1. ACCENT-Based Nomograms to Predict Recurrence and Overall Survival in Stage III Colon Cancer
Lindsay A. Renfro1, Axel Grothey2, Leonard B. Saltz3, Thierry André4, Roberto Labianca5, Steven R. Alberts2, Charles L. Loprinzi2, Greg Yothers6, Daniel J. Sargent1
for the Adjuvant Colon Cancer Endpoints (ACCENT) Group
1Division of Biomedical Statistics and Informatics, Mayo Clinic, Rochester, MN; 2Department of Oncology, Mayo Clinic, Rochester, MN; 3Gastrointestinal Oncology Service, Memorial Sloan-Kettering Cancer Center, New York, NY;
4Hôpital Saint Antoine, Paris, France; 5Oncology Unit, Ospedali Riuniti di Bergamo, Bergamo, Italy; 6National Surgical Adjuvant Breast and Bowel (NSABP) Project Biostatistical Center, Pittsburgh, PA
Abstract
Data
Trial Characteristics
Survival Nomogram
Comparison with AJCC Staging
Background: Current prognostic tools and staging systems in colon cancer (CC) use relatively few patient attributes; including additional covariates may improve prediction. Using the large ACCENT database, we constructed clinically based nomograms for overall survival (OS) and time to recurrence (TTR) in stage III CC that better separate patients into risk groups compared to AJCC v7 staging. 
Methods: 15,995 stage III patients accrued to phase III clinical trials since 1989 were used to construct Cox models for TTR and OS, stratified by study.  Variables included age, sex, race, BMI, performance status, tumor grade, tumor stage, ratio of positive lymph nodes to nodes examined, number/location of primary tumors, and treatment class. Missing data (<18%) were imputed, continuous variables modeled with splines, and clinically relevant pairwise interactions included if p <  Final models were internally validated via bootstrapping for corrected calibration and C-indices for survival data.  Nomogram-defined risk tertiles were compared to AJCC v7 stage III for observed 3-year TTR and 5-year OS for a subset of 7400 patients with complete data. 
Results: All variables were statistically and clinically significant for OS; age and race did not predict TTR.  No meaningful interactions existed.  Nomograms for OS and TTR were well calibrated and associated with C-indices of 0.66 and 0.65, respectively, vs and 0.59 for AJCC.  Nomogram risk tertiles were better separated than AJCC groups.  Removing treatment from the nomograms did not hurt performance (C=0.66 for OS and 0.65 for TTR). 
Conclusions: The proposed ACCENT nomgrams are internally valid and have the potential to aid prognostication, patient/physician communication, and decision making in patients with stage III colon cancer.
15, 936 individual patients (with > 50% variables available) from 12 randomized phase III trials conducted since 1989
Endpoints:
Overall survival (OS), defined as time from randomization to death from any cause
Time to recurrence (TTR), defined as time from randomization to recurrence of primary disease
Candidate Variables:
Age (years)
Sex (Male, Female)
Race (White, Black, Asian, Other)
Body Mass Index (BMI; continuous)
Performance Status (PS; 0, 1, 2+)
Tumor Grade (1, 2, 3+)
Tumor Stage (T-stage; T1, T2, T3, T4)
Ratio of # positive lymph nodes to # examined (continuous)
Number/location tumors: single left, right, or transverse; any multiple
Treatment (5FU Variations, 5FU+irinotecan, 5FU+oxaliplatin)
Trial
Years Accrued
Treatment Arm(s) N
NCCTG
FU/LV +/- LEV for 6 or 12 months
754
INT 0089
FU/LEV vs. FU/LV (HD/LD) vs. FU/LV/LEV
2,704
NSABP C05
