基于运动的干预对肺癌成人患者抑郁症状的影响：随机对照试验的系统评价与荟萃分析。
全文（开放获取） · 中文译文
对应的完整英文原文片段，我再按您的要求进行逐句翻译。 背景与目的：运动越来越多地用于肺癌的支持性护理，但其对抑郁症状的影响仍不确定。本系统评价和荟萃分析综合了随机证据。材料与方法：检索了四个电子数据库，时间范围为2000年1月至2026年7月。符合条件的研究纳入肺癌成人患者，并将基于运动的结构化干预与常规护理或符合条件的主动康复对照进行比较。每个独立试验选取一项抑郁结果，使用预先记录的规则处理评估时间、结果格式和多干预组。随机效应模型使用REML，并采用Hartung-Knapp推断。结果：17项研究符合纳入标准；15项独立试验（1041名受试者）贡献了主要综合分析。合并效应有利于运动（Hedges' g = -0.75，95% CI -1.18至-0.32；p = 0.002），但异质性相当大（τ2 = 0.4582；τ = 0.6769；I2 = 87.1%；H2 = 7.76；Q(14) = 74.11，p < 0.001），且95%预测区间跨越了零（-2.27至0.76）。所有敏感性分析中方向仍然有利，尽管排除两项影响较大的研究后合并效应减弱至-0.49，I2降至约0%。探索性亚组和研究层面调节变量分析未发现异质性的稳定解释。在因偏倚风险、不一致性和间接性降级后，确定性非常低。结论：基于运动的干预措施可能降低某些 a 的抑郁症状评分。 患有肺癌的成人，但对平均效应的大小和可转移性的信心有限。该证据并未确立对重性抑郁障碍的疗效，也未确定最佳的治疗方式、剂量或实施形式。 1. 引言 2022年，肺癌是全球最常诊断的癌症，并且仍然是癌症相关死亡的主要原因，约占250万新发病例和180万死亡病例[]。然而，对许多患者而言，疾病负担不仅限于肿瘤相关症状和治疗不良反应。心理困扰可能在诊断时出现，并在治疗、康复或疾病晚期阶段持续存在。 来自肺癌患者的纵向证据表明，抑郁症状可能在诊断后出现或持续存在，并与较差的生活质量、更大的症状负担、较低的社会支持感知以及较短的生存期相关[]。抑郁症状在成人癌症患者中也很常见，尽管患病率估计值因研究人群和评估方法的不同而有所差异[]。这些症状很少由单一原因引起。呼吸困难、疲劳、疼痛、身体功能丧失、治疗相关不良反应、社会孤立和肺癌相关病耻感均可能有所贡献[,]。因此，肿瘤学指南建议根据症状严重程度和个体临床需求进行定期评估和管理[,]。然而，提供这种护理可能很困难，特别是当抑郁症状与癌症相关的躯体症状重叠、心理肿瘤学服务可及性有限，或患者已经面临沉重的治疗负担时。 在这种背景下，运动具有相关性，因为它可以在同一支持性护理项目中同时改善身体机能和心理健康。国际指南建议，在癌症治疗期间及治疗后，当临床条件合适时，应进行适当处方的有氧运动和抗阻运动[,]。在癌症幸存者中进行的荟萃分析报告显示抑郁症状有所减轻，而更广泛的抑郁症文献中的证据也支持运动可能发挥的作用[,,]。这些发现不能直接推广到肺癌患者。呼吸功能受限、恶病质、合并症、体能状态和抗癌治疗可能会影响参与运动的能力以及对运动项目的反应。 这一区别很重要，因为肺癌试验所代表的既不是单一的临床人群，也不是统一的干预措施。受试者招募自手术前后、化疗或放疗期间、晚期疾病阶段以及生存期阶段。大多数试验在入组时并未要求存在具有临床意义的抑郁症状，且抑郁通常作为次要结局进行评估。因此，这些证据可能既反映了现有症状的改善，也反映了症状恶化的预防，并不能确立对重度抑郁障碍的治疗疗效。干预措施的范围从有监督的有氧和抗阻训练，到步行、呼吸练习和身心运动、居家康复，以及数字化支持的多组分方案。对照条件的范围从常规照护到教育、自我管理和主动康复。干预强度、抑郁评估工具和评估时间点也有所不同。既往针对肺癌的综述报告了可能有利的心理结局，但始终描述了一个样本量小且临床异质性强的证据基础[,,,]。将这些研究置于共同的统计尺度上或许可以估算出平均效应，但这并不能使它们的人群、干预措施、对照条件或临床环境可以互换。 在此背景下，我们对评估成年肺癌患者结构化运动干预的随机对照试验进行了系统评价和荟萃分析。主要目的是估计干预后抑郁症状的平均效应，并量化试验间的变异。我们还检验了合并估计的稳定性，并探讨了运动类别、抑郁量表、干预持续时间以及选定的研究层面参与者特征是否导致研究间异质性。由于证据基础较小且临床多样性，调节变量分析被视为探索性和假设生成性的。 2. 材料与方法 2.1. 注册 本系统评价和荟萃分析依据《系统评价和荟萃分析的首选报告项目》（PRISMA）2020声明[]进行，研究选择过程见.。由于本研究完全基于既往已发表研究数据的系统评价和荟萃分析，未招募新的人类参与者，作者未实施任何干预，也未收集任何可识别个体水平的数据。因此，本研究无需机构伦理委员会批准和知情同意。本综述纳入的各个研究的伦理批准和知情同意程序由各自的原始研究者和机构负责。 该综述于2026年7月17日在开放科学框架注册表（Open Science Framework Registries）中注册（注册版本1；https://osf.io/vuc87/（于2026年7月17日访问））。数据库检索、筛选、数据提取、偏倚风险评估和统计分析已在注册前完成。由于注册是回顾性的，OSF记录并未起到前瞻性方案的作用。 注册版本1与最终报告之间的分析差异仅限于无法可靠实施或为探索性目的而添加的分析。未进行基于分期和治疗方式的亚组分析，因为可用类别稀疏、重叠且报告不完整。未报告失安全系数（Fail-safe N），因为在存在大量研究间异质性的情况下，该指标的可解释性有限。额外的研究层面调节变量分析作为探索性和假设生成性分析予以报告。 注册记录说明，筛选、数据提取和偏倚风险评估由两名评审者（M.S.和Z.P.）独立进行。标题和摘要筛选及全文评估由同一组评审者独立进行，如【对应章节】所述。数据提取由一名评审者（M.S.）进行，该评审者对照原始报告核对了所有条目。偏倚风险评估由两名评审者（M.S.和Z.P.）进行。未进行独立的重复数据提取。综述各阶段实施的操作流程见【对应章节】、【对应章节】和【对应章节】。 2.2. 纳入标准 合格性采用PICOS（人群、干预、对照、结局和研究设计）定义。合格参与者为经确诊为肺癌的成人（≥18岁），无论组织学亚型、疾病分期、治疗阶段或临床环境如何。合格干预措施包括结构化运动或以运动为基础的康复，单独提供或作为多组分康复计划的重要组成部分提供。这些包括有氧运动、抗阻运动、联合运动、肺康复、呼吸练习、身心运动、步行、数字化提供以及其他结构化运动干预。当运动被安排、重复并根据内容和剂量进行处方，而不仅限于一般体力活动建议时，该运动被视为实质性干预组成部分。合格的对照包括常规护理；教育、注意力或自我管理对照；非运动对照条件；以及当干预组接受额外、更强化或实质性不同的结构化运动或康复计划时的主动康复对照。合格的结局是使用经过验证的工具或可单独提取的抑郁亚量表评估的抑郁症状。 符合条件的研究为平行组随机对照试验，以英文全文、同行评审文章形式发表于2000年1月至2026年7月之间。若参与者未患肺癌、治疗分配未采用随机化、干预不包含结构化运动、抑郁结局不可用或无法单独提取，或出版物非原创同行评审研究文章，则排除该研究。多臂随机试验符合条件，并按照既定的数据合成程序进行处理。 2.3. 检索策略 于2026年7月12日在PubMed/MEDLINE、Scopus、Web of Science核心合集以及Cochrane对照试验中心注册库（CENTRAL）中进行了电子检索。检索限于2000年1月1日至2026年7月12日期间发表的英文文献。检索从PubMed/MEDLINE获得118条记录，从Scopus获得620条记录，从Web of Science核心合集获得418条记录，从CENTRAL获得79条记录，去重前共1235条记录。提供了各数据库完整的检索策略，包括受控词汇、字段限制、应用的限定条件、检索日期以及去重前从每个数据库获得的记录数量。 2.4. 研究筛选 数据库记录通过DOI、PMID以及兼容的文献著录字段进行标准化和去重。两名评审员（M.S. 和 Z.P.）根据入选标准独立筛选标题和摘要。任何一名评审员认为可能符合入选条件的报告均进入全文评估。同一组评审员独立评估全文报告，记录每篇排除的主要原因，并通过讨论和共识解决分歧。源自同一项随机试验的多篇报告使用唯一的研究标识符进行关联，以确保每项试验仅向定量汇总提供一个独立的原始数据集。 2.5. 数据提取 一名评审员（M.S.）将研究特征和结局数据提取到结构化表格中，并在构建分析就绪数据集时对照相应原始报告核对了所有条目。