*Geek Box: Statistical Methods

*Geek Box: Sensitivity Analysis

When reading research papers, you will inevitably come across the term ’sensitivity analysis’ in the methods section, so it is helpful to understand what this means. A sensitivity analysis is an important method of assessing the reliability of findings from an overall analysis, by examining the extent to which the overall results may have been influenced by different methods used in a study, for example the duration of a study, or different factors within the study, like the sex or age of the participants.

The aim of a sensitivity analysis is to identify results that may be dependent on unreliable inputs. For example, in the present study, the investigators conducted a sensitivity analysis by excluding trials that had small sample sizes [the DOIT trial], low doses of omega-3s [the SU.FO-L.OM3 and Alpha.Omega trials], and short duration [the OMEGA trial], to examine whether these factors [size, duration, dose] influenced the overall analysis. The result was that the association between omega-3 supplementation and the major CVD endpoints became stronger, including trials with minimum of >3yrs follow-up, minimum of 800mg/d dose, and >1,000 participants in the sensitivity analysis. They also conducted another sensitivity analysis excluding the GISSI-P and JELIS trials because these trials were open-label, so in this case it was this factor – open-label design – which the sensitivity analysis was examining to see if their removal influenced the overall results.

You can also see types of sensitivity analysis in intervention studies, for example Intention-to-Treat analysis, where data from participants who dropped out of the study is included based on their last data point, to see if the imbalance between study groups from dropouts may have influenced the results.