Eyes or patients? Traps for the unwary in the statistical analysis of ophthalmological studies

Abstract

In reports on ophthalmological research the results of measurements on the eye are often expressed as mean and standard deviation based on m patients, n eyes (n > m). This approach leads to t tests that are invalid because the measurements on the two eyes of one subject are usually related, not independent. In a simulation study involving intraocular pressure data analysed in this way, the null hypothesis of no difference between groups was rejected at a nominal alpha = 0.05 level in 39 out of 200 simulations; thus the true alpha was nearly 0.2. This approach is excessively prone to produce false positive results.