未进行完整的独立重复数据提取程序。随机化人群、基线人口学人群、分析人群和抑郁结局人群分别记录。提取的字段包括参与者特征、干预措施和对照措施特征、抑郁评估工具、评估时间、分析人群、汇总统计量、失访、依从性以及报告中的资助信息（如报告）。 所有符合条件的组水平的抑郁结果均被保留。优先采用已发表的终点均值和标准差，其次是有效推导出的终点、具有可兼容标准误到标准差转换的校正终点，以及当没有可兼容终点时的变化分数。中位数转换使用Luo均值估计量以及Shi或Wan标准差估计量，具体根据所报告的数据酌情选择[,,]。只有当可获得相应组特异性样本量时，SD才按SE×√N计算。然后从每个试验中选取一个独立结果。共享一个对照组的相容且合格的干预组使用Cochrane连续结局公式进行合并，其中对照组只纳入一次；否则，选择最能直接分离出额外结构化运动成分的对比[]。 选定的结局是经过验证的抑郁量表或可单独提取的抑郁亚量表。没有定量试验报告超过一种符合条件且相互竞争的抑郁评估工具，因此无需进行工具优先级选择。选取最接近干预结束时的评估结果，并优先采用干预后立即获得的数据。优先采用已发表的未调整终点；随后考虑有效的衍生或调整终点；仅在没有可用的相容终点时才使用变化评分。来自同一参与者的调整后与未调整估计值不会同时纳入。较长期随访结果单独保留。在所有纳入的工具中，评分越高一致表示抑郁症状越严重；因此，负的标准化效应量有利于运动干预。长期随访结局被保留用于定性报告，但未纳入主要荟萃分析。仅报告医院焦虑抑郁量表（HADS）总分、无法单独提取医院焦虑抑郁量表-抑郁（HADS-D）亚量表的研究被排除在定量综合之外，但保留在定性综述中。 选定的抑郁结果的偏倚风险由两名评审员（M.S. 和 Z.P.）使用Cochrane RoB 2工具进行评估[]。分歧通过讨论和协商一致解决。每个纳入研究的信号问题回答、支持证据、领域级判断和总体判断均被记录。GRADE被应用于主要结局，每次降级均与相应的证据体相关联[]。 2.8. 统计分析 所有分析均使用 R（版本 4.4.1；R Foundation for Statistical Computing, Vienna, Austria）中的 metafor 包 [] 完成。对于每个独立比较，计算 Hedges' g 及其抽样方差时，将干预组作为第一组输入，对照组作为第二组输入。由于评分越高表示抑郁症状越严重，负向效应估计值有利于干预组。异常大的效应估计值与源数据进行了核对，没有研究仅因效应量大小而被排除。采用随机效应模型，因为预期各研究间的真实干预效应存在差异。主要 meta 分析采用限制最大似然（REML）估计和 Hartung–Knapp 推断 []。统计显著性定义为双侧 α=0.05。报告了合并效应估计值、95% 置信区间、95% 预测区间、τ2、τ、I2、H2 和 Cochran's Q 统计量。根据 Higgins 等 []，I2 被解释为研究间不一致性指标，而非方法学质量指标。稳健性分析考察了排除该高风险研究、有影响力的研究、需要转换或汇总统计量的研究、需要合并多个干预组的研究，以及未直接报告终点均值、SDs 和组样本量的研究。留一法分析和替代随机效应模型设定分别进行了检验。亚组分析和单变量研究层面 meta 回归均为探索性分析，未对多重检验进行校正，且 w 被解释为产生假设。使用漏斗图检查、Egger回归检验、Begg秩相关检验和剪补法评估小研究效应；由于试验数量少且研究间异质性较大，这些发现被谨慎解读 [,,]。一项额外的敏感性分析将综合分析限制在以运动为主的比较中。当干预措施的主要治疗内容包含有计划的、重复的、剂量明确的需氧运动、抗阻运动、步行、呼吸、身心或肺康复运动时，被归类为以运动为主；单独的数字化交付或监测不导致排除。教育、心理支持、营养或一般自我管理构成同等治疗成分的项目被排除。 3. 结果 3.1 研究筛选结果 对29篇全文报告进行了资格评估。17项独立的随机研究符合纳入标准，而12篇报告被排除()。15项独立试验各贡献了一个抑郁效应量用于主要定量合成。Liu等人(2024)[]被保留在定性合成中，因为所报告的分布无法使用适用的转换方法进行转换。Tenconi等人[]被保留在定性合成中，因为仅报告了HADS总分，而没有可单独提取的HADS-D亚量表分数。 3.2 研究特征 这17项符合条件的研究评估了常规运动或肺康复、身心或基于呼吸的运动，以及数字化、行为或多组分运动康复，覆盖围手术期、积极治疗期、晚期疾病和生存期等情境（）。区分了随机化人群、基线人口学人群、分析人群和选定的抑郁分析人群，因为几篇报告中的这些分母不同。15项定量试验共纳入543名干预组参与者和498名对照组参与者（总N = 1041）。Cheung 等人 [] 和 Wang 等人 [] 是多臂试验。负重深呼吸组与常规护理进行比较，含有PERMA的组被排除在本分析之外。在定量合并分析中，符合条件干预组在各研究内被合并，共同对照组仅计数一次，从而使每项试验保留一个统计学上独立的主要比较。对照条件包括常规护理、教育或注意力对照、自我管理，以及主动康复。在以肺康复或胸部物理治疗作为主动对照的试验中，效应估计值代表相对于主动康复的比较效应或增量效应，而非运动与不运动的比较。 3.3. 偏倚风险 全部17项研究均接受了针对具体结果的RoB 2评估（）。关注点最常涉及非盲法比较中自我报告的抑郁、结局数据缺失，或对报告结果的选择。 3.4. 主要荟萃分析 主要合并分析纳入
全文（开放获取） · 英文原文
【Abstract】 Background and Objectives: Exercise is increasingly used in supportive care for lung cancer, but its effects on depressive symptoms remain uncertain. This systematic review and meta-analysis synthesized randomized evidence. Materials and Methods: Four electronic databases were searched for reports published from January 2000 to July 2026. Eligible studies enrolled adults with lung cancer and compared a structured exercise-based intervention with usual care or an eligible active rehabilitation comparator. One depression result per independent trial was selected using documented rules for assessment time, outcome format, and multiple intervention arms. A random-effects model used REML with Hartung-Knapp inference. Results: Seventeen studies met the eligibility criteria; 15 independent trials (1041 participants) contributed to the primary synthesis. The pooled effect favored exercise (Hedges' g = -0.75, 95% CI -1.18 to -0.32; p = 0.002), but heterogeneity was considerable (τ2 = 0.4582; τ = 0.6769; I2 = 87.1%; H2 = 7.76; Q(14) = 74.11, p < 0.001) and the 95% prediction interval crossed the null (-2.27 to 0.76). The direction remained favorable in all sensitivity analyses, although exclusion of two influential studies attenuated the pooled effect to -0.49 and reduced I2 to approximately 0%. Exploratory subgroup and study-level moderator analyses did not identify a stable explanation for heterogeneity. Certainty was very low after downgrading for risk of bias, inconsistency, and indirectness. Conclusions: Exercise-based interventions may reduce depressive symptom scores in some adults with lung cancer, but confidence in the magnitude and transferability of the average effect is limited. The evidence does not establish efficacy for major depressive disorder or identify an optimal modality, dose, or delivery format. 