smokers have a higher odds of having lung cancer.
· A case-control study looked at the association of alcohol use with the occurrence of coronary heart disease (CHD). There were 300 participants in the study (150 cases and 150 controls). Of the cases, 90 participants drank alcohol; of the controls, 60 participants drank alcohol.
Design the appropriate 2×2 table, calculate and interpret the appropriate measure of association.
No alcohol use
>(90/60) / (60/90) = 2.24. The population of those who drank alcohol are most likely to develop coronary heart disease.
· You suspect that the association between alcohol use and CHD might be confounded by smoking. You collect the following data:
No Alcohol Use
Calculate the appropriate measure of association between alcohol use and CHD in both smokers and non-smokers. Discuss whether smoking was a confounder of the association. What is the relationship of alcohol use to CHD after controlling for confounding?
· A study was conducted in young adults to look at the association between taking a driver’s education class and the risk of being in an automobile accident. 450 participants were included in the study, 150 cases who had been in an accident and 300 controls who had not. Of the 150 cases, 70 reported having taken a driver’s education class. Of the 300 controls, 170 reported having taken a driver’s education class.
Calculate and interpret the appropriate measure of association between driver’s education and accidents.
The question arose as to whether gender might be an effect modifier of this association. When gender was assessed, the data looked like the following:
No Drivers Ed
Perform the appropriate calculations to test for effect modification. Interpret your results.
Course Text: Essentials of Epidemiology in Public Health
Aschengrau, A., & Seage, G. R., III (2008). Essentials of epidemiology in public health (2nd ed.). Sudbury, MA: Jones and Bartlett.
Chapter 11, “Confounding”
Chapter 11 defines confounding and provides several examples. It discusses methods to control for confounding in both the design and analysis stages of a study.
· Chapter 12, “Random Error” pp. 307-321
Chapter 12 deals with a presentation of P values, power, and confidence intervals. These are important concepts used during the evaluation stage of a research study and when reporting results in epidemiologic literature. These concepts should be familiar to you from Biostatistics.
· Chapter 13, “Effect Measure Modification”
Chapter 13 presents an overview of effect modification (also known as interaction). Effect modification occurs when the strength of an association between an exposure and an outcome differs according to the level of another variable. One classic example of effect modification is smoking, which is an effect modifier for exposure to asbestos and lung cancer.
· Confounding Bias
To learn more about how confounding can be controlled, work through this module on confounding bias from the North Carolina Center for Public Health Preparedness. You need to register to access these free materials.
· ERIC Notebook: Confounding Bias, Part I
· ERIC Notebook: Confounding Bias, Part II
This two-part overview from UNC examines confounding bias. Part I talks about the distortion confounding can cause and how to assess if in fact confounding is occurring. Part II discusses strategies for controlling confounding and how to calculate adjusted measures to address the distortions that occur when confounding is present.
· ERIC Notebook: Common Statistical Tests and Applications in Epidemiological Literature