比较机器人辅助与传统全膝关节置换术后的疼痛缓解和功能结局：一项采用GRADE评估并含随机对照试验meta回归的meta分析。
全文（开放获取） · 中文译文
摘要 尽管机器人辅助全膝关节置换术（RA-TKA）因其提高手术精确度的潜力而被越来越多地采用，但其与传统全膝关节置换术（C-TKA）相比对功能结局的影响仍不明确。现有综合研究受限于将观察性数据与随机化数据合并、仅关注术后绝对评分而未评估恢复轨迹，以及缺乏对异质性的严格探索。为克服这些局限，我们仅对随机对照试验（RCTs）进行了方法学严谨的荟萃分析，纳入了全面的元回归、平台特异性亚组分析，并使用推荐意见评估、制定与评价分级（GRADE）框架评估证据确定性。我们全面检索了PubMed、Web of Science、Scopus、Cochrane Library和Google Scholar（第1-50页），检索时间自其建库起至2025年12月27日。数据采用随机效应荟萃分析和Hartung-Knapp校正进行合并。在可行时，进行了不同调节变量的元回归和按机器人平台的亚组分析，以识别异质性来源。共纳入24项RCT，涵盖3，425例患者。合并分析显示，RA-TKA与C-TKA在西安大略和麦克马斯特大学骨关节炎指数（WOMAC）功能评分变化（MD 0.14；95% CI - 5.88至6.16；p = 0.96）、功能性膝关节学会评分（KSS）变化（MD 0.51；95% CI - 3.43至4.44；p = 0.78）或牛津膝关节评分（OKS）变化（MD 2 .07；95% CI -6.82 至 10.96；p = 0.51）。尽管亚组分析显示仅在 6 个月时 OKS 有显著改善（MD 2.25；p = 0.01），但该效应未达到最小临床重要差异（MCID），且在其他时间点未能持续。在活动范围、VAS 疼痛评分变化、KSS 满意度或住院时间方面未观察到显著差异。RA-TKA 的手术时间显著更长（MD 22.34 分钟；95% CI 13.66 至 31.02；p < 0.01），但鉴于极高的异质性（I² = 98.53%）以及该估计可能包含机器人 adoption 早期学习曲线期间实施的病例，此估计应谨慎解读。试验序贯分析提示该发现不太可能归因于随机误差，尽管未达到所需信息量。RA-TKA 与 C-TKA 在功能改善、疼痛缓解及大多数围手术期结局方面相当。最一致的信号是 RA-TKA 手术时间延长，有利于传统手术，但确切幅度仍不确定，并可能受学习曲线影响。所有评估结局的证据确定性为低至极低，强调需谨慎解读。当前证据不支持在 TKA 中常规采用 RA-TKA 替代 C-TKA。未来需要样本量充足、多中心、长期随访且一致报告术者经验的 RCT，以明确 RA-TKA 在当代实践中的作用。 引言 在美国，每年有超过一百万例全膝关节置换术（TKA）实施，目前的预测估计到2040年这一数量将超过三百万例[]。文献支持其有效性；然而，传统全膝关节置换术（C-TKA）在最佳假体对线和软组织平衡方面仍面临挑战，约20%的患者在术后表示不满意[]。这些技术挑战促使了机器人辅助全膝关节置换术（RA-TKA）的发展，该技术结合了三维成像、计算机导航和机器人辅助，以促进对线并可能提高患者满意度[, ]。RA-TKA的基本假设是，更精确的技术对线将带来更优的患者中心结局[]。然而，这种增强的手术精确性的临床意义仍存在争议，研究表明影像学改善与功能结局之间的关系并不一致[, ]。这种脱节说明关注患者中心结局（而非仅关注影像学结局）的重要性。 因此，术后功能恢复是判断手术成功与否的主要基准。这种恢复通常使用经过验证的标准化评估工具进行量化，最著名的是KSS、OKS和WOMAC。这些测量评估患者执行基本日常生活活动的能力、一般功能以及关节活动范围，所有这些都反映了身体独立性和生活质量[]。TKA术后功能恢复的性质和程度是康复过程、所提供术后护理水平以及患者长期健康状况的关键变量。术后疼痛，使用视觉模拟评分（VAS）评估，构成另一个关键指标，并且可能显著影响恢复和患者满意度[]。术后疼痛管理不当或未得到控制可能会延迟功能康复，延迟恢复日常活动，并成为患者不满意和翻修手术的原因[, ]。手术技术与疼痛报告之间的关系已得到证实，但仍未被完全理解。然而，理论上更高的精准度可能优化关节生物力学并减少异常应力分布。患者满意度是首要的以患者为中心的终点指标，涵盖多个领域，包括疼痛缓解、功能改善、术前期望的实现以及对手术成功的总体感知。鉴于五分之一的TKA患者尽管在客观上取得了放射学上的成功却仍不满意，确定最大化这一关键指标（即患者满意度）的策略至关重要。 