Page 50 - PATIENT REGISTRY DATA FOR RESEARCH: A Basic Practical Guide

subsequent analysis, and such a selection which shall be based on the underlying

judgment about the scientific relevance, scope and significance of the study.

i. Some of the variables may have smaller sample sizes. Researchers may have to

decide whether these variables should rightfully be regarded as the independent

variables (i.e. the factors associated with an outcome) or the dependent variables

(i.e. the outcome) for the subsequent statistical analysis, as this decision will affect

their sample size requirements, especially for multivariate analysis.

j. It is also strongly recommended conducting these data analyses by using the

programming codes because it promotes transparency. Besides that, a detailed

step-by-step mechanism of the analysis (and together with the flow diagram of its

procedures) should also be clearly documented for future reference if necessary.

k. Avoid analysing any variable which is not related to the study objective because it

is necessary to conduct statistical analysis according to a well-thought-out data

analysis plan which has been formulated to address hypotheses and aims of the

research, and also to account for possible confounders. Although it seems

tempting for a statistician to analyse all the variables at first glance, due to the

presence of so many different variables in the entire data set; however, it is always

recommended for a statistician to prepare beforehand a set of planned tables and

figures (called dummy tables) which will provide a visual presentation of the

layout of the results. This will avoid confusion among both researcher and

statistician because such dummy tables will illustrate how the results will be

displayed and also help to bring into focus of what both of them are doing. The

researcher, on the other hand, will complement the statistician by carefully

framing the research question and funnelling it down into testable hypotheses and

action steps, which are detailed in a data analysis plan. This can realistically be