I am currently designing a trial to evaluate the performance of a diagnostic test by comparing it against an established standard. The outcomes are ordinal in nature, integer values ranging from 0 to 4. I understand that for binary outcomes, estimating sample size is very straightforward, however i have not found simple explanations on how to calculate the required sample size for ordinal multi class comparisons.
Your sample size requirements (needs) will depend on the chosen inferential test for the assessment of your primary endpoint. As such, the question is vaguely formulated. What kind of test do you intend to use for your analysis? Depending on your assumptions (or knowledge) about the characteristics of the outcome you might use a Mann-Whitney U test or (under certain assumptions) even a t-test (i.e. if the intervals among the outcome responses may be treated as reasonably equal for practical purposes). For the former, you may read the following recent paper: https://arxiv.org/pdf/1805.12249.pdf. For the t-test there is plenty of info freely available on the internet (you may look at this, for instance: https://pdfs.semanticscholar.org/9c63/2970afa8e6dc84db1dc4b7e77ab1a667c2ca.pdf)
Hi Abhishek Aravind
As the goal is to establish inter-rater agreement between two instruments, I suggest the use of Weighted Kappa statistic. For detailed justification see: Article Common pitfalls in statistical analysis: Measures of agreement
For sample size calculations and pre-calculated sample size tables, refer to Article Guidelines of the minimum sample size requirements for Cohen’s Kappa
Indeed, Usman Rashid, using Kappa might be a very sensible option, depending on the design of the trial.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size