ABSTRACT
The problem of what to do with data that appear to be erroneous, but for which no cause is apparent, has puzzled scientists for as long as data have been collected and evaluated. These data can be characterized as outliers, not appearing to be of the same kind as other data collected under the same circumstances. One might suppose that situations exist where such outliers can be considered absurd, e.g., nobody with any knowledge of the process could conceive that such a value could exist. For example, if an automatic device for weighing individual tablets would record a zero, we would be ‘‘certain’’ that the result was not due to a weightless tablet, but rather due to some malfunction of the process. However, in the great majority of cases, the cause for an outlying result cannot be ascertained. In the case of scientific experiments for research purposes, the outlier appears among other experimental results, and the scientist can freely hypothesize reasons and explanations for its presence. Thus, the scientist can make a case for exclusion or inclusion of the outlier, and discuss reasons, implications, etc., with impunity. The future will demonstrate the correctness of his evaluation and judgement; ‘‘Time will tell.’’ In a regulatory environment, time is of the essence. We cannot wait for time to prove a hypothesis about an outlying observation, correct or not. Usually, a decision must be made quickly. Although there is no absolute right or wrong way to proceed, ‘‘judgement’’ seems to be a key word. Under a given set of circumstances, what is to be done with the ‘‘outlier’’ is not easy to answer. These problems were at the heart of a recent litigation involving the Federal Government (FDA) and a generic company (Barr Labs, Inc.) (1) that involved testing of solid dosage forms or products for reconstitution. Much of the government’s case against Barr related to the passing of batches in which a single failing or outlying assay was observed. The government suggested that if a single assay was not within specifications, in the face of all other tests performed on the batch, the product should be rejected. This ‘‘outlying’’ result or test failure could occur as a result of in-process testing or final product testing, either situation resulting in the rejection of the batch. This was the point of much of the trial proceedings, with a willing judge looking for the truth. In fact, there is no truth. What is to be done is a matter of judgement and common sense, grounded in
disagree on what to do situation. Good judgement does not necessarily lead to a single universal truth. Thus, the procedures recommended in this paper represent my judgement and experience.
