Pharmacopsychiatry 2009; 42 - A116
DOI: 10.1055/s-0029-1240188

How to handle the PANSS in statistical analyses

M Obermeier 1, A Mayr 1, R Schennach-Wolff 1, F Seemüller 1, HJ Möller 1, M Riedel 1
  • 1Department of Psychiatry, University of Munich, Germany

Recent metaanalyses concerning efficacy of antipsychotics suggest a subtraction of 30 points to every PANSS (Positive and Negative Syndrome Scale) before analyzing outcome. By contrast, others recommend the traditional approach using unchanged scores. This analysis aims to clarify from a statistical perspective which procedure is most suitable and how outcome criteria are affected. Two real datasets were used to quantify the differences concerning proportional PANSS development, including tests on group-differences: a double-blind randomized controlled medication-trial on 289 patients with first-episode schizophrenia and a naturalistic study on 400 patients with a schizophrenic spectrum disorder. Furthermore simulated data were produced based on different presumptions in their structure concerning distribution and the relation between admission and discharge values. Differences occur considering proportional improvements and related response criteria because subtracting 30 points leads to a change of the scale level from interval to ratio scale. Adjusting the total score to the base level results in more patients being classified as responders. Significance levels of estimated treatment effects might change, depending on the baseline distribution. From a statistical point of view we recommend a subtraction of 30 points before analyzing the data, since for calculating proportional outcome measures only ratio scales are appropriate.