401175 Analytic Approaches in Epidemiology

Question 1:

A large prospective cohort study was conducted among a sample of Australians aged ≥40 years from the general population, investigating the association between tobacco smoking and incidence of pancreatic cancer over a follow-up period of 18 months. The data at the end of the follow-up period are given below, stratified by self-reported family history of pancreatic cancer:

Question 2

In a similar prospective study conducted in the United Kingdom, researchers noted that high levels of red meat consumption, particularly processed meats, were also associated with pancreatic cancer. They also noted that high red meat consumption was more prevalent among smokers than non-smokers. The UK researchers found that those in the highest quintile of red meat consumption had a higher relative risk of pancreatic cancer (RR=1.62) compared to those in the lowest quintile of red meat consumption, and the prevalence of high red meat consumption was higher among smokers (26%) compared to non-smokers (15%).

The investigators in the Australian study in Question 1 did not collect information on red meat consumption or any dietary factors. Given the information above, what would be the association in the total cohort between tobacco smoking and pancreatic cancer had the Australian investigators collected this information and adjusted for it in the analysis?

Question 3

A retrospective cohort study was conducted of all live births over a 2-year follow-up period from an Australian Local Health District to investigate the association between ante-natal depressive symptoms in mothers and low birth weight (LBW) in offspring. The outcome was defined as a binary variable, with offspring categorised as either low birth weight (<2500g) or normal birth weight (>=2500g). The investigators were interested in adjusting for a range of potential confounders, therefore regression models were used to analyse the data. The regression coefficients (β) from a series of univariable logistic regression models (labelled ‘Crude’) and a multivariable logistical regression model (labelled ‘Adjusted’) are provided in the table below:

(a) What was the crude (unadjusted) association between ante-natal depressive symptoms and low birth weight? Note: Express the association as an estimate of the relative risk. Is this a strong association? Why or why not?

(b) What was the crude (unadjusted) association between the potential confounders and low birth weight? How likely is it that the association between ante-natal depressive symptoms and low birth weight might be confounded by smoking during pregnancy and maternal age? Why or why not?

(c) What was the adjusted association between ante-natal depressive symptoms and low birth weight? Note: Express the association as an estimate of the relative risk. Is this a strong association? Why or why not?

(d) What effect did adjusting for smoking during pregnancy and maternal age have on the association between ante-natal depressive symptoms and low birth weight?

(e) Based on the adjusted data, what is the risk of low birth weight for mothers with ante-natal depression, who smoked during pregnancy, and were aged >=40 years?

Question 4

Propose a causal diagram of the relationship between ante-natal depressive symptoms and low birth weight, based on the variables provided in Question 3. Provide a narrative description (1 paragraph only) of the causal diagram incorporating your interpretation of the findings presented in Question 3.