10-22-2012 11:26 PM
I couldn't understand the following content of statistical method with yellow background in the paper:
We evaluated the continuous association between eGFR using both equations and incidence rates of clinical outcomes using a Poisson regression model incorporating linear spline terms for eGFR (knots at 45, 60, 75, 90, and 105 mL/min/1.73 m2) with and without adjustment for age, sex,and race.
note:eGFR (estimated Glomerular Filtration Rate) could be calculated using different equations based on serum creatinine, age, sex, race.
please explain the statistical method and thanks a lot.
In addition, I have another question about how to get adjusted incidence rate ratios for all-cause mortality after adjusting for some covariates like age, sex, diabetes, etc. As far as I am concerned, I could perform Logistic/Cox regression for death or other outcome (Y: die-0, and survive-1; X: covariates). But I have no idea about incidence rate ratio for all-cause mortality using regression.
please give some advice about it. Thank you so much.
10-23-2012 10:46 AM
You can do a Poisson regression using GENMOD. In the simplest terms, it means using a Poisson distribution for the error terms rather than the normal one. More details are in the GENMOD references.
Here is an introduction to splines
You can find more by just searching support.sas.com for linear splines.
I need more context for your second question; it is too unclear to comment on. Ratio between 'what' rates? Logistic regression would give you an estimate of one rate.
10-23-2012 10:32 PM
Thank you so much for your answer to the first question. It is very useful and important for me to learn spline further.
For the second question, I attach a figure of table from the paper that described the multivariate analysis of Incidence Rate for Outcomes like end-stage renal disease, all-cause mortality, stroke, etc. In this context, eGFR (estimated Glomerular Filtration Rate) could be calculated using different equations (MDRD equation and CKD-EPI equation) based on serum creatinine, age, sex, race. So, there were two different eGFR values for one patient. eGFR was divided into 5 groups (>=120, 119-90, 89-60, 59-30, <30) and the group of 119-89 was defined as reference group in the multivariate analysis. I was confused with how to adjust incidence rate of outcome like all-cause death or Stroke. Just as you said, Logistic regression would give me an estimate of one rate. what is y in equation of logisitic regression for estimate of one rate? I know that y is usually variate (Yes or No, 1 or 0, death or survival) . please give some advice about it.please give some advice about it. If the information of figure is not enough, I attached the fulltext of the paper
10-24-2012 09:21 AM
If a case/each observation/ a subject was related to differert time like different follow-up time, We would choose poission regression rather than logistic regression. Is it right?
10-25-2012 09:44 AM
They used Poisson regression to compute the risk adjusted incidence rate of each outcome. The ratio part is in comparing the adjusted rate for a category to the reference rate.
If you want more on the methods used, you should to contact the authors directly; as they didn't even use SAS for the analysis.