Inference and Prediction of Atypical Response to Major Depressive Disorder Treatment with C-Reactive Protein and Interleukin-6 Inflammatory Markers Open Access

Smith, Madoc (Spring 2020)

Permanent URL:


Background: Major depressive disorder is a common mood disorder with complex patient response patterns over time. A patient’s improvement may vary across the treatment period. We aim to predict patient response patterns in a clinically meaningful manner using inflammatory markers historically associated with depression.

Methods: In the PReDICT study (Dunlop, 2012), 316 patients were randomly assigned to three equally effective major depressive disorder treatments over a 12-week treatment period. Clinical and biological measurements were taken to assess possible predictors of patient response to treatment. We classified patients into four groups based on patient responses in early and late treatment intervals. After verifying response group patterns, we conducted pairwise comparisons of potential confounding clinical measures to control in logistic models that predict disagreement between early and late response using C-Reactive Protein (CRP) and Interleukin-6 (IL6). To better predict MDD symptom recurrence and late sustained response, we constructed minimum deviance cross-validation classification trees with inflammatory marker concentrations and potential confounders.

Results: For MDD symptom recurrence, previous antidepressant trials (P=0.050), anxiety diagnosis (P=0.104), baseline Hamilton anxiety score (P=0.077) and employment status (P=0.023) were identified as potential confounders. For late sustained response, age (P=0.064) was identified. When modeling recurrence, there is a relationship between CRP and response group (OR=1.42; P=0.053). The large effect size of IL6 (OR=0.62 P=0.162) suggests a relationship that we are underpowered to detect.

Conclusions: Unemployment and concurrent anxiety may be associated with increased likelihood of patient experience MDD symptom recurrence. Young age may be associated with increased likelihood of a patient experiencing late sustained response. CRP and IL6 appear to have a relationship with a patient experiencing atypical response patterns to MDD treatment. Prediction efforts poorly classified recurrence and late sustained response correctly likely due to small sample size in the recurrence (N=28) and late response (N=20) groups. Tree classification prediction rate was typically better than logistic models at the expense of inferential power.

Table of Contents

1. Introduction

4. Methods

12. Results

18. Discussion

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
Subfield / Discipline
  • English
Research Field
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files