- 著者: Kratz JR, Haro GJ, Cook NR, He J, Van Den Eeden SK, Woodard GA, Gubens MA, Jahan TM, Jones KD, Kim IJ, He B, Jablons DM, Mann MJ
- Corresponding author: Mann MJ (Division of Adult Cardiothoracic Surgery, University of California, San Francisco, CA)
- 雑誌: Journal of Thoracic Oncology
- 発行年: 2019
- Epub日: 2019-04-05
- Article種別: Original Article (clinical methodology study、prognostic biomarker validation)
- PMID: 30959120
背景
Non-small cell lung cancer (NSCLC) staging system は1953年 Denoix の TNM descriptors (tumor size、nodal status、metastasis) 提唱以来100年以上続いており、AJCC TNM 8th edition (2018年導入) も TNM descriptors のみに基づく clinicopathologic staging を維持している (Goldstraw et al. Chest 2017; Detterbeck et al. JThoracOncol 2017)。8th edition の改訂は tumor size cutoff point 増加・stage subdivision 細分化・metastasis型区別を含むが、early-stage NSCLC で 5-year overall survival (OS) が 55-60% に留まり (Tomita et al. JThoracOncol 2018)、external validation は disappointing。Adjuvant chemotherapy adoption (LancetOncol 2008) でも、本来 high-riskの early-stage 患者を effectively identify できないことが mortality 残存の主因。Molecular profilingは prognostic + predictive biomarkers として有望で、staging に molecular descriptors を統合する次世代 cancer staging が期待されているが、TNMと molecular classifierの統合 staging system を rigorous reclassification metrics で validation した報告は限定的である。これまで未解明だった重要な課題として足りなかったのは、(1) 14-gene molecular prognostic classifier (BAG1、BRCA1、CDC6、CDK2AP1、ERBB3、FUT3、IL11、LCK、RND3、SH3BGR、WNT3A + reference genes ESD、TBP、YAP1) を TNM 8th edition に統合した novel TNMB staging system (B=biology) の開発、(2) この TNMB system が conventional 7th/8th editions vs TNM 単独より NSCLC outcome prediction を改善するかの大規模 multicenter validation、(3) Net reclassification improvement (NRI)、integrated discrimination improvement (IDI)、reclassification calibration (RC) などの modern reclassification metrics による定量比較、の3点であり、本研究は n=1,694 NSCLC patients を用いて初めて体系的に実証する。
目的
非扁平上皮 NSCLC patientsの surgical resection後 5-year OS prediction に対し、14-gene molecular prognostic classifierを TNM 8th edition stagingに統合したnovel TNMB staging system (B = biology) を開発・検証し、TNM 7th/8th editions vs TNMB の reclassification metrics (NRI、relative IDI、reclassification calibration RC statistic、C-index、Royston modified Nagelkerke R^2) で前向き比較すること。
結果
Patient cohort characteristics (Table 1)：n=1,694 total cohort、UCSF development cohort n=321 patients (mean age 67.5±10.6 years、59.5% female、67.0% smokers、5-year mortality 42.5%)。Kaiser validation cohort n=418 patients (mean age 66.5±9.3、45.7% female、84.7% smokers、43.9% 5-year mortality)、CCTC cohort n=955 patients (mean age 58.3±10.7、62.5% female、51.2% smokers、47.1% 5-year mortality)。Adenocarcinoma dominant (UCSF 82.2%、Kaiser 77.5%、CCTC 91.0%)。TNM 8th edition stage breakdown spans IA1 to IIIC (Table 1)。
Best supervised reclassification system selection (Table 2)：Testing group (n=81) で 3 candidate staging systems を評価。-1/0/+1 (TNMB) system が最良: Royston R^2 = 0.19 vs TNM 8th edition R^2 = 0.14 (約 36% improvement)、C-index = 0.64 vs 8th edition 0.61 (0.03 absolute improvement)、NRI = 0.51、IDI = 0.05、rIDI = 34.5%。-2/0/+1 system は NRI 0.51、rIDI 29.7%、-2/+1/+2 system は NRI 0.44、rIDI 29.3% で performance劣後。TNMB (-1/0/+1) を validation cohort 用に選択 (Table 3)。
Validation cohort応用 — 78% reclassified：n=1,373 Kaiser/CCTC validation cohort 適用結果、TNMB stagingで 78% of patients が reclassify された (約 1,071 patients、cohort n=1,373 patients)。Predominant effect = upward classification (more advanced disease stages reclassification)、これは early-stage NSCLCの poor survival reality と consistent。
