Advanced Epidemiology

Master of Public Health
Advanced Epidemiology
Mock Coursework 2019-2020

Preamble
This assignment is designed to test your understanding of contemporary epidemiology. Please read the following paper and answer the questions below:

Hart CL, Smith GD, Hole DJ, Hawthorne VM. Alcohol consumption and mortality from all causes, coronary heart disease, and stroke: results from a prospective cohort study of Scottish men with 21 years of follow up. BMJ. 1999;318(7200):1725-9.

Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC31100/pdf/1725.pdf

Questions

1. What exposure are the authors interested in studying (intended exposure)? What exposure are they actually studying (actual exposure)? How might the difference between the actual and intended exposures influence the study findings?
[4 marks]

2. Now, assuming that there is a perfect one to one relationship between the intended exposure and the actual exposure, perform the following tasks (you may restrict your response to a specific outcome):-

a. Write the formulae using counterfactual mathematical notation for: (i) what is being estimated in this study and (ii) what underlying causal question is being addressed.
[12 marks]

b. Now explain your answers to a.(i) and a.(ii) in words . Define any technical epidemiological terms used in the explanation (e.g. association).
[12 marks]

3. In the first paragraph, the authors state: “Non-drinkers are generally found to have a mortality between that of moderate and high consumers.” The authors describe this loosely as “selection bias”. Using a more precise classification based on the three major types of bias that you have
learned about (i.e. confounding, reverse causation, collider bias), explain which of these three types of bias this is. Draw a directed acyclic graph to illustrate the general form of this bias.
[10 marks]

(continued over page)

4. Table 2 provides the results for three different sets of statistical adjustment models. Based on the authors’ reported rationale for running these regression models, draw DAGs for all three models. Assume that the authors’ view of the causal mechanisms underpinning the data remain the same across all three models. For simplicity, you can restrict your DAGs to only include the following variables: alcohol, mortality, age, smoking, body mass index, father’s social class, social class and
diastolic blood pressure. Explain which model would be preferred for the purposes of producing the total causal effect estimate, and why.
[20 marks]

5. How have the authors followed up the cohort? Name two strengths and two limitations of the approach used in this study?
[16 marks]

6. Consider a (notional) randomised controlled trial designed to examine the effects on the 25-year risk of mortality of an intervention designed to reduce alcohol consumption. This study was conducted in a representative sample of the Scottish population.

a. The effect of the intervention in smokers and non-smokers are shown Table 1 below. Assume that the risks below are estimated with complete certainty (ie that the standard errors are zero) and that there is no confounding. The principal investigator who led the trial asserts that the effect of the intervention is different among “never smokers” and “former or current smokers”. The health economist for the trial asserts that the effect of the intervention is the same in both groups. Explain how the principal investigator and health economist can both be correct.
[6 marks]

b. A policy maker in Scotland needs to decide how to allocate this intervention across “never smokers” and “former or current smokers” Scotland. Which of these interpretations (that of the principal investigator or of the health economist) is most relevant to their decision, and why?

[4 marks]

(continued over page)

Table 1: 25-year mortality risk by intervention arm and smoking status at baseline
Intervention to reduce alcohol consumption No intervention Relative risk no intervention versus intervention Absolute risk difference no intervention versus
intervention
Never smoked 10% 14% 1.40 4%
Former or
current smoker 20% 24% 1.20 4%

7. A new treatment for alcohol dependence has been introduced across the country. However, there are concerns that the treatment may increase the risk of suicide – an outcome which is too rare for a randomised controlled trials to detect. You are contacted by a general practitioner (GP) who is interested in carrying out a study using administrative data to compare suicide risk among patients of different GPs. She thinks that GPs may have a different tendency to prescribe the treatment and that this variation in practice could be used as an “instrumental variable”. Draw a DAG to illustrate this type of study design. In relation to the scenario described, what are the assumptions that would need to be met for this analysis to be valid? What are the strengths of this approach over analyses which do not include an instrumental variable?
[16 marks]