Multivariate Modeling of Melanoma

1
Multivariate Modeling of Melanoma Prognosis and
its Web-based Applications
Seng-jaw Soong, Ph.D. Professor and Director
Emeritus Biostatistics and Bioinformatics
Unit Associate Director Emeritus of the
Comprehensive Cancer Center University of Alabama
at Birmingham
Presented at University of Kansas Medical
Center November 17, 2009
2
Melanoma Clinical Presentations
3
Melanoma Clinical Presentations
4
Melanoma Thickness, Ulceration and Clarks Level
5
Regional Nodal Metastases
6
2007 Estimated US Cancer Cases
Women678,060
Men766,860

• 26 Breast
• 15 Lung bronchus
• 11 Colon rectum
• 6 Uterine corpus
• 4 Non-Hodgkin lymphoma
• 4 Melanoma of skin
• 4 Thyroid
• 3 Ovary
• 3 Kidney
• 3 Leukemia
• 21 All Other Sites

Prostate 29 Lung bronchus 15 Colon
rectum 10 Urinary bladder 7 Non-Hodgkin 4
lymphoma Melanoma of
skin 4 Kidney 4 Leukemia 3 Oral
cavity 3 Pancreas 2 All Other Sites 19
Excludes basal and squamous cell skin cancers
and in situ carcinomas except urinary
bladder. Source American Cancer Society, 2007.
7
Incidence rates for cutaneous melanoma for males
and females, in the first 17 ranked developed
countries, as well as China and Japan,
standardized to the world age structure
Per 100,000
Compiled from Ferlay J, Bray F, Pisani P, Parkin
DM. GLOBACAN 2002 Cancer Incidence, Mortality
and Prevalance Worldwide IARC CancerBase No.5,
Version 2.0, IARCPress, Lyon 2004.
http//www-dep.iarc.fr/ accessed 3-19-2007.
8
Melanoma Research UAB and International
Collaboration Experience (1975-present)

• Studies of Natural History, Prognosis and
  Treatment of Melanoma
• Early pioneer multifactorial analyses of
  localized, regional and distant melanoma based on
  Cox regression models at UAB (1978a, 1978b,
  1979a, 1979b, 1980, 1981a, 1981b, 1983a)
• International collaboration on melanoma research
• With Sydney Melanoma Unit, University of Sydney,
  Australia (1982a, 1982b, 1983b, 1984a, 1984b,
  1984c, 1984d, 1985a, 1985b, 1988, 1991a, 1991b,
  1991c, 1992, 1997, 1998a, 1998b, 2003a, 2003b,
  2004, 2006, 2007)
• With World Health Organization (WHO) Melanoma
  Group (1984, 1997)
• With American Joint Committee on Cancer (AJCC)
  and International Union Against Cancer (UICC)
  Melanoma Task Force (2001a, 2001b, 2003, 2003a,
  2003b, 2003c, 2004, 2005, 2008a, 2008b, 2008c,
  2009a, 2009b, 2009c, 2009e)

• 2000 AJCC Database 13 institutions
• 2008 AJCC Database 15 institutions

• NCI-sponsored Multi-institutional Melanoma
  Clinical Trial (1996, 2000, 2001, 2001b)
• Development of New AJCC Melanoma Staging System
  (1997, 2000a, 2001a, 2001b, 2003a, 2003b, 2003c,
  2004, 2005, 2008a, 2008b, 2009a)
• Development of Risk Classification of Melanoma
  Based on Survival Tree (Recursive Partition Tree)
  Methods (1997, 1998, 2007)
• Development and Validation of Predictive Models
  for Melanoma Patients (1979, 1984, 1991, 1992,
  1997, 2003, 2008c, 2009b, 2009c, 2009e)

9
Multivariate Nature of Melanoma Prognosis
Pathological Factors
Clinical Factors
DISEASE RECURRENCES AND PATIENT SURVIVAL
Genetic Factors
Immunological Factors
Nutritional Factors
Social / Economic Factors
10
Multi-state Disease Transitions Model
First Second Third Fourth Event
Event Event Event Distant
Mets Death Regional Metastasis Death
Distant Metastasis Death Death
Regional Mets Distant Mets Death Death
Local Recurrence
Initial State of Melanoma
(Local Stage I/II) (Regional Stage
III) (Distant Stage IV)
11
Primary Objectives of Prognostic Factor Analysis
Identification of Dominant Prognostic
Factors Prediction of the Clinical Course of
Melanoma Patients Development of Melanoma
Staging System
12
Development of Predictive Models for Melanoma
Outcome
Two Important Clinical Questions

• What is the patients chance of surviving for a
  given period, e.g., 5 or 10 years, after
  diagnosis of melanoma?
• If the patient is disease-free for a period of
  time, e.g., 2 or 5 years, what is his or her
  chance of melanoma recurrence or death in the
  successive time intervals?