TNMB vs TNM 8th edition improvement (Tables 4-5、Figure 1)：Net Reclassification Improvement (NRI): TNMB vs TNM 8th edition NRI = 0.33 (95% CI 0.24-0.41、p<0.001) vs TNM 7th vs 6th editions NRI = 0.01 (95% CI -0.04 to 0.03、no improvement) + TNM 8th vs 7th editions NRI = 0.03 (95% CI 0.00-0.06、minimal)。TNMB improvementは conventional TNM revisions の約 11-33-fold larger NRI。relative Integrated Discrimination Improvement (rIDI): TNMB vs 8th edition rIDI = 22.1% (95% CI 8.8-35.3%、p<0.001) vs TNM 7th vs 6th rIDI = 2.1% (95% CI -5.8 to 9.9、ns)、TNM 8th vs 7th rIDI = -2.5% (95% CI -17.6 to 12.4、ns、actually worsened)。TNMB は約 10-fold larger rIDI improvement。Reclassification Calibration (RC) statistic: TNMB = 21 vs TNM 8th edition = 39 (約 46% reduction、cohort n=1,373 patients、lower is better)、TNM 7th = 23 → 8th = 25 (worsened by 2 points、conventional staging revision がcalibrationを improve できない)。Modified Nagelkerke R^2: TNMB 0.17 vs TNM 7th/8th 0.14 (約 21% relative improvement)。Chi-square statistic of log-rank test (trend by stage): TNMB 194 vs TNM 8th edition 155 vs TNM 7th 159 (TNMB 約 25% higher chi-square、stronger survival discrimination)。C-index: TNMB 0.66 vs TNM 7th/8th 0.64 (0.02 absolute improvement、cohort n=1,373 patients)。
Survival curve separation analysis (Figure 1)：Kaplan-Meier overall survival curves for TNMB stages I-IIIC が conventional TNM (7th and 8th editions) より大きな curve separation を示し (Figure 1)、early-stage vs late-stage 間の survival range が拡大。Visual separation の improvement は statistical reclassification metrics (NRI、rIDI、RC) の improvementと consistent。
考察/結論
① 先行研究との違い: 既存報告のNSCLC staging revisions (TNM 5th → 6th → 7th → 8th edition、Goldstraw et al. Chest 2017) は clinicopathologic descriptors (tumor size cutoffs、stage subdivisions、metastasis type) の refinementに focus してきたのと対照的に、本研究はmolecular descriptors (14-gene expression) を staging systemに体系的に統合した点で先行研究と決定的に異なる。既存報告で TNM 7th vs 6th editions、8th vs 7th editions の NRI = 0.01-0.03 vs TNMB の NRI = 0.33 という約 11-33-fold larger reclassification improvement は、 conventional staging revision approach の本質的 limitationを定量的に示した。さらに既存 molecular prognostic classifier studies は単独 prognostic biomarker としての validation に焦点があった (Kratz et al. Lancet 2012) のと相違し、本研究は TNM staging との統合 framework での validation を初めて達成した。② 新規性: 本研究で初めて、14-gene molecular prognostic classifierとTNM 8th edition staging を統合した novel TNMB staging system (B = biology) を開発し、n=1,373 independent multicenter cohort で NRI=0.33、rIDI=22.1%、RC 39→21 という substantial reclassification improvement を実証した novel evidenceを提供した。これまで報告されていない知見として、(a) Supervised reclassification methodology (-1/0/+1 framework) の establishment、(b) 78% reclassification rate in validation cohort、(c) molecular classifier risk → stage shift mapping table (Table 3) の completion、(d) TNM 7th → 8th edition 単独では reclassification metrics が改善しない (実際 calibration が悪化) という critical findings、を実証した。novel paradigm shiftとして “next-gen cancer staging includes molecular biology” を operationally実装。③ 臨床応用と意義: 本研究の bench-to-bedside translational milestone として、(a) Early-stage NSCLC adjuvant chemotherapy 適応判断の refining (TNMB high-risk patients への aggressive adjuvant therapy)、(b) Stage IA-IB の under-treated 患者 identification (現在の guideline では adjuvant therapy 非推奨だが、 TNMB up-staging で adjuvant treatment 検討対象に)、(c) Stage III の over-treated 患者 identification (TNMB low-risk down-stagingで treatment de-intensification 検討)、(d) Cancer staging system development の future template (他癌種への molecular-integrated TNMB platform 拡張可能性)、 (e) Razor Genomics への CLIA-certified assay 商業化 (already 実装済み) で実臨床 application 加速、の臨床応用の意義が大きい。臨床的有用性として 5-year survival prediction の significant improvement が adjuvant treatment decision-makingに直接 informす translational pathway を確立。