I intend to study the growth rate of diseases among patients of different age groups (cohorts) and comment on the growth of diseases in the next 5 years. Can I incorporate vital sign readings such as body temperature, systolic blood pressure, etc. in the growth model? Is any other approach more suitable for this kind of analysis?
$\begingroup$
$\endgroup$
2
-
$\begingroup$ may be this is biological question in addition to statistical question ? $\endgroup$– JohnCommented Jul 24, 2014 at 7:49
-
$\begingroup$ @John I would keep it as a statistical question $\endgroup$– darkageCommented Jul 25, 2014 at 5:18
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Statistically you can use any variable in the model, may not be causal or no reason to be causal but relevant. But question how good / informative model you will be able to get? Does this make any sense fit such model? Will this model be stable ?
For example a model with human weight (y) as function of distance between earth and moon (x), may not seen biologically reasonable and you may get slope ($\beta$) ~ 0. your predictably will go to 0. It does not make sense do so either. Even if you get $\beta$ >0 that may be statistically insignificant or merely due to sampling error and will be unbelievable at least by non-statisticians. Another question is how well such model works in cross validation ?
So my suggestion would be (1) you can fit all variables and see which variables have higher prediction potential (2) finalize the variables you want to use (3) discuss with biologists what would be implication of using such model and with consultation make final list.
