PubMed2004
BACKGROUND: Blinded review has been endorsed by several cytology and pathology organizations as the most appropriate method for the review of cervicovaginal specimens in the context of litigation. Methods for determining the statistical validity of this method were evaluated. METHODS: First, the authors calculated the sample size needed to detect various differences in case difficulty or ease of interpretation, in which ease of interpretation is defined as the percentage of the time a case could be identified as abnormal by routine screening. Very easy cases could be identified most reliably, whereas more difficult cases were detected less regularly and less reliably. Using construct sample sizes, the authors calculated the number of abnormal reviews that may be helpful to conclude that the case's difficulty or ease of interpretation was statistically significantly different from another case of a specified difficulty. Finally, they examined the effect of using two separate cutoff parameters to make these distinctions. RESULTS: Depending on the threshold chosen, it was determined that improvements in the statistically meaningful distinctions may be made using 15 or 20 reviews. To distinguish between routine false-negative cases (ease of detection, 40%) and routine cases (ease of detection, 80%), the thresholds may be set at 5 of 10 reviews (a case that would not be detected regularly and reliably in any given laboratory) and 7 of 10 reviews (defining a case that would be identified regularly and reliably), respectively. CONCLUSIONS: The authors provide data that can be used to interpret the results of a blinded review in a statistically appropriate manner. To improve the utility of blinded reviews, the standards are defined explicitly.
Data Interpretation, Statistical↗