Thyroid Adenocarcinoma: Correlation between molecular cancer subtypes and selected clinical features
Maintained by TCGA GDAC Team (Broad Institute/Dana-Farber Cancer Institute/Harvard Medical School)
Overview
Introduction

This pipeline computes the correlation between cancer subtypes identified by different molecular patterns and selected clinical features.

Summary

Testing the association between subtypes identified by 3 different clustering approaches and 4 clinical features across 137 patients, 8 significant findings detected with P value < 0.05.

  • 3 subtypes identified in current cancer cohort by 'METHLYATION CNMF'. These subtypes correlate to 'Time to Death',  'AGE', and 'HISTOLOGICAL.TYPE'.

  • CNMF clustering analysis on sequencing-based miR expression data identified 3 subtypes that correlate to 'Time to Death',  'AGE', and 'HISTOLOGICAL.TYPE'.

  • Consensus hierarchical clustering analysis on sequencing-based miR expression data identified 3 subtypes that correlate to 'Time to Death' and 'HISTOLOGICAL.TYPE'.

Results
Overview of the results

Table 1.  Get Full Table Overview of the association between subtypes identified by 3 different clustering approaches and 4 clinical features. Shown in the table are P values from statistical tests. Thresholded by P value < 0.05, 8 significant findings detected.

Clinical
Features
Statistical
Tests
METHLYATION
CNMF
MIRseq
CNMF
subtypes
MIRseq
cHierClus
subtypes
Time to Death logrank test 0.0279 0.019 0.00563
AGE ANOVA 0.000287 0.0441 0.162
GENDER Fisher's exact test 0.862 0.405 0.699
HISTOLOGICAL TYPE Chi-square test 9.59e-09 1.65e-07 7.31e-07
Clustering Approach #1: 'METHLYATION CNMF'

Table S1.  Get Full Table Description of clustering approach #1: 'METHLYATION CNMF'

Cluster Labels 1 2 3
Number of samples 39 20 78
'METHLYATION CNMF' versus 'Time to Death'

P value = 0.0279 (logrank test)

Table S2.  Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #1: 'Time to Death'

nPatients nDeath Duration Range (Median), Month
ALL 137 1 0.1 - 66.1 (9.0)
subtype1 39 1 0.2 - 65.9 (7.6)
subtype2 20 0 0.1 - 66.1 (6.4)
subtype3 78 0 0.2 - 66.1 (12.3)

Figure S1.  Get High-res Image Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #1: 'Time to Death'

'METHLYATION CNMF' versus 'AGE'

P value = 0.000287 (ANOVA)

Table S3.  Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #2: 'AGE'

nPatients Mean (Std.Dev)
ALL 137 45.4 (16.3)
subtype1 39 53.7 (17.6)
subtype2 20 38.1 (13.0)
subtype3 78 43.2 (15.0)

Figure S2.  Get High-res Image Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #2: 'AGE'

'METHLYATION CNMF' versus 'GENDER'

P value = 0.862 (Fisher's exact test)

Table S4.  Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #3: 'GENDER'

nPatients FEMALE MALE
ALL 38 99
subtype1 12 27
subtype2 5 15
subtype3 21 57

Figure S3.  Get High-res Image Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #3: 'GENDER'

'METHLYATION CNMF' versus 'HISTOLOGICAL.TYPE'

P value = 9.59e-09 (Chi-square test)

Table S5.  Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #4: 'HISTOLOGICAL.TYPE'

nPatients OTHER THYROID PAPILLARY CARCINOMA - CLASSICAL/USUAL THYROID PAPILLARY CARCINOMA - FOLLICULAR (>= 99% FOLLICULAR PATTERNED) THYROID PAPILLARY CARCINOMA - TALL CELL (>= 50% TALL CELL FEATURES)
ALL 7 76 39 15
subtype1 4 7 26 2
subtype2 1 14 4 1
subtype3 2 55 9 12

Figure S4.  Get High-res Image Clustering Approach #1: 'METHLYATION CNMF' versus Clinical Feature #4: 'HISTOLOGICAL.TYPE'

Clustering Approach #2: 'MIRseq CNMF subtypes'

Table S6.  Get Full Table Description of clustering approach #2: 'MIRseq CNMF subtypes'

Cluster Labels 1 2 3
Number of samples 32 46 31
'MIRseq CNMF subtypes' versus 'Time to Death'

P value = 0.019 (logrank test)

Table S7.  Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

nPatients nDeath Duration Range (Median), Month
ALL 109 1 0.1 - 66.1 (8.1)
subtype1 32 1 0.3 - 47.4 (7.8)
subtype2 46 0 0.2 - 66.1 (19.2)
subtype3 31 0 0.1 - 65.9 (6.8)

Figure S5.  Get High-res Image Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #1: 'Time to Death'

'MIRseq CNMF subtypes' versus 'AGE'

P value = 0.0441 (ANOVA)

Table S8.  Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #2: 'AGE'

nPatients Mean (Std.Dev)
ALL 109 45.6 (16.1)
subtype1 32 51.2 (18.4)
subtype2 46 41.9 (13.8)
subtype3 31 45.2 (15.8)

Figure S6.  Get High-res Image Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #2: 'AGE'

'MIRseq CNMF subtypes' versus 'GENDER'

P value = 0.405 (Fisher's exact test)