【1. Introduction】 Lung cancer was the most frequently diagnosed cancer worldwide in 2022 and remained the leading cause of cancer-related death, accounting for approximately 2.5 million new cases and 1.8 million deaths []. For many patients, however, the burden of the disease extends beyond tumor-related symptoms and treatment adverse effects. Psychological distress may arise at diagnosis and persist during treatment, recovery, or advanced disease. Longitudinal evidence from patients with lung cancer indicates that depressive symptoms may develop or persist after diagnosis and are associated with poorer quality of life, greater symptom burden, lower perceived social support, and shorter survival []. Depressive symptoms are also common among adults with cancer, although prevalence estimates vary according to the population studied and the method of assessment []. These symptoms rarely have a single cause. Dyspnea, fatigue, pain, loss of physical function, treatment-related adverse effects, social isolation, and lung cancer-related stigma may all contribute [,]. Oncology guidelines therefore recommend regular assessment and management according to symptom severity and individual clinical needs [,]. Delivering this care can nevertheless be difficult, particularly when depressive and cancer-related somatic symptoms overlap, access to psycho-oncology services is limited, or patients already face a substantial treatment burden. Exercise is relevant in this context because it may address physical functioning and psychological well-being within the same supportive-care program. International guidelines recommend appropriately prescribed aerobic and resistance exercise during and after cancer treatment when it is clinically suitable [,]. Meta-analyses conducted in cancer survivors have reported reductions in depressive symptoms, while evidence from the broader depression literature also supports a potential role for exercise [,,]. These findings cannot be transferred directly to people with lung cancer. Respiratory limitation, cachexia, comorbidities, performance status, and anticancer treatment may affect both the ability to participate in exercise and the response to an exercise program. This distinction is important because trials conducted in lung cancer represent neither a single clinical population nor a uniform intervention. Participants have been recruited before and after surgery, during chemotherapy or radiotherapy, with advanced disease, and in survivorship settings. Most trials did not require clinically significant depressive symptoms at enrollment, and depression was commonly assessed as a secondary outcome. The evidence may therefore reflect both improvement in existing symptoms and prevention of symptom-worsening and does not establish treatment efficacy for major depressive disorder. Interventions have ranged from supervised aerobic and resistance training to walking, breathing-based and mind–body exercise, home rehabilitation, and digitally supported multicomponent programs. Comparator conditions have ranged from usual care to education, self-management, and active rehabilitation. Intervention