*Geek Box: Post-Hoc Analysis
The literal translation of the Latin term ‘post-hoc’ is ‘after this’. Post-hoc means something which occurs after an event. A post-hoc analysis is therefore an analysis which is undertaken after a trial has concluded, using data from that trial, to look at a question that was not pre-specified.
So of course, while all data analysis is conducted after a study is concluded, the key feature of a post-hoc analysis is that it is secondary to the study, and the question it is addressing was not a primary research question.
For example, the DIETFITS trial was designed to test the effects of baseline insulin secretion and genetic factors on response to either low-carbohydrate or low-fat diets. It was not designed to test the effects of different percentages of saturated fat on cardiovascular risk factors.
However, the researchers had data on both diet and blood lipids. Thus, they could do a post-hoc analysis secondary to the intervention, looking at the relationship between saturated fat and blood lipids. It is important to distinguish that, because it is a secondary analysis and not part of the pre-planned study, in effect a post-hoc analysis is an observational study. It can look at relationships, but not causality.
Post-hoc analyses are very useful, and informative, tools in research, particularly where a well-conducted controlled trial generates a rich data set. Like all research designs, it is important to understand their place, and what they can show (associations) and cannot show (causation).