This lecture was created by Gemma Sharp from the University of Bristol.

The focus of this lecture is on minimising bias and confounding when using observational data to understand causal effects.
We start by discussing why epidemiologists might want to identify causal effects, and some of the features of observational studies that makes this difficult. These include confounding, reverse causation, selection bias and reporting bias.

We then discuss some of the methods/approaches that can be used to minimise these issues, either at the design or the analysis stage. Methods including negative control studies, cross context comparisons and matched (e.g. within sibship) designs are described, as well as multivariable regression analyses. We will consider the strengths and limitations of these approaches and also how evidence garnered using these different approaches (each of which has different limitations and key sources of bias) can be “triangulated” to infer causality from observational data.

By the end of this lecture, students will be able to:
1. Explain confounding, bias and reverse causality;
2. Describe the limitations of observational epidemiological studies;
3. Describe ways to reduce bias and confounding in observational epidemiological studies;
4. Interpret associations from an observational study considering the potential for confounding and bias.

Essential reading:
• Lawlor, DA, Tilling, K, Davey Smith, G. Triangulation in aetiological epidemiology. International Journal of Epidemiology. 2016;45(6):1866-86.
Recommended reading:
• Richmond, RR et al. (2014) Approaches for drawing causal inferences from epidemiological birth cohorts: A Review Early Hum Dev. 2014 Nov; 90(11): 769–780.
• Brion MJ et al. (2011) What are the causal effects of breastfeeding on IQ, obesity and blood pressure? Evidence from comparing high-income with middle-income cohorts. Int J Epidemiol. 2011 Jun;40(3):670-80

свернуть