TCGA-LUAD + GSE72094
Clinical and molecular data harmonization, four-model comparison, repeated nested cross-validation, and exploratory external validation of two models.整合临床与分子数据,比较四种模型,进行重复嵌套交叉验证,并对其中两个模型完成探索性外部检验。
GENMABS · TREATMENT BENEFIT RESEARCH健脉 · 治疗获益研究
Compare treatment strategies. Investigate who may benefit. We assess whether clinical and molecular data support the research question, then develop and evaluate models with explicit uncertainty and limitations.比较治疗方案,研究哪些患者可能获益。先评估临床与分子数据能否支持研究问题,再开发和评估模型,明确呈现不确定性与限制。
THE FRAMEWORK研究流程框架
Explore completed research stages and the current validation status. The report separates measured results from claims that still require evidence.点击查看已完成的研究阶段和当前验证状态;报告将实际结果与尚待证实的结论分开呈现。
Treatment arms · Clinical variables · Molecular features · Longitudinal outcomes治疗方案 · 临床变量 · 分子特征 · 纵向结局
Completeness · Event counts · Treatment-group overlap完整性 · 事件数 · 治疗组重叠性
02EXPLORE NOW点击查看 ↗Mechanism · Biomarker coding · Timing机制 · 标志物编码 · 时序
03EXPLORE NOW点击查看 ↗Five Cox models · Fixed specification · Internal evaluation五个Cox模型 · 固定设置 · 内部评估
04VIEW STATUS查看状态 ↗Internal evaluation complete · External validation pending内部评估已完成 · 外部验证待完成
Explore the five-model comparison, uncertainty and current conclusion ↗查看五模型比较、不确定性及当前结论 ↗
Stages 1–3 and the evidence report show completed research work. Stage 4 describes the validation boundary. These are inspectable case materials, not an automated patient treatment recommender.第1–3阶段及证据报告展示已完成的研究工作;第4阶段说明验证边界。这些是可查阅的案例材料,不是自动患者用药推荐系统。
STAGE 01 · AVAILABLE NOW第一阶段 · 现可展示
Two independent public-data feasibility and modeling projects show how our analysis moves from data review to a defensible go/no-go decision. These are research examples, not clinical treatment products.两项基于公开数据的研究展示了从数据审查到可解释 GO/NO-GO 判断的分析过程。这些是研究案例,不是临床用药推荐产品。
Clinical and molecular data harmonization, four-model comparison, repeated nested cross-validation, and exploratory external validation of two models.整合临床与分子数据,比较四种模型,进行重复嵌套交叉验证,并对其中两个模型完成探索性外部检验。
Patient-level audit of clinical, genomic, treatment, and timeline variables; biology-guided pathway-state design and eligibility screening for two treatment strategies.审查患者级临床、基因组、治疗与时间线变量;依据生物机制设计通路状态,筛查两种治疗方案的可比较患者。
The 625-person MSK example above is an earlier pathway/chemotherapy feasibility analysis. The PD-L1 × ICI study below uses a separately defined 489-person LUAD cohort; the two denominators and treatment comparisons are not interchangeable.上方625人MSK案例是早期通路/化疗可行性研究;下方PD-L1×ICI研究使用另行定义的489人肺腺癌队列,两者人数与治疗比较不可互换。
Continue to the PD-L1 study ↓继续查看PD-L1研究 ↓ILLUSTRATIVE RAW INPUT原始数据结构示例
patient_id | drug_name | start_date
case_001 | regimen A | 2024-01-10
case_001 | regimen B | missingRESEARCH-READY OUTPUT研究就绪矩阵示例
patient_id | index_date | exposure | qc_flag
case_001 | 2024-01-10 | reviewed | checkIllustrative synthetic rows above—not actual patient records. Raw datasets are not uploaded on this webpage.上方仅为合成示例,不是真实患者记录。本页面不会接收原始患者数据上传。
02 · MSK-CHORD LUAD02 · MSK-CHORD 肺腺癌
A mechanism-informed biomarker choice, translated into explicit patient-level definitions.从生物机制选择标志物,再转化为明确的患者级变量定义。
PD-L1 on tumor cells can bind PD-1 on T cells and suppress T-cell activity. Anti-PD-1 or anti-PD-L1 antibodies can interrupt this checkpoint interaction.肿瘤细胞上的PD-L1可与T细胞上的PD-1结合,抑制T细胞活性。抗PD-1或抗PD-L1抗体可阻断这一免疫检查点相互作用。
The model uses the latest documented tumor-cell PD-L1 positive/negative result on or before the treatment index date. It does not measure PD-1 expression, a mutation-defined pathway state, or the entire pathway’s activity.模型使用治疗开始当日或之前最近一次记录的肿瘤细胞PD-L1阳性/阴性结果。它不是PD-1表达、由突变定义的通路状态,也不是完整通路活性。
The recorded ICI category pools checkpoint agents and combinations; it is not limited to one anti-PD-L1 antibody. Not every ICI directly targets the PD-1/PD-L1 binding interface.本轮ICI类别合并了多种检查点药物及联合用药,不限于一种抗PD-L1单抗;并非所有ICI都直接作用于PD-1/PD-L1结合界面。