④ 残された課題と今後の検討: limitationとして、(1) Retrospective design — prospective trial で TNMB-guided adjuvant therapy が actual survival outcome を改善するか validation が今後の研究として必要、(2) Only non-squamous NSCLC patients (squamous NSCLC は molecular biology が異なるため除外) — squamous-NSCLC 専用 classifier の development が future studyとして残された課題、(3) Adjuvant therapy era での generalizability validation — immunotherapy (PD-1/PD-L1 inhibitors) 普及前 cohort であり、modern era での re-validation が必要、(4) Cost-effectiveness analysis — qRT-PCR assay コスト vs benefit、(5) Other molecular features (tumor mutation burden、PD-L1 expression、ctDNA、neoantigen load) との combinatorial improvement の解明、(6) Stage IV / metastatic disease での similar TNMB platform development、(7) Adjuvant pembrolizumab / atezolizumab era での TNMB stratification の clinical utility、(8) Regulatory pathway development (FDA Breakthrough Designation、CMS reimbursement) for routine clinical adoption、が今後の展望。これらlimitationにもかかわらず、本研究は NSCLC staging system に molecular biology を 体系的に統合した clinical methodology landmark evidence であり、 next-generation cancer staging development の clinical-methodology milestoneである。
方法
Study design: Multi-cohort retrospective biomarker-staging validation study。Patient cohorts: 計 n=1,694 patients of resected non-squamous NSCLC (stage I-IIIC) — Development cohort: University of California San Francisco (UCSF) n=321 patients (resected 1997-2007)、training group n=240 + testing group n=81 (3:1 split)。Validation cohort: Kaiser Permanente Northern California (Kaiser) n=418 patients (1998-2005) + China Clinical Trials Consortium (CCTC、Guangzhou等 3 leading academic institutions) n=955 patients (2000-2008)、合計 n=1,373 patients。Patient sex/smoking history/NSCLC subtype/5-yr mortality 3 cohort 間で大きな差なし、CCTC patients は slightly younger + more advanced disease stage。Molecular prognostic classifier (14-gene panel): 11 cancer-related target genes (BAG1、BRCA1 [DNA repair associated]、CDC6 [cell division cycle 6]、CDK2AP1 [cyclin dependent kinase 2 associated protein 1]、ERBB3 [erb-b2 receptor tyrosine kinase 3]、FUT3 [fucosyltransferase 3]、IL11 [interleukin 11]、LCK [proto-oncogene SRC family tyrosine kinase]、RND3 [Rho family GTPase 3]、SH3BGR [SH3 domain binding glutamate rich protein]、WNT3A [Wnt family member 3A]) + 3 reference genes (ESD [esterase D]、TBP [TATA-box binding protein]、YAP1 [Yes associated protein 1]) を formalin-fixed paraffin-embedded (FFPE) tumor tissues で RT-PCR assay (Clinical Laboratory Improvement Amendments [CLIA]-certified laboratory、Razor Genomics)。Algorithm + cutoff points for low/intermediate/high risk of recurrence は prior publication (Kratz 2012 Kratz et al. Lancet 2012) で確立。TNMB staging system development: Supervised reclassification method で TNM 8th editionに classifier risk (low/intermediate/high) を統合。Training group (n=240) で various upstaging/downstaging optionsを評価 (-1/0/+1、-2/0/+1、-2/+1/+2 等)。Top 3 staging systems を testing group (n=81) で評価し、best performance system を validation で testing。Final TNMB system: TNM 8th edition stage を base に、high-risk recurrenceで +1 upstage、intermediate-riskで no change、low-riskで -1 downstage (Table 3 で完全 mapping)。Statistical methods (5 metrics): Primary outcome = OS after surgical resection (death from any cause、survival time from surgical resection to death or censoring、administrative censoring 5 years post-resection assuming independent censoring)。 (a) Kaplan-Meier analysis + Cox proportional hazards models for observed/predicted survival risk、(b) Time-dependent C-index (concordance index = area under ROC curve) + Royston modified Nagelkerke R^2 statistic、(c) Categorical Net Reclassification Improvement (NRI) for ordered staging systems、(d) Integrated Discrimination Improvement (IDI) + relative IDI (rIDI)、(e) Reclassification Calibration (RC) statistic using Greenwood-Nam-D’Agostino goodness-of-fit test in reclassification table cells with ≥5 events。95% CI by bootstrap resampling、two-sided α=0.05、Stata 15.1 software (StataCorp LP、College Station、TX) + R software (version 3.4.4)。No cell line / mouse strain (clinical retrospective study)。