intensity, depression instruments, and assessment time points have also differed. Previous lung cancer-specific reviews have reported potentially favorable psychological outcomes but have consistently described a small and clinically heterogeneous evidence base [,,,]. Placing these studies on a common statistical scale may allow an average effect to be estimated, but it does not make their populations, interventions, comparator conditions, or clinical settings interchangeable. Against this background, we conducted a systematic review and meta-analysis of randomized controlled trials evaluating structured exercise-based interventions in adults with lung cancer. The primary objective was to estimate the average effect on post-intervention depressive symptoms and quantify variation across trials. We also examined the stability of the pooled estimate and explored whether exercise category, depression instrument, intervention duration, and selected study-level participant characteristics contributed to between-study heterogeneity. Because the evidence base was small and clinically diverse, moderator analyses were treated as exploratory and hypothesis-generating. 【2. Materials and Methods】 【2.1. Registration】 This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [], and the study selection process is shown in . As this study was a systematic review and meta-analysis based exclusively on data from previously published studies, no new human participants were recruited, no interventions were administered by the authors, and no identifiable individual-level data were collected. Therefore, institutional ethics committee approval and informed consent were not required for the present study. Ethical approval and informed consent procedures for the individual studies included in this review were the responsibility of the respective original investigators and institutions. The review was registered in the Open Science Framework Registries on 17 July 2026 (registration version 1; https://osf.io/vuc87/ (accessed on 17 July 2026)). Database searching, screening, data extraction, risk-of-bias assessment, and statistical analysis had been completed before registration. Because registration was retrospective, the OSF record did not function as a prospective protocol. Analytical differences between registration version 1 and the final report were limited to analyses that could not be implemented reliably or were added for exploratory purposes. Stage- and treatment-based subgroup analyses were not performed because the available categories were sparse, overlapping, and incompletely reported. Fail-safe N was not reported because it was considered to have limited interpretability in the presence of substantial between-study heterogeneity. Additional study-level moderator analyses were reported as exploratory and hypothesis-generating. The registration record stated that screening, data extraction, and risk-of-bias assessment were conducted independently by two reviewers (M.S. and Z.P.). Title-and-abstract screening and full-text assessment were conducted independently by the same reviewers, as described in . Data extraction was performed by one reviewer (M.S.), who checked all entries against the source reports. Risk-of-bias assessment was performed by two reviewers (M.S. and Z.P.). Independent duplicate data extraction was not performed. The procedures implemented at each stage of the review are reported in , , and . 