Recorded positive vs negative; measured before index treatment. Missing is not negative. No common TPS threshold is imposed on heterogeneous source reports.使用治疗前报告的阳性或阴性;缺失不当作阴性。不同来源报告不强行套用统一TPS阈值。
Any named ICI within the 28-day initiation window vs none recorded in that window. This is not an ever/never ICI comparison.起始28天窗口内记录已命名ICI,与该窗口未记录ICI比较;不是整个随访期间曾用/从未用ICI。
Does the ICI-associated relative hazard differ by pretreatment PD-L1 status? An interaction is a hypothesis to test, not proof of patient-level benefit.ICI相关相对死亡风险是否随治疗前PD-L1状态变化?交互项用于检验假设,本身不证明个体获益。
PD-L1 negativity does not mean that cancer arose through a different single pathway, and PD-L1 positivity does not guarantee response. RTK/RAS/MAPK or PI3K mutations are not direct measurements of checkpoint blockade sensitivity.PD-L1阴性不代表肿瘤由另一条特定通路引起,阳性也不保证响应。RTK/RAS/MAPK或PI3K突变不能直接等同于免疫检查点阻断敏感性。
Documented stage IV LUAD before index; entry by the landmark; pretreatment PD-L1; no targeted drug/ADC recorded in the index window; observed beyond index + 28 days. OS is remaining survival from that landmark. Patients without ICI may still have chemotherapy, research agents, or other therapies.治疗前记录为IV期肺腺癌;患者在landmark前进入观察;PD-L1在治疗前;起始窗口排除记录靶向药/ADC者;观察超过治疗开始后28天。OS从该landmark计算剩余生存。未记录ICI者仍可能接受化疗、试验药或其他治疗。
The current 489-person cohort is not the earlier 625-person feasibility cohort. It does not require an identical chemotherapy backbone, so it cannot isolate an add-on ICI effect.当前489人不是早期625人可行性队列。本轮没有要求相同化疗基础,因此不能单独识别“增加ICI”的效应。
Biology reference: NCI · Immune Checkpoint Inhibitors ↗机制参考:NCI · 免疫检查点抑制剂 ↗
03 · MSK-CHORD LUAD03 · MSK-CHORD 肺腺癌
Five model comparisons separate clinical risk, biomarker information, treatment exposure, and their interaction.五模型比较分别考察临床风险、标志物信息、治疗记录及其交互。
| Model模型 | Inputs输入 | Question比较内容 |
|---|---|---|
| M1 | C | Clinical and background-treatment baseline临床及背景治疗基线 |
| M2 | C + P | Baseline + pretreatment tumor PD-L1基线 + 治疗前肿瘤PD-L1 |
| M3 | C + T | Baseline + index-window ICI exposure基线 + 起始窗口ICI记录 |
| M4 | C + P + T | Both main effects; no interaction同时加入两个主效应,无交互 |
| M5 | C + P + T + P × T | Main effects + biomarker–treatment interaction主效应 + 标志物与治疗交互 |
C = expanded clinical/background-treatment baseline; P = pretreatment PD-L1; T = named ICI in the initiation window. P and T are excluded from C.C为扩展临床/背景治疗基线;P为治疗前PD-L1;T为起始窗口已命名ICI。C中不含P和T。
489 patients · 302 deaths · 187 censored. The same eligibility rules and fold assignments apply to M1–M5.489人 · 302例死亡 · 187人删失。M1–M5使用同一入组规则和交叉验证分折。
Ridge Cox, alpha = 1; Breslow ties. Standardization and missing-value handling use training folds only. No tuning to obtain a smaller p-value.Ridge Cox,alpha=1,Breslow并列处理。标准化与缺失处理仅在训练折估计;不为降低p值调参。
Event-stratified 5-fold cross-validation, repeated 3 times. Out-of-fold discrimination and IPCW prediction error; calibration from the first repeat.按事件分层的5折交叉验证,重复3次。使用折外预测计算区分度和IPCW误差;校准采用首轮折外预测。
Sex; ECOG categories and missingness; liver/brain site documentation in the 90 days before index; log(1 + documented extra-pulmonary site categories); no-site-documentation indicator; outside-MSK prior-treatment documentation within 365 days and missingness; cisplatin indicator in the initiation window. Histology is constant in this LUAD-only cohort and is dropped.性别;ECOG分类及缺失;治疗前90天肝/脑部位文档;log(1+已记录肺外部位类别数);无部位文档标记;治疗前365天院外既往治疗文档及缺失;起始窗口顺铂指示。组织学因全体为肺腺癌而为常数,被剔除。
No positive site record is not proof of no metastasis. The cisplatin indicator is not full adjustment for chemotherapy. Current age, undated smoking summaries and post-treatment laboratory results are not included.没有阳性部位文档不等于没有转移;顺铂指示不能完整控制化疗差异。本轮未纳入当前年龄、无日期的吸烟汇总和治疗后血液指标。
04 · MSK-CHORD LUAD04 · MSK-CHORD 肺腺癌
Internal evaluation completed. External validation of the original PD-L1 × ICI model remains outstanding.内部评估已完成。原PD-L1×ICI模型的外部验证仍待完成。
