- Thread starter Chicken01
- Start date

It was in an assignment I did a few weeks ago but I still cant get my mind around it.

By strong criteria I mean no comorbidity.

I was here at the university of Padua but since we don't really have a good chatroom I cant really ask my fellow students for help. Thats why Im here.

Greetings,

Chicken

If we stick to comorbidity as an example, then excluding comorbid patients would mean

that factors are excluded which would (probably) not be affected by the treatment. But

these factors can independenttly affect the outcome, e.g. symptom severity. Now the

problem is, every factor independently affecting the outcome will increase variablilty of

the outcome across treatments (since allocation to groups is randomized, comborbidity

will evenly be distributed across groups). This additional variabilty will be independent

of treatment effects. So you will have more "random noise" in your outcome, which

means that difficulty increases to detect the "signal" (the treatment effect). The probabilty

to detect the signal (treatment effect) is another description of statistical power.

It is a general rule that in order to increase statistial power to detect the effect of a variable,

you can try to reduce random noise, i.e. exclude subject characteristics which increase

variability. The downside of this is, in extreme you might study populations which

have little to do with those in the real world (for example, many medications

are used by people age 70+, but studies are carried out with subjects aged < 60;

or, comorbidities are excluded, but 80% of subjects with a certain condition

do have comorbidities).

Hope that helps.

Good look to you all in Italy. Hopefully the worst will soon be over for you.

With kind regards

Karabiner

If we stick to comorbidity as an example, then excluding comorbid patients would mean

that factors are excluded which would (probably) not be affected by the treatment. But

these factors can independenttly affect the outcome, e.g. symptom severity. Now the

problem is, every factor independently affecting the outcome will increase variablilty of

the outcome across treatments (since allocation to groups is randomized, comborbidity

will evenly be distributed across groups). This additional variabilty will be independent

of treatment effects. So you will have more "random noise" in your outcome, which

means that difficulty increases to detect the "signal" (the treatment effect). The probabilty

to detect the signal (treatment effect) is another description of statistical power.

As a general rule, in order to increase statistial power to detect the effect of a variable,

you can try to reduce random noise, i.e. exclude subject characteristics which increase

variability. The downside of this is, in extreme you might study populations which

have little to do with those in the real world (for example, many medications

are used by people age 70+, but studies are carried out with subjects aged < 60;

or, comorbidities are excluded, but 80% of subjects with a certain condition

do have comorbidities).

Hope that helps.

Good look to you all in Italy. Hopefully the worst will soon be over for you.

With kind regards

Karabiner