Which test/study to use?

#1
Hey Guys,

I was wondering if I could get some help. Lets say you have a group of patients who are positive for a disease based on the gold standard diagnosis (i.e. blood culture, biopsy, etc) or in other words a group of True Positives. Now you want to take these patients and for each one, look back at the screening test to see if it was positive or negative. Not only do you want to know if it was positive or negative, but you want to know the likelihood that it would be positive or that you would have a positive screen in a known diseased cohort of patients. Its a little tricky because you can only calculate true positives and false negatives. Lets also say that you are trying to compare two new screening tests and wanted to compare the old screening test and the new screening test against the cohort of known diseased patients. So in summary, here are my questions.


1. What is the best test or method to look back at the likelihood that a group of diseased patients (true positives) will have a positive screen? And how would you determine if this likelihood is good or bad?

2. Assume you would just repeat the calculation for Question #1 for a second type of screening test, how would you compare the two screening tests?


Thanks
 

hlsmith

Omega Contributor
#2
I will check out your question tomorrow, but pretty sure you don't have enough information to answer your questions. I slightly recall there is an approach, maybe matching that is called Case-"something" (not "control") that you can use for examining a question, but I do recall the particulars.
 

hlsmith

Omega Contributor
#3
LR(+) = Sensitivity / (1 - Specificity)

or ((True Positives / False Negatives) / (True negatives / False Postives))

Which you don't appear to have all required data elements if you construct the 2 x 2 classification table.
 
#4
Thank you for the reply. I do have the Sensitivity and Specificity of the new screening test if that helps. I think I can try to show how the S/S of the new screening test compares to known true positives, but I am not sure how to go about it.
 
#6
Sorry, Sensitivity and Specificity. Based on validation of the screen I have the sensitivity, specificity, Negative and predictive values and Likelihood Ratios (Positive and Negative).
 

hlsmith

Omega Contributor
#7
Hey Guys,


1. What is the best test or method to look back at the likelihood that a group of diseased patients (true positives) will have a positive screen? And how would you determine if this likelihood is good or bad?

2. Assume you would just repeat the calculation for Question #1 for a second type of screening test, how would you compare the two screening tests?


Thanks
Calculating the positive likelihood ratio as I presented in my prior post gets you this value. I use some code I have, to get the 95% CI around the value. The test is interpreted as higher odds of a positive test being in a person with the condition respective to a person without the condition. Interpretations are context based, but typically a value around 10 or greater is deamed of interest in medicine. You can calculate the +LR for both screening and see if the 95% CI don't overlap and crown the larger value superior given the goal.

There might be a way to better formulize a comparison statistic between the +LR. I know you can compared nested statistic via Chi-sq test, perhaps. But I am unsure if you applied both tests to the exact same samples.