Sample size: diagnostic test (sensitivity & specificity)
Sample size for a diagnostic-accuracy study
Use this tool when you're evaluating a new test against a reference standard and you need to estimate sensitivity and specificity with a target precision. Based on Buderer (1996).
Probability the new test correctly identifies a diseased person.
How to justify this number
Use a published estimate from a comparable population, or a pilot. If you have only a guess, set sensitivity at 0.80 — this maximises required sample (most conservative).
Probability the new test correctly identifies a non-diseased person.
Half-width of the 95 % confidence interval around Sn and Sp. 0.05 = ± 5 percentage points.
Proportion of patients in your recruitment setting who actually have the disease (by reference standard).
How to justify this number
Determines how many people you must screen to obtain enough diseased / non-diseased participants. If recruitment is by case-control design (you separately enrol cases and controls), prevalence is irrelevant — recruit n_diseased and n_healthy directly.
You need
What does this calculation actually do?
Buderer (1996) treats sensitivity and specificity as proportions estimated from disjoint subsets (diseased vs non-diseased):
n_diseased = z² · Sn · (1 − Sn) / w² n_healthy = z² · Sp · (1 − Sp) / w² n_total = max( n_diseased / prev, n_healthy / (1 − prev) )
The total reflects the number of people you must screen to
naturally accrue enough diseased and non-diseased participants. With a
case-control design (separate streams), recruit
n_diseased cases and n_healthy controls
directly — ignore the total.
References: Buderer NMF. Acad Emerg Med 1996;3:895–900. · Hajian-Tilaki K. J Biomed Inform 2014;48:193–204. · Bujang MA, Adnan TH. J Clin Diagn Res 2016;10:YE01–YE06.