Partitioning Around Medoids Clustering in Follicular Lymphoma Patients: Comparison with FLIPI and FLIPI2 in PFS Open Access

Kim, Chaejin (Spring 2018)

Permanent URL: https://etd.library.emory.edu/concern/etds/5138jd86g?locale=en
Published

Abstract

Background: The Follicular Lymphoma International Prognostic Indices (FLIPI and FLIPI2) are widely used prognostic indices in follicular lymphoma (FL), with some limitations because they dichotomize continuous variables and classify patients based on the number of adverse factors. In this thesis, Gower’s distance and Partitioning Around Medoids (PAM) clustering were employed to overcome these issues, and progress free survival (PFS) analysis was performed based on PAM clustering results to compare with PFS using the conventional methods.

 

Methods: The demographic and baseline characteristics are summarized descriptively using frequencies (%) and means (SD). Gower’s distance was calculated to measure dissimilarity between observations, considering that the prognostic factors of FLIPI and FLIPI2 are mixed type data. Using the distance matrix, the silhouette width was calculated to find the optimal number of clusters. Based on the results, PAM clustering was conducted. The clustering results were compared to the classification of FLIPI or FLIPI2 in terms of PFS.

 

Results: When using FLIPI prognostic factors, PAM in 3 clusters showed the smaller P-value (P- value=0.094) from the log-rank test than that in 3 risk groups of FLIPI (P-value=0.27). When using FLIPI2 prognostic factors, PAM in 3 cluster also showed the smaller P-value (P=0.03) from the log-rank test than that in 3 risk groups of FLIPI2 (P-value=0.50). The Kaplan-Meier (KM) curves and comparison tables between PAM and FLIPI or FLIPI2 indicated that, although PAM reflected the scale of FLIPI or FLIPI2 in some sense, it showed somewhat counterintuitive survival results considering the composition of patients in FLIPI or FLIPI2 scales. PFS stratified by PAM showed better differentiation in survival.

 

Conclusion:  Classification based on FLIPI or FLIPI2 versus PAM clustering provided us with different results. For both FLIPI and FLIPI2, PAM clustering showed better classification in terms of PFS. This may indicate that the issues observed in process of establishing FLIPI and FLIPI2 may indeed contribute to the loss of power to classify patients with FL. A large scale study may be warranted to validate these results.

Table of Contents

Table of Contents

1. Introduction. 1

2. Methods. 3

2.1 Patient Sample. 3

2.2 Statistical Analyses. 4

2.2.1 Descriptive Summarization. 4

2.2.2 Gower’s Distance (Gower’s Dissimilarity Coefficient). 4

2.2.3 Partitioning Around Medoids (PAM). 5

2.2.4 Kaplan-Meier Estimator. 7

2.2.5 Log-Rank Test. 8

3. Results. 9

3.1 Descriptive Summarization. 9

3.2 Classification using Prognostic Factors in FLIPI 9

3.2.1 Number of Clusters: 2. 10

3.2.2 Number of Clusters: 3. 10

3.2.3 Number of Clusters: 4. 11

3.3 Classification using Prognostic Factors in FLIPI2. 11

3.3.1 Number of Clusters: 2. 11

3.3.2 Number of Clusters: 3. 11

3.3.3 Number of Clusters: 4. 12

4. Discussion and Conclusion. 12

5. References. 16

6. Tables and Figures. 18

 

 

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.
School
Department
Subfield / Discipline
Degree
Submission
Language
  • English
Research field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Partnering Agencies
Last modified

Primary PDF

Supplemental Files