Efficient Study Designs to Assess the Accuracy of Screening Tests

Evaluating a screening test often requires estimation of test sensitivity and specificity with appropriately narrow confidence intervals and at least cost. If the major cost is the reference ("gold") standard, savings arise from reducing the large number of test negatives that are verified by the reference standard. On the basis of the formulae of Begg and Greenes (Biometrics 1983;39:207-15), the authors determine the optimal sampling strategy for test positives and test negatives to minimize the total sample size that needs to be verified for a given confidence interval width for sensitivity. Unless sensitivity is very high, verifying more test positives and fewer test negatives than would occur with equal sampling fractions is appropriate. For example, if the sensitivity is 0.7 and the specificity is 0.99, the optimal sampling strategy is for 6.2% of those verified to be test positives, compared with 1.7% in the case of equal sampling fractions. At a disease prevalence of 0.01, the 3.3-fold increase in test positives results in a saving of about 15% in the test negatives and 11% in the total verified sample size. Overall, savings are about 50% for a sensitivity of 0.3, but are negligible when sensitivity is greater than 0.8. Optimal sampling strategies for sensitivity do not materially alter confidence intervals for specificity. Figures are presented from which readers can easily obtain the optimal sampling strategy given an estimate of specificity, approximated by the proportion of screenees who are test negative, and the range of likely sensitivity.