Terms for comparison groups

[chapb]

Across the iAM.AMR project, we use the term factor to describe "a practice or circumstance that influences the occurrence of AMR". We quantify the impact of a factor on the occurrence of AMR using a measure of association – the odds ratio – to compare the odds of resistance in two groups: those who experience or are exposed to the factor, and those who don't, or who aren't exposed to the factor.

But what are the most appropriate names or terms for the two groups we are comparing? More verbosely:

What is the most appropriate, and broadest term for each comparison group from an experimental or observational study, used in the calculation of a measure of association? Specifically, in the context of the iAM.AMR, what are the best terms to use for a factor's comparison groups (for which we calculate an odds ratio)?

Existing Terms

The existing terms used are: "exposed" and "referent".

Suggested Replacement

The suggested replacements are "comparator" and "referent".

Other options include:

• treatment/intervention/exposed & control/comparison
• dependent & independent
• antecedent & consequent

[chapb]

I like "comparator" and "referent", because they're abstract and we get to define them ourselves. This means we avoid some pitfalls with other terms, but they aren't intuitive. As was pointed out in last meeting -- and apologies, I didn't understand what was meant at the time --, but yes, they could easily mean the same thing, given they're rarely used and not formally defined.

Here are some thoughts on other options:

exposed group

• As Ben suggested, exposed is a commonly used and understood term in the risk literature
• sounds weird when looking at the inverse comparison in the model: "the exposed group received ceftiofur" works, but "the exposed group received no ceftiofur" is a an exposure to a non-exposure
• generally implies the factor is exogenous, which is weird in comparisons of innate features like "the exposed group is breed x, sex y", "the exposed group is ages i - j"

intervention or treatment group

• similar problems as exposed, and not applicable outside of experimental studies
• implies control over group allocation (not applicable in natural experiments, some observational studies)
• also implies the "treatment" group is a positive "treatment"

dependent group & independent group

• also abstract, but maps onto regression, operations on a 2x2 table
• implies causation, not correlation

[BenSmithRA]

Can you please clarify for my understanding, particularly point C, using a simple example?

Study A reports on the odds of AMR in barns with boot brushes vs no boot brushes.

A. In this case, would you call the boot brush group the "comparator" group, and the boot brush-less group the "referent" group (assuming that is what the study authors reported)?

B. In the model, if one wanted to examine the effect of 100% of barns using boot brushes, one would use the odds in the study's comparator group over the odds in the referent group?

C. If one wanted to model the opposite (no boot brushes) on exposure, one would use the inverse (odds in boot brush-less group over odds in boot brush group)? How would one write that up in text, say, when describing their model methods in results? Would it be something like: "we examined the impact of no boot brush use 100% of the time, so the odds ratio was calculated as (Study A's) referent group over the comparator group?

[chapb]

A) Yep, the boot brush group would be the comparator, vs the brush-less group as the referent (assuming that's as reported).

B) Correct!

C) Yes, if we went to use the factor in the "alternate" direction (vs "as described"), we'd take the inverse. In the results we'd say something akin to "To consider the impact of eliminating the use of boot brushes, we set 'brush use' as the referent group, and took the reciprocal of the odds ratio to describe the change in odds of resistance when not using boot brushes".

[gjin5]

Sorry I'm late on this, just wanted to throw in my 2 cents. Of the options listed I think "comparator" and "referent" are the best and I agree with @chapb on the reasoning. I also think it's quite intuitive actually and the flexible/subjective definition plays well with the fact that we use odds ratios in the model ie, the comparator can be the referent and vice versa if we choose so.

[cp-murphy]

Hello everyone
Throwing a wrench into this conversation. Despite our earlier decision, upon reflection, I wonder if we want to consider calling the "comparator" group simply the "factor" group with the "referent group". There will be a need to define these in the papers (Charly working in definitions for her paper) and if we already define factor (official definition: a measured observation and it can be extended to a measured observation with a potential association or relationship to antimicrobial resistance). By extending factor to "factor group" it will harmonize the language for the audience. I think it will also work for situations where the group allocation changes (e.g., factor/factor group could be drug use compared to referent of no use; alternative could allocate the factor/factor group could be absence of drug use compared to referent of drug use). Happy for thoughts and we can run this by the larger group. I think the conversation will go better when we have definitions.

[chapb]

Going with "factor group" and "comparator" based on the iAM meeting on 2021-06-01!

We'll work through the updates through the documentation. The models are a different story though; this will take some time to figure out the least intrusive method of updating it.