FU/LV vs. FU/LV + IFN
1,191
NCCTG
FU/LV + HD or standard LEV
648
SWOG 9415
Bolus vs. infusional FU/LEV/LV
795
GERCOR
Bolus vs. infusional FU/LV
512
MOSAIC
FU/LV vs. FOLFOX
1,344
X-ACT
FU/LV vs. Cap
1,982
NSABP C06
IFU/LV vs. UFT+LV
829
NSABP C07
1,733
CALGB 89803
FU/LV vs. FU/LV + Iri
1,239
PETACC-3
FU/LV (AIO or LVFU2) vs. +/- Iri
2,264
Total Original ACCENT
15,995
AJCC v7 Staging
Nomogram OS
C-index = 0.58
C-index = 0.66
TTR
C-index = 0.59
C-index = 0.65
Age
Sex
Race
BMI PS
Grade
T-Stage
Nratio
Location
Treat
Total
3-Yr OS
5-Yr OS
Patient A 58 M
White 35 2 T3
0.20
Left
Oxali
Points 3 11 5 37
119
86%
78%
Patient B 82 45 1 3+
0.65
Right
5FU 12 15 8 80 14
251
32%
15%
Recurrence Nomogram
Final Models OS
TTR
Variable
Coef
Std. Err HR
Factor
P-value
Age*
0.0149
0.0025
0.0028
<
Sex
Female
Male -
0.1740
0.0271
1.190
0.0866
0.0265
1.090
0.0011
Race
White
Black
Asian
Other
0.1869
0.1220
0.0361
0.0580
0.1080
0.1038
1.205
1.130
1.037
0.0094
BMI*
0.0370
0.0069
0.0077
0.0418
0.0066
0.0074 PS 1 2+
0.1330
0.3514
0.0318
0.1021
1.142
1.421
0.1110
0.2850
0.0315
0.1129
1.117
1.330
0.0002
Grade 2 3+
0.0725
0.2262
0.0468
0.0519
1.075
1.254
0.1069
0.2008
0.0451
0.0507
1.113
1.222
T-Stage T1 T2 T3 T4
0.3748
0.9086
1.3469
0.0951
0.0945
0.0993
1.455
2.481
3.845
0.5317
1.1514
1.5682
0.1044
0.1030
0.1072
1.702
3.163
4.798
Node Ratio*
2.9665
0.1870
0.3306
2.8326
0.1822
0.3235
Location
Left
Right
Trans
Multi
0.2297
0.1907
0.2380
0.0306
0.0377
0.0813
1.258
1.210
1.269
0.0796
0.0578
0.1627
0.0298
0.0371
0.0788
1.083
1.059
1.177
0.0178
Treatment
5FU
5FU+Ox
5FU+Iri
0.0661
0.0608
0.842
0.944
0.0219
0.0626
0.0554
0.751
0.933
<0.0001
Methods
Continuous Predictor Effects
Multivariable Cox proportional hazards models were created for TTR and OS, stratified by study
All statistically and clinically significant variables and pairwise interactions were included in the final models
Continuous variables were modeled with nested splines
Models were internally validated via bootstrapping to obtain optimism-corrected discrimination (C-index) and calibration plots
7,400 patients with complete data were used to compare nomogram-defined risk tertiles versus AJCC v7 (A,B,C)
Age
Sex
Race
BMI PS
Grade
TStage
Nratio
Location
Treat
Total
3-Yr RF
5-Yr RF
Patient A 58 M
White 35 2 T3
0.20
Left
Oxali
Points - 6 5 7 73
126
78%
73%
Patient B 82 45 1 3+
0.65
Right
5FU 15 13 77 18
214
38%
30%
Calibration
Conclusions
Objectives
Results
The proposed stage III colon cancer nomograms:
Involve more patient characteristics than existing tools
Are internally valid (good discrimination and calibration)
In the future, these nomograms may be used for:
Stratified randomization in clinical trials to balance risk across treatment arms
Treatment decision-making
Enhanced physician/patient communication
To create internally valid nomograms for predicting overall survival (OS) and time to recurrence (TTR) in stage III colon cancer using widely available patient characteristics
To compare outcomes by nomogram-defined risk tertiles against AJCC v7 staging (IIIA, IIIB, IIIIC) among patients where both classifications are available
All variables were included in the final OS model
All variables except age and race included in TTR model
No relevant two-way interactions existed for OS or TTR
Both models are internally valid with good calibration and better discrimination than AJCC
The OS and TTR models showed improved discrimination (correct ordering of pairs of patients by risk) relative to AJCC v7 staging