1. Lymph Node (LN) Ratio (LNR) Based Risk Classification (RC) in Stage III Colon Cancer (CC) with Internal and External Validation: Finding from the ACCENT Database Q. Shi 1, J.M. Hubbard 1, G. Yothers 2, T. Andre 3, L. Saltz 4, G. Francini 5, B. M. Bot 1, C. Twelves 6, M. Buyse 7, A. Grothey 1, and D. J. Sargent 1, The Adjuvant Colon Cancer Endpoints (ACCENT) Collaborative Group Purpose: The degree of LN involvement is one of the most important prognostic factors in CC. Current staging does not consider the number of total LN examined (TLN) and its predictive ability of nodal associated disease RC. Our aim was to develop a RC model based on LNR and other key prognostic factors (e.g. T stage) with internal and external validation. Methods: Individual patient (pt) data from 17 randomized clinical trials were used. LNR is defined as number of positive LNs divided by number of TLN. The primary endpoint was disease-free survival (DFS). The development dataset consisted of pts receiving 5FU based chemotherapy only. This dataset was randomly split into a training set (6822 pts) and an internal validation set (5047 pts). Pts receiving surgery alone (1001 pts) and 5FU plus other agents (oxaliplatin, irinotecan, or oral-based) (3688 pt) comprised two external validation populations. Optimal cutoffs were selected based on Cox model goodness of fit. Validation was based on two key aspects of a predication model - discrimination and calibration. In addition, the Area under the Curve (AUC) was used to assess the accuracy of 3 and 5 year prediction. All analyses were stratified by study. Results: Overall 54% of pts were male with median age of 61. Median follow-up was 8 yrs. 7320 pts had recurrence or died, with 5 yr DFS of 58.6% (95% CI: 57.8%, 59.4%). The optimal RC based on LNR was identified with the same discrimination (c statistics = 0.62) but improved prediction at 3 and 5 years (AUC = 0.65) over AJCC N staging. The hazard ratios comparing higher risk subgroups to reference group for LNR based RCs range from 1.5 to 5.4 (all p <.0001). Conclusions: LNR, especially combined with T stage can aid risk classification of DFS and provides better prognostic prediction of DFS in stage III CC than AJCC N and T staging. ABSTRACT To assess the prognostic values of LNR in stage III colon cancer To establish optimal risk group classifications based on LNR alone or combination of LNR and T stage OBJECTIVES Patients: Stage III CC patients from 18 trials with available LN related measurements (LN+ and TLN) Treatments including: surgery only, 5FU-based, 5FU plus other (oxaliplatin, irinotecan, or oral-based) chemotherapy Outcomes: Disease-Free Survival (DFS): defined as the time from the first occurrence of recurrence or death due to all cause Statistical analyses Stratified Cox PH model was used to assessing association between LN RCs and DFS Optimal cutoffs searching was based on lowest AIC of the model fitting on training set Internal validation – optimism-corrected performance External validation – predicted performance Comparisons between candidate LNR RCs and AJCC RCs were based on the model performances measures below from internal and external validations: C statistics – the ability of discriminating high vs. low risk patients (range from 0 to 1, close to 1 the better) Calibration slop – the reliability of the predicted effect based on the risk classification (range from - ∞ to + ∞, close to 1 the better) AUC of 3 & 5 year rate prediction – the accuracy of prediction based on the risk classification at 3 and 5 year for individual patients (range from 0 to 1, close to 1 the better) METHODS Statistical Analysis Flow Chart Characteristics Training set Internal Validation set External validation sets P 5FU or variation of 5FUsurgery only * 5FU + other No. patients (%) 6822 (41.2)5047 (30.5)1001 (6.0)3688 (22.3)NA Age, years Median Range 62.0 15.0 to 91.0 60.0 15.0 to 85.0 61.0 18.0 to 84.0 60.0 17.0 