Citation

Schlenger WE, Williams RL, Blitstein JL. J. Am. Med. Assoc. JAMA 2005; 294(20): 2578-9; author reply 2579-80.

Copyright

(Copyright © 2005, American Medical Association)

DOI

10.1001/jama.294.20.2578-c

PMID

16304069

Abstract

The study by Dr Gould and colleagues addressed whether asking high school students about suicidal ideation or behaviors was harmful. We are concerned about aspects of the study methods. The study describes a group-randomized trial design in which 181 classrooms, containing 2342 students enrolled at 6 purposely selected New York State high schools, were randomized to 1 of the 2 study conditions. Rather than using data analytic techniques developed for group-randomized designs that we believe are the current standard, the authors concluded, based on their observation that the within-school intraclass correlation coefficients (ICCs) were all close to 0, that traditional multivariable regression models would be appropriate. Although it is arguable whether the schools in this trial should be considered to be random or fixed effects in the analysis, it is nevertheless paramount that the group randomization be explicitly accounted for.