Table 3 Time-to-event melanoma recurrence prediction by GB-T and RF-T models.

From: Prediction of early-stage melanoma recurrence using clinical and histopathologic features

Data

Sample size

Validation

GB-T

RF-T

Time-dependent AUC

Concordance index

Time-dependent AUC

Concordance index

Original cohortsa

MGB:

215 recurrences

645 non-recurrences

DFCI:

95 recurrences

285 non-recurrences

Internal

0.853

CIe:0.849–0.857

0.820

CI:0.816–0.824

0.845

CI:0.841–0.850

0.813

CI:0.809–0.816

External

0.820

CI:0.819–0.821

0.809

CI:0.808–0.810

0.810

CI:0.809–0.811

0.807

CI:0.807–0.808

p-valued

<0.001

0.012

<0.001

0.248

Core complete cohortsb

MGB:

142 recurrences

516 non-recurrences

DFCI:

74 recurrences

276 non-recurrences

Internal

0.829

CI:0.819–0.840

0.788

CI:0.783–0.793

0.788

CI:0.783–0.793

0.783

CI:0.775–0.791

External

0.804

CI:0.802–0.806

0.808

CI:0.807–0.810

0.781

CI:0.778–0.783

0.793

CI:0.791–0.795

p-valued

0.001

0.001

0.057

0.312

Complete cohortsc

MGB:

116 recurrences

365 non-recurrences

DFCI:

45 recurrences

175 non-recurrences

Internal

0.808

CI:0.800–0.816

0.768

CI:0.760–0.777

0.795

CI:0.786–0.804

0.769

CI:0.761–0.776

External

0.673

CI:0.671–0.675

0.752

CI:0.75–0.753

0.675

CI:0.67–0.68

0.734

CI:0.732–0.736

p-valued

<0.001

0.007

<0.001

<0.001

  1. aThe original cohorts (Supplementary Table 5) in which there were no missing values for sex, insurance type, age at diagnosis, HPCM, histological type, tumor site, AJCC stage, Breslow thickness, anatomic level, and laterality (Supplementary Table 1).
  2. bThe “core complete” cohorts were based on the original cohorts with no missing values for additional core features: negative or not indicated regional lymph node histology, no missing value for ulceration, and no missing value for mitotic rate.
  3. cThe “complete” cohorts were based on the core complete cohorts with additional constraints: no missing values for median income and all tumor characteristics. Considering the small sample size for time-to-vent prediction, twofold cross-validation was used for internal validation.
  4. dp-value: comparison between the internal validation and the external validation.
  5. eCI: 95% confidence interval.