to 84.0 <.0001 † Gender, n (%) Female Male 3135 (49.9) 3687 (54.1) 2344 (46.5) 2702 (53.5) 469 (46.9) 532 (53.1) 1680 (45.5) 2008 (54.5) 0.8066 # TLN Median Range 12.0 1.0 to 99.0 12.0 1.0 to 99.0 11.0 1.0 to 77.0 13.0 1.0 to 97.0 <.0001 ‡ Number of positive LN Median Range 2.0 1.0 to 44.0 2.0 1.0 to 37.0 2.0 1.0 to 32.0 2.0 1.0 to 32.0 0.3766 ‡ LNR Median Range 0.25 0.01 to 1.00 0.24 0.01 to 1.00 0.25 0.01 to 1.00 0.20 0.01 to 1.00 <.000 ‡ T stage, n (%) T1/2/is T3 T4/Perforation 821 (12.2) 5034 (74.6) 892 (13.2) 651 (14.8) 3556 (80.9) 190 (4.3) 82 (10.1) 696 (86.0) 31 (3.8) 418 (11.3) 2828 (76.8) 438 (11.9) <.0001 # Patient Characteristics Abbreviations: TLN: total number of lymph node examined; LN: lymph nodes; * Regimens include FLOFOX4, FOLOX, IFL, FOLFIRI and Uracil/tegafur; † ANOVA; ‡ Kruskal-Wallis test; # Chi-square test; AJCC N & T StageLNR & T Stage Nodal Risk Classification Permanence Measures Internal validation External validation 1 External validation 2 With T Stage AJCC N and T stage N1, T1/2 N1, T3/4 N2a, T1 N2a, T2/3 N2a, T4 N2b, T1/2 N2b, T3/4 C statistics Calibration slop 3 year rate pred. 5 year rate pred. 0.62 0.98 0.51 0.27 0.60 0.89 0.52 0.60 0.87 0.44 0.45 LNR and T stage T1/2/is lnr ≤ 0.4 lnr > 0.4 T3 lnr ≤ 0.17 0.17 < lnr ≤ 0.36 0.36 < lnr ≤ 0.67 lnr > 0.67 T4 lnr ≤ 0.3 lnr > 0.3 C statistics Calibration slop 3 year rate pred. 5 year rate pred. 0.63 0.98 0.69 0.68 0.61 0.87 0.62 0.94 0.65 Comparisons of Performance among RC Models LNR is a significant prognostic factor of DFS in stage III colon cancer patients. Strong non-linear relationship between LNR and DFS p <.0001 for quadratic effect HR = 1.49, 95% CI = [1.43, 1.55] for LNR = 0.3 vs. 0.1 Significant non-linear relationship remains after adjusting for age, gender, T stage, PS, grade and treatment. Based on LN data only, LNR RC achieves similar performance as AJCC N staging. Comparing to AJCC N & T RC, the LNR & TRC achieves same overall performance in discriminating high vs. low risk patients, and predicting DFS (C stat = 0.61 to 0.63, Calibration slop = 0.87 to 0.98) predicts 3 and 5 year DFS rates more accurately (AUC = 0.62 to 0.69) Two RCs agree on 85% of the patients. Among the patients whose risk differs between classifications, the discrepancy within ‘high risk’ patients is greater than low or intermediate risk group. KEY FINDINGS This is a large scale risk classification analysis with internal and external validations. LNR based RC discriminate patient regarding long-term adverse outcome risk well. The RC based on LNR which take account of TLN provide more accurate prediction in DFS than the RC based on number of LN+ alone. DISCUSSIONS Each of the rectangle of area is proportional to % of patients Blue, red, and green boxes represent low, intermediate, and high risk pts classified as by AJCC N and T stage, respectively. Columns of “Low”, “Intermediate”, and “High” represent 3 risk groups classified by LNR and T stage. Low, Intermediate, and High risk groups consist pts who belong to different shades of blue, red and green lines in the K-M curves on the left. Comparison of Patient Risk Classifications based on Two RCs ACCENT Initially established in 2003, to validate DFS as an endpoint in adjuvant colon cancer Individual patient data from 24 large adjuvant randomized clinical trails, 33,574 pts Jointly owned by all contributors This study was funded by NIH Grant CA-25224 Authors Affiliation 1. Mayo Clinic, USA; 2. NSABP Biostatistical Center, USA; 3. Gr Hospitalier Pitie-Salpetriere, France; 4. Memorial Sloan-Kettering Cancer Center, USA; 5. University of Siena, Italy; 6. University of Bradford, UK; 7. International Drug Development, Belgium;  Two RC classify 84.6% patients into the same risk groups.  LNR based RC classified  15.1% AJCC-low risk pts into intermediate risk group  8.6% AJCC-intermediate risk pts into high/low risk group  49.6% AJCC-high risk pts into intermediate risk group