"Outlier" intraocular pressure readings: valid or invalid for statistical analysis?

Wide variations in the response to intraocular (IOP)-lowering medications may be due to the normal response of the subjects or to variables unrelated to the effect of medication, eg, noncompliance or recording errors. We used data from a previous study of medications designed to lower IOP to investigate the influence of readings showing an unusually large increase in IOP from baseline on the statistical analysis and conclusions. These readings, which could have been due to variables unrelated to the effect of medication, were defined as "outliers" by a statistical method. A statistically insignificant difference between two IOP-lowering medications became significant on excluding two subjects with outlier IOP readings. We explain how reporting results obtained with and without including outliers in the analysis may provide clinicians with more complete information.