一个好的荟萃分析（Meta-analysis）应该有哪些要素？

声明

1) The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement was used to report this systematic review and meta-analysis.

本句声明论文是按照PRISMA规范写的。所有医学论文大类，包括临床试验、队列研究、病例对照，甚至个案报道，都有相应的写作规范。几乎所有的正规期刊，都要求投稿的论文遵从相应的规范。

2) The meta-analysis protocol was registered in PROSPERO.

本句表明荟萃分析方案在PROSPERO平台上做了注册，这一点非常重要。正规期刊都有明确的规定，不注册不审稿。

Data Sources and Search Strategies

1) A comprehensive literature search was performed in EMBASE, PubMed, the Cochrane Central Register of Controlled Trials, and the Cochrane Database of Systematic Reviews for English-language articles published in print or online in peer-reviewed journals through November 1, 2023; proceedings of main international meetings (ie, American Society of Clinical Oncology, European Society for Medical Oncology, International Association for the Study of Lung Cancer World Conference on Lung Cancer, and European Lung Cancer Congress annual meetings) from January 1, 2008, to November 1, 2023, were included.

需要注意的是，检索范围除了EMBASE、PubMed、Cochrane library外，还包括主流国际会议的公报。如果是IPD（individualized patient data）就更好。

2) The detailed search terms and procedures are described in eTable 1 in Supplement 1; the data sources searched are in eTable 2 in Supplement 1.

本句说明了检索策略，包括具体的关键词和步骤，用表格的方式给出了具体描述。这点做得非常好。细节是判断论文质量的重要指标。

Risk of Bias

The random sequence generation, allocation concealment, blinding, determination of incomplete outcome data, and selection of outcome reporting domains for assessing the risk of bias in the study were performed using the revised Cochrane risk-of-bias RoB 2 tool. Two independent reviewers (M.A.H., A.S.) separately rated trial quality, and differences were resolved by appealing to a third reviewer (A.A.).

此处交代了偏倚评估使用的工具，以及具体执行方案。在荟萃分析写作中，需要详细说明偏倚的向量和评估内容等，具体包括：The random sequence generation, allocation concealment, blinding等。

Statistical Analysis

此部分需要注意的是，荟萃分析使用的具体模型，不仅要在正文方法里写清楚，在论文摘要的方法部分也需要写明。具体到这篇论文的摘要，就有提到：A random-effects model was used for meta-analysis.——本文使用的就是随机效应模型（random-effects model）。

此外，随机效应模型中的“效应”，英文必须是复数“effects”，这一点有很多人犯错。

Study Selection and Characteristics

此部分详细地描述了纳入分析的原始论文：入组患者的人口学特征、关键入排标准、基本试验设计（优效/非劣，公开标签/双盲，对照组，一级终点等）等。荟萃分析的对象是纳入分析的原始论文，因此，必须有详尽的描述，不能一带而过。

Two-year EFS and pCR by Systematic Review

Seven of the RCTs reported EFS data, but for the RATIONALE-315 trial, these data were not available yet. All 7 RCTs with EFS data consistently reported higher 2-year EFS rates with neoadjuvant ICI-chemotherapy (from 62.4% to 76.9%) compared with chemotherapy with or without placebo (from 38.7% to 67.6%), with significant differences in all 5 RCTs performing a statistical comparison (Table).

Two-year EFS and pCR by Meta-analysis

In the meta-analysis, neoadjuvant ICI-chemotherapy was associated with better 2-year EFS (HR, 0.57; 95% CI, 0.50-0.66; P < .001) (Figure 1A) compared with control arms with neoadjuvant chemotherapy. There was also an association with increased pCR rate (RR, 5.58; 95% CI, 4.27-7.29; P < .001) (Figure 1B).

EFS and pCR by Systematic Review是对核心结果的描述——关键指标EFS和pCR；Two-year EFS and pCR by Meta-analysis是对荟萃分析本身结果的描述——风险比HR是多少。

这两部分对比来看，Meta-analysis的关键结果，确实既有直观的描述，也有统计结果。但更重要、花费更多篇幅描述的是前者Systematic Review。Meta-analysis只是研究所采用的数学统计方法而已，Systematic Review才是研究真正的重点。从临床医生看问题或者审稿人看文章的角度，比起统计结果（风险比），同行更关心核心结果的具体指标。

当然，并非所有的论文里都需要安排by Systematic Review和by Meta-analysis 两个独立的小节。外行看热闹，内行看门道，万变不离其宗，只要看破了门道，就能把论文写得更好。