Lewei Lin

Date of Award

January 2011

Document Type

Open Access Thesis

Degree Name

Medical Doctor (MD)



First Advisor

Robert Rosenheck

Subject Area(s)



HYPOTHESIS/AIMS: Apply health state analysis to a large clinical trial dataset of patients with schizophrenia to provide added insights into clinical and quality of life characteristics. We also evaluated the hypothesis that there is no difference in the distribution of health states of patients across several antipsychotic medications. METHODS: This study was a secondary analysis of data from the CATIE trial, a multi-site double blind clinical trial of antipsychotic treatments for schizophrenia. We applied K-means clustering to the CATIE data, creating discrete clusters with symptom and side effect characteristics that were then validated using a panel of quality of life measures. A comparison was made across medications for differences in cluster distributions at baseline and 6 months. RESULTS: 1049 patients from the CATIE trial dataset were included for initial cluster analysis. By examining cluster profile plots, it was determined that 5 was the optimum number of health states. Using intent to treat, the model was applied to compare 6-month outcomes for patients on perphenazine, olanzapine, risperidone and quetiapine. Chi square tests of independence showed significant difference (p=0.0090) in the distribution of patients across health states for the 4 medications at 6 months. Chi squared pairwise comparisons were significant for only perphenazine vs. risperidone (p = 0.012 < alpha of 0.025 with Hochberg correction) and for olanzapine vs. risperidone (p= 0.0010 < alpha of of 0.05) but not for any other pairs. At baseline, almost 20% of patients were in the worst health state (HS+Dp+Ak), but decreased at 6 months, with the greatest decreases in the pherphenazine (9.2% decrease) and olanzapine (11.1%) groups compared to risperidone (4.7%) and quetiapine (6.7%). There was a large increase in the best health state (LS+LSE) for patients taking perphenazine (15.0%), olanzapine (18.5%) and quetiapine (12.0%) but less for patients taking risperidone (4.5%). CONCLUSION: This study demonstrated health state analysis is a useful tool that provides information on the overall clinical state of patients and can potentially be used to help guide clinicians in treatment decisions.


This is an Open Access Thesis.

Open Access

This Article is Open Access