各位好！今日与大家分享一篇近期发表在JTCVS上分析局部晚期食管癌新辅助放化疗后达到临床完全缓解（cCR）的文章。研究从马尔科夫链决策分析出发，分析了达到cCR患者行食管癌根治术或积极随访的潜在获益与风险分析。一起来学习下！
Locally Advanced Esophageal Cancer Patients With Complete Clinical Response Post Neoadjuvant Chemoradiotherapy: A Markov Decision Analysis of Esophagectomy versus Active Surveillance
Adom Bondzi-Simpson, MD MSc, Tiago Ribeiro, MD, Angelo Grant, BHSc, Michael Ko, MD PhD, Natalie Coburn, MD MPH, Julie Hallet, MD MSc, Girish S. Kulkarni, MD PhD, Biniam Kidane, MD MSc
JTCVS 19 Apr 2024
Objective: Chemoradiation followed by esophagectomy is a standard treatment option for locally advanced esophageal cancer (LAEC) patients. Esophagectomy is a high-risk procedure, and recent evidence suggests select patients may benefit from omitting or delaying surgery. This study aims to compare surgery versus active surveillance for LAEC patients with complete clinical response (cCR) after neoadjuvant chemoradiotherapy (nCRT).
目的：放化疗后食道切除术是局部晚期食道癌(LAEC)患者的标准治疗方案。食道切除术是一种高风险的手术，最近的证据表明，特定的患者可能会从省略或推迟手术中受益。本研究旨在比较新辅助放化疗(NCRT)后临床完全缓解(CCR)的LAEC患者的手术和积极监测。
Methods: Decision analysis with Markov modelling was utilized. The base case was a 60-year-old male with T3N0M0 esophageal cancer with cCR after nCRT. The decision was modelled for a 5-year time horizon. Primary outcomes were life-years (LY) and quality-adjusted life-years (QALYs). Probabilities and utilities were derived through literature. Deterministic sensitivity analyses were performed using ranges from literature with consideration for clinical plausibility.
方法：运用马尔可夫模型进行决策分析。基础病例为60岁男性，T3N0M0食道癌，nCRT术后CCR。这一决定是以5年的时间范围为模型的。主要结果是寿命年(LY)和质量调整寿命年(QALY)。概率和效用是通过文献得出的。确定性敏感性分析使用文献中的范围进行，并考虑到临床的可信性。
Results: Surgery was favoured for survival with an expected LY of 2.89 versus 2.64. After incorporating quality of life, active surveillance was favoured with an expected QALY of 1.70 versus 1.56. The model was sensitive to probability of recurrence on active surveillance (threshold value 0.598), probability of recurrence being resectable (0.318) and disutility of prior esophagectomy (-0.091). The model was not sensitive to perioperative morbidity and mortality.
结果：手术对生存有利，预期LY为2.89，而不是2.64。在纳入生活质量后，积极监测受到青睐，预期的QALY分别为1.70和1.56。该模型对主动监测的复发概率(阈值0.598)、可切除的复发概率(0.318)和既往食道切除术无效(-0.091)敏感。该模型对围手术期发病率和死亡率不敏感。
Conclusions: Our study finds that surgery increases life expectancy but decreases quality-adjusted life years. Although the incremental change in QALY for either modality is insufficient to make broad clinical recommendations, our study demonstrates that either approach is acceptable. As probabilities of key factors are further defined in the literature, treatment decisions for patients with LAEC and a cCR after nCRT should consider histology, patient values, and quality of life.
结论：我们的研究发现，手术增加了预期寿命，但减少了质量调整后的寿命。尽管两种方法的QALY的增量变化都不足以提出广泛的临床建议，但我们的研究表明，这两种方法都是可以接受的。由于文献中进一步定义了关键因素的概率，因此对LAEC患者和nCRT后CCR患者的治疗决策应考虑组织学、患者获益和生活质量。
Keywords: Esophageal cancer, Complete clinical response, Decision analysis
Central message
Surgery and active surveillance are both appropriate options in patients with locally advanced esophageal cancer and a complete clinical response after neoadjuvant chemoradiotherapy.
Perspective statement
In this study we have applied decision analysis with Markov modelling to mathematically compared surgery and active surveillance in locally advanced esophageal cancer patients with a complete clinical response after neoadjuvant chemoradiotherapy. This work has highlighted key factors to decision making through probabilistic rationality. Together this work can guide future research directions.