【2.2. Eligibility Criteria】 Eligibility was defined using PICOS (Population, Intervention, Comparator, Outcomes, and Study Design). Eligible participants were adults (≥18 years) with a confirmed diagnosis of lung cancer, irrespective of histological subtype, disease stage, treatment phase, or clinical setting. Eligible interventions comprised structured exercise or exercise-based rehabilitation delivered either alone or as a substantive component of a multicomponent rehabilitation program. These included aerobic, resistance, combined, pulmonary rehabilitation, breathing-based, mind–body, walking, digitally delivered, and other structured exercise interventions. Exercise was considered a substantive intervention component when it was scheduled, repeated, and prescribed in terms of content and dose rather than limited to general physical-activity advice. Eligible comparators included usual care; education, attention, or self-management controls; non-exercise control conditions; and active rehabilitation comparators when the intervention arm delivered an additional, more intensive, or substantively different structured exercise or rehabilitation program. Eligible outcomes were depressive symptoms assessed using validated instruments or separate extractable depression subscales. Eligible studies were parallel-group randomized controlled trials published as full-text, peer-reviewed articles in English between January 2000 and July 2026. Studies were excluded if participants did not have lung cancer, treatment allocation was not randomized, the intervention did not include structured exercise, depression outcomes were unavailable or not separately extractable, or the publication was not an original peer-reviewed research article. Multi-arm randomized trials were eligible and were handled according to the documented data-synthesis procedures. 【2.3. Search Strategy】 Electronic searches were conducted in PubMed/MEDLINE, Scopus, Web of Science Core Collection, and the Cochrane Central Register of Controlled Trials (CENTRAL) on 12 July 2026. Searches were limited to English-language records published from 1 January 2000 to 12 July 2026. The searches retrieved 118 records from PubMed/MEDLINE, 620 from Scopus, 418 from Web of Science Core Collection, and 79 from CENTRAL, resulting in a total of 1235 records before deduplication. provides the complete database-specific search strategies used, including controlled vocabulary, field restrictions, applied limits, search dates, and the number of records retrieved from each database before deduplication. 【2.4. Study Selection】 Database records were standardized and deduplicated using DOI, PMID, and compatible bibliographic fields. Two reviewers (M.S. and Z.P.) independently screened titles and abstracts against the eligibility criteria. Reports considered potentially eligible by either reviewer proceeded to full-text assessment. The same reviewers independently assessed the full-text reports, recorded one principal reason for each exclusion, and resolved disagreements through discussion and consensus. Multiple reports originating from the same randomized trial were linked using a unique study identifier to ensure that each trial contributed only one independent primary dataset to the quantitative synthesis. 【2.5. Data Extraction】 One reviewer (M.S.) extracted study characteristics and outcome data into structured forms and checked all entries against the corresponding source reports during construction of the analysis-ready dataset. A completed independent duplicate data-extraction procedure was not performed. Randomized, baseline-demographic, analyzed, and depression-outcome populations were recorded separately. Extracted fields included participant characteristics, intervention and comparator characteristics, depression instrument, assessment time, analysis population, summary