Testing significance of infection rate decrease?

AP8

New Member
#1
Good morning, everyone.

I'm just starting out on a MSc in biomedical research. For my project, I'm looking into a recent seemingly impressive fall in central line-associated infection rates. My research mentor thinks my first step should be to establish whether or not the change that was seen was significant.

I have the reported rates for about the last 10 years. Could someone please give me some advice on how to calculate a p-value for a fall in these infection rates?

Thank you very much!
 
#2
Good morning, everyone.

I'm just starting out on a MSc in biomedical research. For my project, I'm looking into a recent seemingly impressive fall in central line-associated infection rates. My research mentor thinks my first step should be to establish whether or not the change that was seen was significant.

I have the reported rates for about the last 10 years. Could someone please give me some advice on how to calculate a p-value for a fall in these infection rates?

Thank you very much!
What does your advisor mean as “significant”? I’m not looking for the vague answer of statistical significance. Maybe he means clinically important or maybe me means, “is there some kind of real change that has occurred and what is our evidence against the same old same old?” Being in a similar field, medical people don’t usually know what they mean when they say something’s significant or not nor do they understand the implications of either. I would also consider regression to the mean which is a well known and real phenomenon.
 

AP8

New Member
#3
Thanks for your response.

Basically, central line infection rates were pretty high for a number of years and then a number of policies were implemented (nursing education, enforcement of policies around central line maintenance and handling, resident education) and the rates immediately following these interventions, the rates fell to about half of what they had been previously.

My mentor wants me to find a p-value for the change in the post-intervention infection rate to establish weather the fall in infection rates was 'real'.

Thanks again for your help.
 

hlsmith

Omega Contributor
#5
In essence you are concerned with slopes. Tell your mentor you will not provide a pvalue, but slopes or slope difference estimates. You may get in to trouble for this problem if there were lots of interventions spread across time, since you may need to control for each individually. Is your data, rates per month for about 120 months?

Yes, time series or interrupted times will likely need to be implemented.
 
#6
Without a control and with many changes implemented, it will be hard to determine whether regression to the mean is at play (might very well be) and if not regression to the mean, then which interventions were related to the outcome.