Table S9.  Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #3: 'GENDER'

nPatients FEMALE MALE
ALL 33 76
subtype1 12 20
subtype2 11 35
subtype3 10 21

Figure S7.  Get High-res Image Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #3: 'GENDER'

'MIRseq CNMF subtypes' versus 'HISTOLOGICAL.TYPE'

P value = 1.65e-07 (Chi-square test)

Table S10.  Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #4: 'HISTOLOGICAL.TYPE'

nPatients OTHER THYROID PAPILLARY CARCINOMA - CLASSICAL/USUAL THYROID PAPILLARY CARCINOMA - FOLLICULAR (>= 99% FOLLICULAR PATTERNED) THYROID PAPILLARY CARCINOMA - TALL CELL (>= 50% TALL CELL FEATURES)
ALL 6 57 34 12
subtype1 3 4 23 2
subtype2 2 32 8 4
subtype3 1 21 3 6

Figure S8.  Get High-res Image Clustering Approach #2: 'MIRseq CNMF subtypes' versus Clinical Feature #4: 'HISTOLOGICAL.TYPE'

Clustering Approach #3: 'MIRseq cHierClus subtypes'

Table S11.  Get Full Table Description of clustering approach #3: 'MIRseq cHierClus subtypes'

Cluster Labels 1 2 3
Number of samples 29 37 43
'MIRseq cHierClus subtypes' versus 'Time to Death'

P value = 0.00563 (logrank test)

Table S12.  Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #1: 'Time to Death'

nPatients nDeath Duration Range (Median), Month
ALL 109 1 0.1 - 66.1 (8.1)
subtype1 29 1 0.3 - 46.7 (7.7)
subtype2 37 0 0.2 - 66.1 (8.2)
subtype3 43 0 0.1 - 66.0 (12.3)

Figure S9.  Get High-res Image Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #1: 'Time to Death'

'MIRseq cHierClus subtypes' versus 'AGE'

P value = 0.162 (ANOVA)

Table S13.  Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #2: 'AGE'

nPatients Mean (Std.Dev)
ALL 109 45.6 (16.1)
subtype1 29 50.5 (18.4)
subtype2 37 43.8 (14.2)
subtype3 43 43.8 (15.8)

Figure S10.  Get High-res Image Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #2: 'AGE'

'MIRseq cHierClus subtypes' versus 'GENDER'

P value = 0.699 (Fisher's exact test)

Table S14.  Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #3: 'GENDER'

nPatients FEMALE MALE
ALL 33 76
subtype1 10 19
subtype2 12 25
subtype3 11 32

Figure S11.  Get High-res Image Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #3: 'GENDER'

'MIRseq cHierClus subtypes' versus 'HISTOLOGICAL.TYPE'

P value = 7.31e-07 (Chi-square test)

Table S15.  Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #4: 'HISTOLOGICAL.TYPE'

nPatients OTHER THYROID PAPILLARY CARCINOMA - CLASSICAL/USUAL THYROID PAPILLARY CARCINOMA - FOLLICULAR (>= 99% FOLLICULAR PATTERNED) THYROID PAPILLARY CARCINOMA - TALL CELL (>= 50% TALL CELL FEATURES)
ALL 6 57 34 12
subtype1 3 4 21 1
subtype2 1 26 7 3
subtype3 2 27 6 8

Figure S12.  Get High-res Image Clustering Approach #3: 'MIRseq cHierClus subtypes' versus Clinical Feature #4: 'HISTOLOGICAL.TYPE'

Methods & Data
Input
  • Cluster data file = THCA.mergedcluster.txt

  • Clinical data file = THCA.clin.merged.picked.txt

  • Number of patients = 137

  • Number of clustering approaches = 3

  • Number of selected clinical features = 4

  • Exclude small clusters that include fewer than K patients, K = 3

Clustering approaches
CNMF clustering

consensus non-negative matrix factorization clustering approach (Brunet et al. 2004)

Consensus hierarchical clustering

Resampling-based clustering method (Monti et al. 2003)

Survival analysis

For survival clinical features, the Kaplan-Meier survival curves of tumors with and without gene mutations were plotted and the statistical significance P values were estimated by logrank test (Bland and Altman 2004) using the 'survdiff' function in R

ANOVA analysis

For continuous numerical clinical features, one-way analysis of variance (Howell 2002) was applied to compare the clinical values between tumor subtypes using 'anova' function in R

Fisher's exact test

For binary clinical features, two-tailed Fisher's exact tests (Fisher 1922) were used to estimate the P values using the 'fisher.test' function in R

Chi-square test

For multi-class clinical features (nominal or ordinal), Chi-square tests (Greenwood and Nikulin 1996) were used to estimate the P values using the 'chisq.test' function in R

Download Results

This is an experimental feature. The full results of the analysis summarized in this report can be downloaded from the TCGA Data Coordination Center.

References
[1] Brunet et al., Metagenes and molecular pattern discovery using matrix factorization, PNAS 101(12):4164-9 (2004)
[3] Bland and Altman, Statistics notes: The logrank test, BMJ 328(7447):1073 (2004)
[4] Howell, D, Statistical Methods for Psychology. (5th ed.), Duxbury Press:324-5 (2002)
[5] Fisher, R.A., On the interpretation of chi-square from contingency tables, and the calculation of P, Journal of the Royal Statistical Society 85(1):87-94 (1922)
[6] Greenwood and Nikulin, A guide to chi-squared testing, Wiley, New York. ISBN 047155779X (1996)