1

Supplemental Methods

Development of the Prognostic Score

We have previously validated the CCP score, originally applied to prostate cancer,1,2as an independent prognostic marker in early stage, resectable lung adenocarcinoma.3In this study, the CCP score contributed significant prognostic information above clinical variables. These analyses also confirmed that pathological stage remains a significant, independent predictor of lung cancer mortality. Thus, a combination of pathological stage and CCP score would be the best predictor of outcome, capturing both the prognostic contribution of the current standard of care, pathological stage, as well as the added prognostic information conferred by the CCP score.

For the derivation of an algorithm combining pathological stage and CCP score, we selected untreated patients from two data sets used in the CCP score validation study.3 The Directors’ Consortium (DC) analysis of lung adenocarcinoma consisted of 179 patients, including 36 lung cancer deaths.4The data set from The University of TexasMD Anderson Cancer Center (MDACC) and European Institute of Oncology (IEO) included 316 patients with 54 lung cancer deaths.3 To adjust for the difference in expression ranges in microarray data (DC) compared to quantitative PCR data (MDACC/IEO), the microarray scores were centered by processing site and then scaled by 1.83, the ratio of the standard deviation of the CCP score in MDACC/IEO to the standard deviation of the CCP score, after centering by processing site, in the DC data.

Cox proportional hazards models were then constructed with the CCP score as continuous variable and stage as a numerical covariate (1=IA, 2=IB, 3=IIA, 4=IIB), using five year disease specific survival as outcome. As shown in Supplemental Table 3,the hazard ratios for pathological stage and CCP score were consistent between the data sets. For modeling, we combined untreated patients from DC and MDACC/IEO for a final prognostic score training cohort of 495 patients with 90 events. The coefficients for stage and the CCP score were derived from the hazard ratios in a Cox PH model of stage and CCP score, stratified by cohort. Stratification had six levels, two for the MDACC/IEO cohort and four representing the four contributing sources in the Directors’ Consortium. The combined score of pathological stage and CCP was then scaled and shifted to yield the final algorithm for the prognostic score:

Prognostic Score = 20 x (0.33 x CCP + 0.52 x Stage) + 15

where the CCP score is rounded to the nearest decimal and the prognostic score is rounded to the nearest integer.

Combining Cohorts

To assess the appropriateness of combining the Brigham and Women’s Hospital (BWH) and the Royal Infirmary of Edinburgh cohorts, we tested whether clinical differences between cohorts were relevant to five year disease-related death. To this end, we constructed Cox proportional hazards models, for each clinical variable, consisting of the clinical variable in question, a variable designating cohort, and an interaction term. None of the interaction terms were significant at the 5% level in two-sided likelihood ratio tests after adjusting for multiple testing.

Tests for Heterogeneity in the CCP Score Hazard Ratio

To test for heterogeneity in the CCP score hazard ratio, we constructed Cox proportional hazards models for each clinical variable consisting of the clinical variable in question, CCP score, and an interaction term. None of the interaction terms reached significance at the 5% level after adjusting for multiple comparisons. Prior to adjusting for multiple comparisons, the interaction between CCP and age was significant at the 5% level (P=0.018).

Tests for Non-Linear Effects

To test for non-linear effects of the CCP score and the prognostic score, Cox proportionalhazards models with second- and third-order terms of each score were evaluated. Wefound no evidence supporting non-linear effects, for either score, at the 5% level.

Supplemental Table 1: Distribution of predominant adenocarcinoma subtypes observed among patients in the BWH cohort. Data were available for 268 of 474 patients.

Predominant Subtype / N / (%)
Solid / 21 / 4.4
Acinar / 112 / 23.6
Papillary / 34 / 7.2
Bronchioalveolar / 101 / 21.3
Not Specified / 206 / 43.5

Supplemental Table 2: Cell cycle progression (CCP) and housekeeper genes used to calculate the CCP score.

