0

I am doing a study proposal for a project that looks at two different measures for evaluating the presence of a health condition, one clinical exam one diagnostic exam. the clinical exam will be performed by clinican A and the diagnostic exam will be performed by clinician A (newly trained) and clinician B (expert). For both the clinical exam and the diagnostic test the response will be limited to a yes/no answer on the presence or absence of the condition in question. The clinical exam will be compared to the diagnsotic exam to look for differences in ability to document the disease status in question and the performance and interpretation of the diagnsotic exam will be compared between expert and newly trained user. For the study proposal I need to explain how I am going to establish a correlation and I have no idea how to do that. Any assistance you can provide is greatly appreciated.

Karen
  • 1

2 Answers2

1

If I've understood your research goals correctly, you have two questions you'd like to answer:

  1. Is there a difference between clinician A's ability to correctly diagnose a condition using a clinicial exam vs. a diagnostic exam?
  2. Do experts correctly diagnose the condition more often than trainees when both use a diagnostic exam?

Both of these can be answered using separate chisquare contingency tables (assuming your sample sizes are likely to be on the small side). A google search will turn up many examples of how to do this (the calculations can often be performed by hand), and there are a number of discussions on CrossValidated such as this one:

Contingency tables: what tests to do and when?

Community
  • 1
Geoffrey
  • 43
  • 6
  • Thank you and yes, my sample size is limited to 40 subjects due to the costs of the examinations. I will have a look online and see what I can come up with and will post my findings to see if I am on the right track. Thank you – Karen Mar 04 '14 at 16:50
  • You are welcome. Look for Fisher's tea tasting experiment - you'll see it extrapolates to your situation very well. – Geoffrey Mar 04 '14 at 17:27
0

You should look into metrics for quantifying agreement. If you were to take the expert judgement as "truth", you could quantify your scenario using binary classification metrics. For examples, see here.

Kyle.
  • 1,550
  • 1
  • 11
  • 22