13
Cox Regression Model Hazard Function
Survival Function
14
AJCC Melanoma Databases
15
AJCC Collaborative Melanoma DatabaseMultivariate
Cox Regression Analysis for Stage I/II Melanoma
Patients (n25,734)
Reference
16
AJCC Collaborative Melanoma DatabaseMultivariate
Cox Regression Analysis for Stage I/II Melanoma
Patients (n25,734) (continued)
Reference
17
AJCC Collaborative Melanoma Database Stage I/II
Survival Curve by Tumor Thickness
Revised 1/17/08
18
(No Transcript)
19
AJCC Collaborative Melanoma Database Stage I/II
Survival Curves by Stage
Revised 1/17/08
20
Predictive Model Development and Validation For
Stage I/II Melanoma
21
AJCC Melanoma Collaborative Database Stage
I/II Datasets Used for Model Development and
Validation

• 1. Model Development (Training) Dataset
  (n14,760)
• This dataset consists of data from the
  following 10 institutions and groups
• 1) Memorial Sloan Kettering Cancer
  Center
• 2) MD Anderson Cancer Center
• 3) University of Pennsylvania
• 4) Sunbelt Melanoma Group
• 5) Sentinel Lymph node Working Group
• 6) University of Michigan
• 7) Moffitt Cancer Center
• 8) University of Alabama at Birmingham
• 9) InterGroup Melanoma Clinical Trial
• 2. Model Validation (Test) Dataset (n10,974)
• This dataset consists of patients treated
  at Sydney Melanoma Unit (SMU), Australia

22
AJCC Collaborative Melanoma Database Stages I and
II Comparisons of Patient Characteristics by
Dataset
23
AJCC Collaborative Melanoma Database Stages I and
II Comparisons of Patient Characteristics by
Dataset
24
AJCC Collaborative Melanoma Database Multivariate
Cox Regression Analysis For Stages I and II
Melanoma by Dataset
25
AJCC Collaborative Melanoma DatabaseStage
I/II Survival Curves by Tumor Thickness Model
Development Dataset
26
AJCC Collaborative Melanoma Database Stage
I/II Survival Curves by Tumor Thickness Model
Validation Database (SMU)
27
AJCC Collaborative Melanoma Database Stage
I/II Cox Regression Model by Tumor
Thickness Developed Based on the Model
Development Dataset
Model Tumor Thickness (mm) Covariates in the
Model (1) 0.00 0.50 Ulceration (Y/N), Age
(lt60/60) (2) 0.51 1.00 Ulceration (Y/N),
Lesion Site (A/E) Age (lt60/ 60) (3) 1.01
2.00 Ulceration (Y/N), Lesion Site
(A/E) Age (lt70/70) (4) 2.01
3.00 Ulceration (Y/N), Lesion Site (A/E) Age
(lt70/70) (5) 3.01 6.00 Ulceration (Y/N),
Lesion Site (A/E) Age (lt70/70) (6)
gt 6.00 Ulceration (Y/N)
28
AJCC, Stage I/II, Validation
Correlation Plot Observed Survival Rate and
Predicted Survival Rate by Combination of
Prognostic Factors within Thickness Groups
Predicted survival rates based on the models
developed from the model development
dataset. Observed survival rates calculated
using the model validation dataset.
29
AJCC Collaborative Melanoma Database Predicted 5-
and 10-Year Survival Rates by the Cox
Models Developed Based on the Model
Development Dataset (n14,760)
Tumor __Survival Rate (95
C.I.)__ Patient Thickness Ulceration Lesion
Site Age 5-Year 10-Year____
1 0.8 No Extremity 54 0.98 0.96 (
0.97-0.99) (0.94-0.97) 2 1.5 Yes Extremity
45 0.89 0.80 (0.87-0.92) (0.75-0.84)
3 2.2 No Axial 75 0.72 0.60 (0.65
-0.79) (0.51-0.69) 4 3.7 Yes Axial
39 0.63 0.50 (0.58-0.68) (0.44-0.56)
5 7.0 Yes Extremity 60 0.47 0.30
(0.40-0.55) (0.22-0.40)
30
Web-based Melanoma Outcome Prediction
Tools (www.melanomaprognosis.org)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
Melanoma outcome prediction tools that have been
or will be included in the website

• Predicting outcome for patients with localized
  (Stage I/II) melanoma.
• Predicting outcome for patients with regional
  (Stage III) melanoma.
• Predicting outcome for patients with distant
  (Stage IV) melanoma.
• Predicting outcome following a disease-free
  interval for patients with localized melanoma.
• Predicting sentinel lymph node positivity based
  on a stratified analysis of logistic model.
• Generating clinically useful hazard function for
  individual patients based on the generalized
  gamma (GG) parametric model.

