Mortality risk model creation

#1
Hi there,

I have a few questions to throw out there. A bit of back ground. I am starting to build a multiple regression model for looking at the risk factors associated with in hospital mortality (eg age, sex, race etc [all catagorical data]). Firstly, I wish to carry out univariate analysis on each factor and see how this is related to death. I wish to generate a Pearson Chi 2ed value , odds ratios and 95% confidence intervals. I will then test their independence using stepwise logistic regression. Once established, these factors will be used to put together the multiple regression model.

Ok, so far I have been using the CROSSTABS command to construct 2x2 tables in SPSS which gives the Chi 2ed value, odds ratios and 95% CI. I think this is the best way to do this. Any suggestions or input????

I am having some trouble with the stepwise logistical regression though. I am using DIED as the dependant and my catagories as the independant variables. Any suggestions??

So, my questions now!!!!!!

1) Is there a better way to doing the univariate analysis other than with the CROSSTABS in SPSS??? I can do it just fine, but was wondering for future reference??

2) Any suggestions with the logistic regression???

3) Under the GLM heading, there is UNIVARIATE ANALYSIS. Although I have done some research, can anyone help me with this analysis I.e how to use and what for???

4) Here's my last one: I have data about many risk factors from approx 100, 000 patients. What I have available for example is the percentage of death for each group. Eg

n= 100, 000
% %died
age= 50 65 1.2
sex
male 45 12.0
female 55 58.6

So I have summary of the data for 100, 000 in the above form, not case by case. I know that one of the assumptions for statistcal analysis is that the data is random and unprocessed, but am I able to carry out my experiments (as explained above) using this data. If so how??? Any suggestions??? All the data is stored on paper records. I am just not in the mood to be entering data for this number of cases by 100 varibale risk factors!!!

I am using SPSS FYI

Cheers all,

lhcpr
 
#2
Formatting error

Sorry, the formatting turned out really bad.

It is like this

n= 100, 000

age= 50: % (65) died %(1.2)

sex
male % (45) male died (12.0)
female % (55) female died (1.2)

I know it looks bad but couldn't get the formatting right!!!
 
#3
Hi lhcpr,

Welcome to the forum.

SPSS has specialized tabs/functions for regression. If you click Statistics > Regression > Linear, you can enter the independent and dependent variables in the dialog box and do the regression. This link may be helpful:

http://www2.chass.ncsu.edu/garson/pa765/regressa.htm

For logistic regression, follow Analyze > Regression > Binary Logistic. Check out:

http://www2.chass.ncsu.edu/garson/pa765/logispss.htm

There are OCR (Optical Character Recognition) software for transferring your paper records to electronic format.