statistics, attrition, adherence, and funding information when reported. All eligible arm-level depression results were retained. Published endpoint means and SDs were preferred, followed by validly derived endpoints, adjusted endpoints with compatible SE-to-SD conversion, and change scores when no compatible endpoint was available. Median conversions used the Luo mean estimator and the Shi or Wan SD estimator, as appropriate to the reported data [,,]. SD was calculated as SE × √N only when the corresponding group-specific sample size was available. One independent result was then selected from each trial. Compatible eligible intervention arms sharing one control group were combined using Cochrane continuous-outcome formulas, with the control group included only once; otherwise, the contrast that most directly isolated the additional structured exercise component was selected []. 【2.6. Outcome Definition】 The selected outcome was a validated depression scale or separately extractable depression subscale. No quantitative trial reported more than one competing eligible depression instrument, so no instrument-priority choice was required. The assessment closest to intervention completion was selected, preferring immediate post-intervention data. Published unadjusted endpoints were preferred; valid derived or adjusted endpoints followed, and change scores were used only when no compatible endpoint was available. Adjusted and unadjusted estimates from the same participants were not entered together. Longer follow-up results were retained separately. Higher scores consistently indicated more severe depressive symptoms across all included instruments; therefore, negative standardized effect sizes favored exercise. Longer-term follow-up outcomes were retained for qualitative reporting but were not included in the primary meta-analysis. Studies reporting only the total Hospital Anxiety and Depression Scale (HADS) score, without a separately extractable Hospital Anxiety and Depression Scale–Depression (HADS-D) subscale, were excluded from the quantitative synthesis but retained in the qualitative review. 【2.7. Risk of Bias and Certainty of Evidence】 Risk of bias for the selected depression result was assessed by two reviewers (M.S. and Z.P.) using the Cochrane RoB 2 tool []. Disagreements were resolved through discussion and consensus. Signaling-question responses, supporting evidence, domain-level judgments, and overall judgments were recorded for each included study. GRADE was applied to the primary outcome, with each downgrade linked to the corresponding body of evidence []. 【2.8. Statistical Analysis】 All analyses were conducted in R (version 4.4.1; R Foundation for Statistical Computing, Vienna, Austria) using the metafor package []. For each independent comparison, Hedges’ g and its sampling variance were calculated with the intervention group entered first and the comparator group second. Because higher scores indicated more severe depressive symptoms, negative effect estimates favored the intervention. Unusually large effect estimates were checked against the source data, and no study was excluded solely because of effect magnitude. A random-effects model was used because the true intervention effects were expected to vary across studies. The primary meta-analysis used restricted maximum likelihood (REML) estimation with Hartung–Knapp inference []. Statistical significance was defined using a two-sided α of 0.05. The pooled effect estimate, 95% confidence interval, 95% prediction interval, τ2, τ, I2, H2, and Cochran’s Q statistic were reported. Following Higgins et al. [], I2 was interpreted as an index of