16 Jul How to prepare a file for statistical analysis of a clinical study?

1- Before collecting the data, take time to think about which data are really going to be of interest

Too often, the number of variables to be collected is much higher than those actually analyzed. This considerably increases the work of the person in charge of the collection (which is always tedious), and the risk of obtaining missing data that could reduce the power of the study. For a “small” study, it is not unreasonable to have less than 10 columns! Avoid open comments, but classify them a priori. If this is not possible, perform this operation once the data entry is complete (see below).

It is also important to limit the number of classes for a non-numerical column: the ideal is to have less than 5 classes, and never more than 10.

2- Only one row per patient, only one characteristic per column

And that’s it: you don’t want to have any summary lines or tables inside your Excel sheet. There should also be no merged columns or rows.

3- The first row must be the name of the variable

We recommend that you use an explicit and short name. This will allow you to produce graphics without having to rewrite the axis titles. The statistical software pvalue.io allows names up to 40 characters (truncated beyond) and supports spaces.
The classes of categorical (non-numerical) variables should also be explicit (not numbers). By default, pvalue.io considers that variables with less than 5 different values are categorical variables.

4- No column is the opposite of another

In the example below, it is not necessary to have both columns. A single column “Sex”, coded “Male” for men and “Female” for women is more appropriate.

5- There cannot be any two columns, one of which is a transformation of another

In the example below, “Age in class” is a transformation of “Age”. Only one should be kept; in this example “Age” (see rule #6).

6- Do not code numeric variables in classes

It is always preferable to keep a numerical column as long as the conditions of statistical methods are met. There is indeed a significant loss of information through this transformation.

This rule allows an exception if the thresholds are defined in the literature or usually used.

7- No unit of measurement or percentage in a numerical column.

As a general rule, no non-numeric characters should be put in a numerical column. For example, do not write 10% or 134mmHg.

8- Keep the same spelling for the classes of categorical variables

Your statistical software may be very smart, it will still identify MI and myocardial infarction as two different classes.

9- Leave the cells empty for missing data

Do not use other values such as “?”, “.”, NA, etc.

10- Do not use any personal data

Ideally, you should use an unique identifier for each of your patients, unrelated to its name or birth date. This identifier can simply be a sequential inclusion number in the study. To find the patient who matches this identifier, you can use another Excel file or write on plain paper (but be careful not to lose it) with two columns: the patient’s name and the identifier. This file must be kept safe in the department from which the patients are extracted and is not allowed to be moved or copied.