36
Practical Applications
Disease Staging and Clinical Scoring
System Treatment Planning and Follow-up
Evaluation Clinical Trial Planning - Identifi
cation of Adequate Study Population - Stratifica
tion Criteria Clinical Trial
Analysis - Adjustments in Treatment
Comparisons - Refinements in Subgroup Analyses
37
Clinical score is defined as a patients
projected 10-year survival rate, and can be
considered as a composite prognostic indicator of
several dominant prognostic factors in the
predictive model.
38
(No Transcript)
39
(Example) Clinical Trial Design
40
(Example) Stratified Analysis of Clinical Trial
Treatment A
Treatment B
Treatment A
Patientsin thestudy
Treatment B
Treatment A
Score0 - 19
Treatment B
41
Predicting Disease Recurrence and Patient
Survival After Disease-Free Intervals
42
(Soong, et. al., World J. of Surgery, 1992)
43
Predicting Outcome of an Individual Patient With
Localized Melanoma
Example Patient 'X' Characteristics
Age - 48 years Tumor Thickness - 1.2 mm Sex -
Female Clark's Level - III Lesion Location
- Leg Ulceration - Yes Surgical Treatment
- WLE only Growth Pattern Superficial
spreading Key characteristics that predict the
clinical outcome of Patient X's localized
melanoma.
44
Estimated Probabilities of Melanoma Recurrence
and Death after Disease-Free Interval (Example
for Patient 'X')
21
Melanoma Recurrence
7
Melanoma Death
25
2 years
0
16
5 years
2 years
0
4
2
2 years
10 years
5 years
0
4
1
2 years
5 years
0
10 years
15 years
45
Parametric Modeling of Melanoma Prognosisand
Outcome

• An Alternative to Cox Regression Model
• Shouluan Ding, Seng-jaw Soong, Hui-Yi Lin,
• Renee Desmond, Charles Balch
• Biostatistics and Bioinformatics Unit,
  Comprehensive Cancer Center,
• The University of Alabama at Birmingham
• and Johns Hopkins University
• Correspondence to Seng-jaw Soong
• The University of Alabama at Birmingham,
• 1530 3rd Avenue South, WTI 153,
• Birmingham, AL 35294-3300
• Email sjsoong_at_uab.edu

Journal of Biopharmaceutical Statistics,
19732-747, 2009
46
Generalized Gamma (GG) Distribution



47
Comparing the test statistics for the
parameters GG Model vs. Cox Model

Cox Model
Generalized Gamma Model
p-value
p-value
Wald
(SE)


IA
Tumor thickness (mm)


Clark level (IV/V vs. II/III)
-
Gender (female vs. male)
Ulceration (yes vs. no




Site (axial vs. extremity


Age (?60 vs. 60)


IB
-
48
Comparing the test statistics for the
parameters GG Model vs. Cox Model (cont)




Cox Model
Generalized Gamma Model






49
Generalized Gamma Regression Model
Hazard function of GG model for localized
melanoma by substage at diagnosis
50
Hazard Function Plots for Individual Patients
Patients Thickness Clark Gender
Ulceration Site Age Maximum
Time (mm) Level

Hazard (year)
III
51
Acknowledgments

• AJCC Melanoma Task Force (35 members) under the
  leadership of
• Charles Balch, MD, Jeff Gershenwald, MD,
  Seng-jaw Soong, PhD
• Institutions/study groups that contributed data
  to the AJCC Melanoma Database
• Memorial Sloan-Kettering Cancer Center
• MD Anderson Cancer Center
• Sydney Melanoma Unit - Australia
• University of Alabama at Birmingham
• Sunbelt Melanoma Trial Group
• National Cancer Institute Milan, Italy
• Moffitt Cancer Center
• Intergroup Melanoma Surgical Trial Group
• Eastern Cooperative Oncology Group
• San Pio X Hospital Milan, Italy
• Sentinel Lymph Node Working Group
• Netherland Cancer Institute, Netherland
• National Cancer Institute Naples, Italy
• University of Pennsylvania
• University of Michigan