1.首先这里先对马尔科夫链模型进行简单介绍，马尔科夫链模型是基于贝叶斯思想，其基本假设是未来状态仅取决于当前状态，而不取决于之前发生的事件。事件套事件，事情推动事情（套娃的感觉出来了）。目前在医学领域常见的应用场景是流行病学，如根据某地某疾病传播链，预测某临近地区发生疾病流行的程度。感兴趣的可以参看下面这两本书。
2.上细节：
首先文章临床问题的把握较为准确，选择了目前悬而未决的临床决策困境。针对临床完全缓解的患者，是否需要接受食管癌根治术或可行积极随访监测。
其次方法上，不同于以往临床研究。本文几乎是开局一个案例，后面全靠调研。先基于多阶段多状态、贝叶斯思想回顾了此场景下的疾病转归史。定义了此临床场景下的多种状态，不同状态之间有不同的转化率。这里其实要重点学习下本研究是如何进行前提假设、敏感性分析和Validation。针对很多外科困难决策场景的研究 其实可以学习下这种思路。别看此文章看似考虑了很多因素，真正一个病人放在你面前，你就知道风险与获益没有文章中提的那么简单。
说简单点这东西有点像临床路径，但是比目前的临床路径更加关注疾病的不同状态和不同状态间的转化率。基于目前发表的临床研究数据，带入了不同阶段的条件概率。从而得出预期寿命和质量调整寿命年。
结论中规中矩，积极随访有更好的质量调整寿命年，但是手术有更好的预期寿命。有趣的是作者也知道这个研究并非基于严谨的临床病例数据，某种意义上算是一种富有循证概率的思想实验，整整三大段的批评与自我批评，可能想传递的一个概念：千万别当真，咱这就是算命。
3. 众所周知，在食管癌cCR处理策略上，SANO (Surgery versus Active surveillaNce for Oesophageal cancer) 系列trials是最值得期待的研究之一。
这篇文章其实是从统计角度出发的解读，同类话题还有从真实世界、从residual tumor、从ctDNA角度出发的研究和文章。Waiting to watch的叫法听起来还是比SANO好多了，cCR能不能对手术说No有时候也得评估个体抗肿瘤免疫与肿瘤进化的强弱。
https://www.sanotrial.nl/
SANO-1 trials
SANO-2 trials
10.3390/jcm12082873
10.1016/j.urolonc.2020.10.006
目录
1. INTRODUCTION
2. Methods
2.1 Decision analysis with Markov Modelling
2.2 Model Overview
2.3 Base Case
2.4 Surgery Arm (Figure 1-2)(Table 1-2)
2.5 Surgery Arm Surveillance - Markov cycle
2.6 Active Surveillance Arm - Markov Cycle
2.7 Assumptions
2.8 Outcomes
2.9 Model Probabilities and Utilities
2.10 Sensitivity Analyses
2.11 Validation
2.12 Ethical Considerations
3. Results
3.1 Main Findings (Table 3)
3.2 Recurrence on active surveillance and resectability (Figure 3)
3.3 Validation (Figure 4)
4. DISCUSSION (Figure 5)
— 图表汇总—
2. Methods
2.4 Surgery Arm
Figure 1. Schematic diagram representing the decision analysis tree and entry into Markov cycles. The base case patient begins at the decision node subsequently entering the active surveillance arm or surgery arm. Subscript in Markov cycles correspond to Figure 2 Markov cycle schematics.
Legend: “M” = Markov cycle, [] = Decision node, 0 = chance node, ypT0N0 = complete pathologic response post-resection, ypT1-4N0 = node negative with residual disease post-resection, ypT1-4N1-3 = residual disease and node positive post-resection
Figure 2. Markov cycles are denoted by the subscript as per Figure 1.
A) M1:: schematic for the active surveillance arm after the decision node in the DA tree.
B) M2: schematic for patients with a complete pathologic response (ypT0N0) after surgery.
C) M3: schematic for patients with residual disease after surgery.
Table 1. Model probability estimates and reference sources. Range provided were used in deterministic sensitivity analysis and selected based on clinical plausibility and literature review.
Legend: cCR = complete clinical response, MIS = minimally invasive surgery
Table 2. Model utility estimates and reference sources. Range provided were used in deterministic sensitivity analysis and selected based on clinical plausibility and literature review. Disutility’s were subtracted from the transition health state disutility’s and applied to a cycle length while in a given health state.
Legend: cCR = complete clinical response, MIS = minimally invasive surgery
3. Results
3.1 Main Findings
Table 3. Model threshold values determined through one-way deterministic sensitivity analysis.
Parameters presented represent all parameters that model was sensitive to.
3.2 Recurrence on active surveillance and resectability
Figure 3. One-way (A, B) and Two-way (C) deterministic sensitivity analysis for the probability of recurrence on active surveillance and the probability of recurrence being local (resectable) with thresholds values highlighted.
Legend: QALYs = Quality-adjusted life years
3.3 Validation
Figure 4. Survival analysis according to Markov cycle for active surveillance depicting overall survival (OS) and progression free survival (PFS). Number of Markov cycles correspond to the number of 3-month cycles, labels denote successive years.
4. DISCUSSION
Figure 5. Graphical Abstract