between-study inconsistency rather than methodological quality. Robustness analyses examined the exclusion of the high-risk study, influential studies, studies requiring converted or summary statistics, studies requiring the combination of multiple intervention arms, and studies without directly reported endpoint means, SDs, and group sizes. Leave-one-out analyses and alternative random-effects model specifications were examined separately. Subgroup analyses and univariable study-level meta-regressions were exploratory, were not adjusted for multiple testing, and were interpreted as hypothesis-generating. Small-study effects were assessed using funnel-plot inspection, Egger’s regression test, Begg’s rank-correlation test, and trim-and-fill analysis; these findings were interpreted cautiously because the number of trials was small and between-study heterogeneity was substantial [,,]. An additional sensitivity analysis restricted the synthesis to exercise-dominant comparisons. Interventions were classified as exercise-dominant when their principal therapeutic content comprised scheduled, repeated, and dose-specified aerobic, resistance, walking, breathing, mind–body, or pulmonary rehabilitation exercise; digital delivery or monitoring alone did not result in exclusion. Programs in which education, psychological support, nutrition, or general self-management constituted coequal therapeutic components were exclude. 【3. Results】 【3.1. Study Selection Results】 Twenty-nine full-text reports were assessed for eligibility. Seventeen unique randomized studies met the eligibility criteria, whereas 12 reports were excluded (). Fifteen independent trials contributed one depression effect each to the primary quantitative synthesis. Liu et al. (2024) [] was retained in the qualitative synthesis because the reported distribution could not be converted using the applicable conversion methods. Tenconi et al. [] was retained in the qualitative synthesis because only the total HADS score, without a separately extractable HADS-D subscale, was reported. 【3.2. Study Characteristics】 The 17 eligible studies evaluated conventional exercise or pulmonary rehabilitation, mind–body or breathing-based exercise, and digital, behavioral, or multicomponent exercise rehabilitation across perioperative, active-treatment, advanced-disease, and survivorship settings (). distinguishes randomized, baseline-demographic, analyzed, and selected depression-analysis populations because these denominators differed in several reports. The 15 quantitative trials contributed 543 intervention and 498 comparator participants (total N = 1041). Cheung et al. [] and Wang et al. [] were multi-arm trials. The loaded-deep-breathing arm was compared with routine care, and the PERMA-containing arm was excluded from this analysis. For quantitative synthesis, eligible intervention arms were combined within each study and the shared comparator group was counted only once, thereby retaining one statistically independent primary comparison per trial. Comparator conditions ranged from usual care, education or attention control, and self-management to active rehabilitation. In trials using pulmonary rehabilitation or chest physiotherapy as active comparators, the effect estimates represent comparative or incremental effects relative to active rehabilitation rather than exercise versus no exercise. 【3.3. Risk of Bias】 All 17 studies underwent result-specific RoB 2 assessment (). Concerns most often involved self-reported depression in an unblinded comparison, missing outcome data, or selection of the reported result. 【3.4. Primary Meta-Analysis】 The primary synthesis included