Repeated out-of-fold evaluation, calibration review, paired conditional bootstrap comparisons, and an exploratory interaction analysis. These assess performance within the development source.已完成重复折外评估、校准审查、配对条件bootstrap比较及探索性交互分析;这些属于开发数据来源内部的评估。
An independent cohort still needs compatible pretreatment PD-L1, clinical inputs, treatment definitions and OS timing. We do not label the current model externally validated.仍需要治疗前PD-L1、临床输入、治疗定义和OS起点兼容的独立队列。目前不标注本模型已获外部验证。
Two independent published trial datasets were used for a newly simplified sex/ECOG prognostic model. Its discrimination and calibration were limited. It omits PD-L1 and its treatment interaction, and the trial treatment contrast differs from this cohort.另用两个已发表试验数据集测试了新建的性别/ECOG简化预后模型,区分度和校准均有限。该模型不含PD-L1及其治疗交互,试验治疗比较也与当前队列不同。
That work is a separate transferability assessment, not external validation of the five models shown here. No external interaction p-value or validated patient-level benefit is claimed.该工作属于另行开展的可迁移性评估,不是这里五个模型的外部验证。本页面不宣称外部交互p值或经验证的患者级治疗获益。
05 · OUTPUT · MSK-CHORD LUAD05 · OUTPUT · MSK-CHORD 肺腺癌
The result is an auditable research finding, with uncertainty and limitations presented alongside the measured performance.将实际性能、统计不确定性和限制一同呈现,形成可审计的研究结论。
M5 mean internal C-index. M1: 0.5994. The improvement is small.M5内部平均C-index;M1为0.5994。增量较小。
Mean ΔC, M5 − M4. The first-repeat paired interval crosses zero.M5−M4平均ΔC;首轮配对增量区间跨0。
Exploratory evidence report available; a validated individual treatment-benefit model has not been established.可交付探索性证据报告;尚未建立经验证的个体治疗获益模型。
| Model模型 | C-index ↑ | Brier 12m ↓ | Brier 24m ↓ | Calibration slope 24m*24月校准斜率* |
|---|---|---|---|---|
| M1 | 0.5994 | 0.2201 | 0.2446 | 0.553 |
| M2 | 0.6001 | 0.2160 | 0.2408 | 0.596 |
| M3 | 0.5973 | 0.2208 | 0.2453 | 0.547 |
| M4 | 0.5979 | 0.2167 | 0.2414 | 0.592 |
| M5 | 0.6061 | 0.2139 | 0.2367 | 0.668 |
C-index and Brier: means over three repeats. *Calibration slopes: first-repeat 24-month IPCW logistic estimates; ideal slope = 1. These are internal results, not external performance.C-index和Brier为三次重复均值。*校准斜率为首轮24月IPCW逻辑校准估计,理想值为1。全部为内部评估,不是外部性能。
An exploratory adjusted PD-L1 × ICI hazard interaction was observed (Wald p = 0.0099). Predictive improvement remains modest and uncertain. This is an observational association, not proof that ICI benefits or harms a particular patient.观察到探索性调整后的PD-L1×ICI风险交互(Wald p=0.0099)。预测增量仍较小且不确定。这是观察性关联,不证明ICI使某位患者获益或受害。
The detailed report includes the four-cell audit, exact definitions, model comparison, calibration, uncertainty, limitations and source references. Only aggregate results are downloadable here.详细报告包含四格审计、精确定义、模型比较、校准、不确定性、限制和参考来源。此处仅提供汇总结果下载。
BEFORE A PROJECT开始项目之前
We first define the decision, check the structure and limitations of the available dataset, and agree on a realistic scope and deliverables.我们首先明确需要回答的研究决策,审查数据结构和限制,再确定合理的分析范围及交付成果。
Cohort & IDs队列与标识符Patient/sample linkage, duplicates, inclusion criteria.患者与样本对应、重复记录、入组标准。
Clinical & molecular variables临床与分子变量Definitions, missingness, assay availability, timing.变量定义、缺失率、检测覆盖和测量时间。
Treatment & outcomes治疗与结局A/B definitions, follow-up, events, confounding and overlap.A/B定义、随访、事件、混杂与重叠性。
Decision-ready report决策支持报告What is supported, what is uncertain, and what data are needed next.哪些结论受支持、哪些仍不确定以及下一步数据需求。
START WITH THE QUESTION从研究问题开始
Share a brief, non-identifying description of your research question and available data. We will discuss the proposed review scope before any secure data transfer.请先介绍不含患者身份信息的研究问题及可用数据概况。确认项目范围及安全交付要求之后,再讨论数据传输。
For research use. No direct patient care recommendations. Do not email identifiable health records or upload PHI through this page.Public-source analyses do not imply endorsement by NIH, NCI, TCGA, MSK, or GEO.仅供科研使用,不提供直接临床用药推荐。请勿通过此页面或普通邮件发送可识别患者健康信息。公共数据分析不代表 NIH、NCI、TCGA、MSK 或 GEO 背书。