The Signature: A Never-Sent Letter to the Editors of Neurology, Updated With Commentary and Guidelines for the Future

A recent paper in Neurology by Annette Schrag et al. [1,2] with an accompanying editorial by D.L. Gilbert and R. Kurlan [2] caught my attention as being a perfect example of what modern medicine is doing, right and wrong in its current policies regarding the statistical treatment of patient data. Schrag et al. did a magnificent job surveying a database of 678,862 UK patients to search for associations between streptococcal infection and certain neuropsychiatric disorders including tics, Tourette syndrome and obsessive-compulsive disorder collectively termed PANDAS (Pediatric Autoimmune Neuropsychiatric Disorders Associated with Streptococcus B infection; Swedo et al. [3]). Schrag et al. [1] used conditional logistic regression to identify controls matched for age and sex and wound up with 254 cases and 2,211 controls. They sliced and diced this data for such other variables as antibiotic treatment and ethnicity, but in the end could not confirm a statistically significant association between streptococcal infection and the PANDAS [1].