# Terminology for different types of epidemiological variable

If I have a data set with $n$ patients and I have a categorical variable $X$ for each patient, I want to drop all patients whose $x$ is a particular value.

Question 1: what is the term (preferably the epidemiological term) for variable $X$? (Is it an exclusion criteria variable [I've never seen this term before, I'm just guessing], or is there a commonly used term?)

Next, I want to subset the resulting data set into two data sets according to the values of a dichotomous variable $Y$ for each patient such that all patients for which $y = 0$ will go in data set A whereas all patients for which $y = 1$ will go in data set B. I will then treat data sets A and B separately and do statistical tests within each. Following this, I would perform comparisons between A and B.

Question 2: what is the term (again preferably the epidemiological term) for variable $Y$? (I was thinking something along the lines of subsetting variable, but again, I've never seen this term.)

Let us assume we're talking about a particular categorical variable. Lets say a "residence" variable that indicates if someone lives in a private home, an apartment, a dorm or group living space, a nursing home, a prison, or other.

For your first example, I don't know that I'd call the variable anything. I'd probably end up just saying what value of the variable I ended up excluding. For example: "Subjects who were incarcerated in correctional institutions were excluded from the study."

In the second example, depending on what you're doing you might say you're "stratifying by Y". That being said, I'd never be really all that comfortable with splitting them into two entirely two data sets and analyzing one utterly without the context of the other. Unless it's just a way to partition things that really should be two separate studies - but in that case you don't need to talk about the variable, as you're really just running two separate studies whose data you happened to get in the same file.

• In the first instance, I'd run tests within A and B separately. However, ultimately, I'd want to compare the results of A with the results of B. So, perhaps stratifying variable? I just need a single, understandable term to summarize my study design in tabular format. – jetistat001 Sep 30 '12 at 8:23
• Yeah, I'd just say "stratified by Y". – Fomite Sep 30 '12 at 8:24

For the first question, there is no particular term for X. Exclusion criteria is a term used to refer to rules used to exclude patients from being enrolled in the trial. There may be an analogy here to the process of excluding data when a mathematical condition exists but it is a different situation and I have never heard the term exclusion criteria used in that case.

In my experience with clinical trials the analysis you are describing in the second question is most commonly called subgroup analysis. I imagine the epidemiologist use that term also.

• The only reason I wouldn't call it a subgroup analysis is the suggestion that A & B are treated entirely separately, rather than B being a subset of A. – Fomite Oct 1 '12 at 3:37
• The term subgroup analysis is used to define an analysis of any subset of the original data set. – Michael R. Chernick Oct 1 '12 at 10:16
• I'm not saying its wrong, I'm saying I'd find it odd. – Fomite Oct 1 '12 at 10:18
• Why odd? Each analysis involves a subgroup of a full larger data set. – Michael R. Chernick Oct 1 '12 at 10:38
• Generally speaking, as I mentioned, I see "subgroup" analysis used to look at a particular group of interest as part of a greater whole. Partitioning the data into a number of mutually exclusive subgroups and comparing them, without any indication of a "whole study" analysis, feels far more like stratification than a subgroup analysis. – Fomite Oct 1 '12 at 10:39