Effect of different methods for outlier detection and rejection when calculating cut off values for diagnosis of lupus anticoagulants

Background: Lupus anticoagulant (LA)-detection is performed by testing plasma with activated partial thromboplastin time (APTT)-derived and dilute-Russell-viper-venom (dRVV)-derived tests. Results are interpreted by comparison to cut-off values determined by testing plasma from healthy-subjects. Several issues are concerned with the determination of LA cut-offs. Among them, the identification/rejection of outliers (i.e., aberrant values) prior statistical analysis is one of the most important and poorly defined. Objectives: The primary aim of this collaborative study was to evaluate whether outliers' identification/removal by using three algorithms prior cut-off calculation would have any effect on the between-laboratory variability of cut-offs. As a secondary aim, we evaluated whether and to what extent outliers' removal would affect LA-detection rate. Methods: Data sets stemming from 120 plasmas from healthy-donors, collected and tested at each of 11 laboratories were evaluated by three algorithms for outliers' identification. Furthermore, a well characterized set of plasmas certified as being positive at increasing LA-potency, were concomitantly tested and their results used to assess any effect of outlier detection/rejection on LA-detection rate. Results: Relatively few outliers were identified and their elimination (regardless of algorithms) showed no appreciable effects on the inter-laboratory variability of cut-offs nor on the LA-detection rate, indicating that outliers are not the main cause of the inter-laboratory variability of cut-offs for LA-detection. Conclusions: These results strengthen the recommendation that cut-offs should be determined locally after outlier removal (to avoid inclusion of gross, obvious outliers) and that they cannot be interchangeably used in other laboratories even when using the same platform.