Recursive Partitioning for Latent Recurrent Events: Application to Multiple Sclerosis Data

• Activity Number: 566
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Shannon Stock*+ and Brian C. Healy and Victor G. DeGruttola
• Companies: Harvard University and Massachusetts General Hospital and Harvard School of Public Health

Abstract

The progression of autoimmune diseases often involves jumps in disease severity followed by periods of stability. In the case of multiple sclerosis, the Expanded Disability Status Scale is used to track and classify patient disease progression. However, since patients may experience small exacerbations of their disease severity that eventually subside, there is uncertainty whether an observed increase in EDSS is due to an actual jump in the disease process. Therefore, analyses involving such data must accommodate a latent recurrent response. In this paper, we introduce a recursive partitioning method that accommodates latent recurrent events by weighting observed changes in outcome measurements by the probability one or more events occurred since the previous time of measurement. We obtain estimates of these probabilities using the Expectation-Maximization algorithm. This method is evaluated using simulation studies and illustrated using a motivating data set involving multiple sclerosis patients.