choosing age, BMI and tumour size to select a historical cohort


New Member
Hi all ,
I have very basic knowledge of statistics and I would be grateful if someone can guide me
please on how I can use SPSS to select a historical cohort?

I have retrospective data with two groups
1. group of patients who had laparoscopic surgery for cancer n=17 which is what we are interested in
2. group of patients who had open surgery for the same cancer n= 35
from the open group I would like to select a sub group based on age , BMI and tumour size that will match the age, BMI and tumour size of laparoscopic group.

Once I have selected the sub group according to the matching criteria I would like to compare lap group vs the sub group of open surgery to see if there is any difference in outcomes between the two groups in terms of complications , survival ,recurrence in cancer etc.

Can anyone suggest how to doI go about it ? Specially what tests I need to do on SPSS to choose the subgroup?



Less is more. Stay pure. Stay poor.
Whoa, you are asking a lot of this small dataset. You could fit a logistic model regressing treatment approach on age, BMI, and tumor size. Then create inverse propensity treatment weights to match on. You will need to select a caliper value (distance) between IPTWs to match on. I would also think about just using the IPTWs in model instead of matching, if there is reasonable overlap (visualized in overlaid histograms for weights by treatment group) in the IPTWs.

However, I will say you can't look at all of those outcomes, that would be even more ridiculous than trying to match in this small sample. Because if you go the matching route you sample will likely only get smaller. Or you could look at all of these outcomes, but you would be obligated to penalized your level of significance to prevent false discovery. Also, it looks like you will need to use a time to event for modeling the outcome. Thus given something like recurrence you will also need to control for competing risks, if applicable. I went to do this same thing last week with 750 surgical outcomes, and after applying a sparsity penalty my confidence intervals were obnoxiously wide, and I didn't have to adjust for false discovery. I think you likely should abort since you won't have enough data for meaningful generalizations. What did you say you were going to do in your IRB/Ethic Committee protocol?


New Member
Thanks a lot for such a quick response but I have to say I was hoping an answer for statistical idiots not stats wizards
any chance you can simplify in small steps and simple terms ?
Or is that asking too much as well ?


New Member
AS far as IRB protocol is concerned we did not apply for it as it is retrospective data on patients who already had intervention so IRB/ethic committee approval is no necessarily needed. Patient identifiable data will not e presented so once again no need for that . Although I do understand some journals will specifically ask for it at the time of publication