CCP Genes / Housekeeper Genes
ASPM / RPL38
CDCA8 / UBA52
MCM10 / RPL4
FOXM1 / RPS29
CDC20 / SLC25A3
CDKN3 / CLTC
BIRC5 / RPL37
DLGAP5 / PSMA1
KIF20A / RPL8
BUB1B / PPP2CA
PRC1 / TXNL1
TK1 / MMADHC
CEP55 / PSMC1
PBK / RPL13A
RAD54L / MRFAP1
NUSAP1
RRM2
KIAA0101
ORC6L
RAD51
CENPM
SKA1
CENPF
KIF11
PTTG1
CDC2
DTL
PLK1
CDCA3
ASF1B
TOP2A

1

Supplemental Table 3. Hazard ratios for the CCP score are given per unit of CCP score. HR for stage represents the increase in hazard associated with any incremental increase in stage.

Cohorts
(Events/N) / Cox PH Model / Stage HR
(95% CI) / CCP HR
(95% CI)
MDACC/IEO
(54/316) / pStage + CCP + strata (cohort) / 1.66
(1.17-2.35) / 1.36
(1.06-1.75)
DC
(36/179) / pStage + CCP + strata (cohort) / 1.71
(1.24-2.35) / 1.44
(1.06-1.96)
MDACC/IEO/DC
(90/494) / pStage + CCP + strata (cohort) / 1.69
(1.33-2.13) / 1.39
(1.15-1.69)

MDACC=The University of Texas MD Anderson Cancer Center

IEO= European Institute of Oncology

DC=Director’s Consortium

1

Supplemental Table 4. Univariate and multivariate Cox proportional hazards analysis of CCP score and other parameters as predictors of overall survival within 5 years of surgery.

Univariate / Multivariate
HR
(95%) CI / P-value / HR
(95%) CI / P-value
CCPa / 1.62
(1.34-1.97) / 9.8 X 10-7 / 1.37
(1.10-1.70) / 0.0051
Age / 1.03
(1.02-1.04) / 2.2 X 10-5 / 1.03
(1.02-1.05) / 2.5 X 10-5
Gender / 0.027 / 0.29
Male / 1 / 1
Female / 0.74
(0.57-0.97) / 0.86
(0.65-1.14)
Stage / 1.0 X 10-6 / 0.021
IA / 1 / 1
IB / 1.32
(0.96-1.81) / 1.34
(0.85-2.11)
IIA / 2.64
(1.83-3.75) / 2.36
(1.37-4.04)
IIB / 2.41
(1.41-3.90) / 2.29
(0.97-5.16)
Tumor Size / 1.16
(1.08-1.23) / 4.3 X 10-5 / 1.03
(0.92-1.15) / 0.62
Pleural Invasion / 0.21 / 0.48
No / 1 / 1
Yes / 1.21
(0.90-1.60) / 0.87
(0.59-1.28)
Cohort / 0.0036 / 0.63
BWH / 1 / 1
RIE / 1.52
(1.15-1.99) / 0.92
(0.64-1.30)

a) HR for CCP is per IQR of the CCP score.

b) Events/N=211/633 for multivariate analysis, and univariate analysis of pleural invasion. Other univariate analyses had Events/N=222/650.

Supplemental References

1.Cuzick J, Swanson GP, Fisher G, et al: Prognostic value of an RNA expression signature derived from cell cycle proliferation genes in patients with prostate cancer: a retrospective study. Lancet Oncol 12:245-255, 2011

2.Cuzick J, Berney DM, Fisher G, et al: Prognostic value of a cell cycle progression signature for prostate cancer death in a conservatively managed needle biopsy cohort. Br J Cancer 106:1095-1099, 2012

3.Wistuba, II, Behrens C, Lombardi F, et al: Validation of a proliferation-based expression signature as prognostic marker in early stage lung adenocarcinoma. Clin Cancer Res 19:6261-6271, 2013

4.Shedden K, Taylor JM, Enkemann SA, et al: Gene expression-based survival prediction in lung adenocarcinoma: a multi-site, blinded validation study. Nat Med 14:822-827, 2008

1

Supplemental Figure 1. Risk estimates based on the prognostic score versus risk estimates based on stage. The solid vertical line indicates the cutpoint between low and high mortality risk.