How to write descriptive statistics?

Question

I am using logistic regression to look at the association between plaque and smoking, with adjustment for a bunch of potential confounders. In the descriptive statistics section, is it normal practice to look at the distribution of the outcome within different levels of predictor variables, or the distribution of the main exposure (smoking) among the different predictor variables? For example, for table 1, should I have Predictor variables (to the left) against Outcome (Yes/No) (at the top), or Predictor variables (to the left) against Smoking Status (Never/Ex/Regular) (at the top)?

Answer 1

The answer to your question depends on the purpose of your Table 1.

One purpose of Table 1 would be to help your readers understand more about the subjects in whom you are investigating the association between smoking and plaque, adjusted for cofounders. In this case, Table 1 could include so-called demographic variables (e.g., age, gender) and possibly the confounders you are adjusting for in your model. You could set up this table so that it has the following columns:

Table 1: Descriptive statistics.

Variable            Subjects       Subjects            All Subjects
                    With Plaque    Without Plaque
                    (n = 20)       (n = 80)            (n = 100)

Age, mean (sd)  

Gender, count (%)
   Males
   Females
   Other

Etc.

Another purpose of Table 1 would be to help readers understand more about the variables that were included in your model. In that case, your table would look similar to the one suggested above, except that would include only variables that made it into your model.

If your outcome variable were to include missing values, you would have to capture that in your table (e.g., insert a column titled Subjects With Missing Plaque Status).

For your numeric variables (e.g., Age), you should report measures of center and spread that are most suitable given the distribution of the data.