Help with Cox Regression

#1
Hi all,

it's my first time here, but I hope to be able to contribute to other threads soon. I'm working on a dataset at the moment, and need a little help. I am using SPSS.

I want to know whether an earlier age of onset of smoking leads to an earlier age of onset of psychosis. I was thinking Cox is the best way to go.
So time is "age of onset of psychosis", status is "status".

As a categorical covariate I have "smoking type" (split into categories of non-smoker, light and heavy).

Now, because I want to know whether age of onset of smoking effects the age of onset of psychosis, would "age of onset of smoking" be placed as a covariate too?

Am I right?
 

hlsmith

Omega Contributor
#4
I guess what I am getting at, is it a dataset where an individuals self-reported when they started smoking, how much they smoke, and date of onset of smoking? Or were these prospectively collected data and nicotine level collected, etc.?
 

Karabiner

TS Contributor
#6
I want to know whether an earlier age of onset of smoking leads to an earlier age of onset of psychosis.
There are several mistakes in that sentence, IMHO. You cannot make
causal claims such as "leads to", since you only have observational data
(and smoking, or early smoking in particular, certainly is a sign for presence
of additional risk factors), and you do not have subjects who [still] are without
psychosis, and you cannot claim much about psychosis in general, since you
only have a subgroup of people with psychosis (current in-patients, which
presumably are not representative for people with psychosis in general).

Regarding self-reported onset of smoking, if not all in-patients are smokers,
and/or if some patients started smoking after onset of psychosis, then it will be
difficult to use "onset of smoking" as covariate.

With kind regards

K.
 

hlsmith

Omega Contributor
#7
researchworker,


Per Karabiner's concerns and comments, is this a real analysis or for fun/school? Many times you can attempt to go through the motions, but results may not be consistent or unbiased for making any inferences. Also, is Karabiner correct in that you only have patients with a diagnosis of psychosis?
 

hlsmith

Omega Contributor
#9
Cox regression is built on a predicted binary outcome, which generates probabilities. You do not appear to have a binary outcome!
 

Miner

TS Contributor
#10
You could perform a separate survival analysis for the three smoking categories, then test whether the difference in parameters are statistically significant.
 

Karabiner

TS Contributor
#11
Ok, those current-in patients with a psychosis all present with nicotine use. So, psychosis in its prodromal stage "leads to" nicotine use? Or there are factors associated with psychosis, which are also associated with nicotine use. - Or whatever. Since all subjetcs use nicotine, and all of them started nicotine use before onset of pschosis, you maybe can perform a simple linear regression of age at onset of psychosis on age at onset of smoking.

With kind regards

K.