Epidemiology: challenges in the interpretation and analysis with - - PowerPoint PPT Presentation

Epidemiology: challenges in the interpretation and analysis with some examples

Lars T. Fadnes Centre for International Health University of Bergen

Example

• Zinc sirup

– Metallic and bitter taste – Accepted mostly by the most sick and less by the healthier

• Placebo sirup

– Taste like fruit juice – Accepted my most children

Epidemiology: challenges in the interpretation and analysis with s Randomised to treatmentarm Taking all doses Proportion taking all doses in analysis zinc Most sick 55 45 82% Least sick 110 55 50% placebo Most sick 55 50 91% Least sick 110 100 91%

Intention-to-treat analysis Per protocol What are the challenges?

Interactions/ effect modification

• The effect of one factor on the outcome is dependant on

an other factor

• How to assess?

– Stratified analysis – Multivariable analysis

• E.g. effect of treatment (zinc) on diarrhoea severity is

depending on whether the child had fever or not

– Those with fever had protective effect of zinc – Those without fever had no positive effects of zinc

• Observation of associations between

investigated exposure factors and their

• utcomes does not necessarily imply causation.

In the absence of random error and bias, there are three possible explanations for these associations in nature:

1) The exposure causes the outcome 2) The outcome causes the exposure (reverse causation) 3) The exposure and outcome share a common cause

unlikely if the exposure always comes before the outcome

What is this effect called? Confounding…

Confounding

• Exposure and outcome share a common cause

– Confounding can be seen as a mixing of effects – A confounder is a variable that causes confounding or a variable which removes confounding when adjusted for – hinders our ability to see the true causal effect of the exposure on the outcome

• When do we need to suspect it?

– Imbalance in factor/ determinant between the exposed and unexposed groups

• Design

– Randomisation (equal distribution of common causes of both the exposures and outcome) – matching of cases

• How to assess?

– Stratified analysis – Multivariable analysis

• regression methods to control for multiple confounders at the same time

Mediation

Accuracy versus precision

High accuracy, but low precision High precision, but low accuracy

Before you finish…

• Take 1 minute to fill in short evaluation of

these sessions in the computer lab on the following web site http://evaluation.fadnes.net/

Questions and comments

Installation of RcmdrPlugin.Cih.Epi

• Install R (if not done already)

– http://www.r-project.org/

• Install package Rcmdr (if not done already)

– Package  Install package  scroll down to Rcmdr  click OK

• Install package Epi

– Package  Install package  scroll down to Rcmdr  click OK

• Download RcmdrPlugin.Cih.Epi from

http://statistics.fadnes.net/epi/

• Install package(s) from local zip file (choose the package you just downloaded)
• Open program

– Load package Rcmdr – Tools -> load Rcmdr plug-in(s)  RcmdrPlugin.Cih.Epi

• click ok and Yes
• Now you are ready