手术期间钙终点是一项研究重点 [, ]。 尽管在关节成形术实践中，比较这两种技术的荟萃分析数量不断增加，但仍存在若干关键空白[–]。首先，许多既往的合成分析将随机研究和观察性研究合并在一起，可能引入混杂因素，并掩盖源自高等级证据的真实治疗效果。然而，本研究仅限于随机对照试验（RCT），从而确保更稳健的因果推断，并最大限度地减少观察性数据中固有的混杂。其次，虽然既往荟萃分析主要关注术后绝对结局，但此类指标不能充分反映患者改善的程度。相比之下，我们的研究 uniquely 聚焦于自基线的变化评分，这更能准确反映恢复程度，并考虑了不同研究中患者基线特征的潜在差异。这一区分具有临床意义，因为两名术后评分相同的患者，根据其术前状态，可能经历了截然不同的改善幅度。第三，我们进行了全面的荟萃回归分析，以发现多个临床相关协变量间异质性的来源，这种方法在既往研究中有限或缺失。这使得能够更严格地探索潜在效应修饰因素及研究间变异来源。第四，在可行的情况下，我们按特定机器人平台进行亚组分析，从而能够识别潜在的特定系统效应，而不是将机器人辅助视为一个整体。 将全膝关节置换术视为一种同质性干预措施。第五，我们针对手术时间这一异质性极高的结局指标进行了试验序贯分析（TSA），以评估观察到的差异是否能够抵御随机误差的影响，从而为该重要结局提供更确切的评价。最后，我们应用了GRADE框架，以透明且临床可解释的方式评估了研究结果的证据强度。总体而言，这些方法学创新为当前证据基础提供了更为严谨且具有临床意义的综合，并解决了既往荟萃分析的关键局限性。 方法 检索策略 本综述严格遵循PRISMA 2020建议[]，并在PROSPERO（CRD420261301318）上注册。两名作者独立对五个主要生物医学数据库进行了彻底检索，涵盖截至2025年12月27日所有符合条件的已发表文献：PubMed、Web of Science、Scopus、Cochrane Library和Google Scholar（第1-50页）。此外，我们通过手工检索纳入研究及相关综述的参考文献列表来补充检索，以查找电子数据库检索未捕获的任何额外符合条件记录。详细检索策略见补充材料中的补充表1。 纳入和排除标准 本综述仅限于随机对照试验（RCT），以避免观察性数据中存在的混杂因素，并对治疗影响进行严格的因果评估。纳入人群为计划因任何适应证接受全膝关节置换术的成年患者。干预措施为机器人辅助全膝关节置换术（RA-TKA），可使用任何可用的机器人系统。对照措施为采用手动或常规技术、无机器人辅助的常规全膝关节置换术（C-TKA）。为保证纳入资格，研究需提供以下一项或多项临床终点的数据：疼痛与功能测量：通过视觉模拟评分（VAS）变化评估术后疼痛严重程度；标准化功能指标，包括牛津膝关节评分（OKS）、美国膝关节协会评分（KSS）和西安大略和麦克马斯特大学骨关节炎指数（WOMAC）功能子量表的变化。手术及围手术期结局：通过关节活动度（ROM）评估关节活动性；住院时长（LOS）；手术时间（分钟）；估计的术中或累计失血量（mL）；术后伤口引流量（mL）；围手术期血红蛋白波动；全身炎症指标；以及经过验证的患者满意评分。排除标准为非随机对照试验、病例系列、研究方案或摘要；未比较RA-TKA与C-TKA或缺乏对照组的研究；未报告相关临床结局的研究。资格评估由两名研究者使用Rayyan []独立进行。 数据由两名研究者使用预先试用的Microsoft Excel电子表格独立提取。为全面捕捉功能恢复情况，我们提取了关键功能结局（如WOMAC、KSS、OKS）的较基线变化评分，因为这些指标反映了随时间改善的幅度，并考虑了各研究间基线患者特征的潜在差异。与仅表明所达到最终功能水平的术后绝对评分不同，较基线变化评分对早期恢复速率和幅度的差异更为敏感，能更有意义地评估某项技术是否加速了改善和/或改变了最终功能结局。 为解决潜在的单元分析问题，对按膝报告的结果进行了仔细评估。纳入的一项随机对照试验（Song 等）涉及同期双侧手术，并报告了膝关节水平的数据。尽可能提取患者水平的数据；否则，在独立性假设下纳入膝关节水平的数据，这可能会低估方差并高估精确度。为评估该方法，本研究对所有结局进行了逐一排除敏感性分析。此外，为系统评估并防止同一随机对照试验队列的受试者在多篇文献中被重复计数（例如，随访时间不同或作者共享的研究），我们为可能存在患者重叠的研究构建了补充表格（补充表 S2–S6）。每张表格均列出相关出版物的关键信息，包括研究编号、按干预组（RA‑TKA/C‑TKA）划分的总样本量、研究机构或医院、患者入组时间、简要纳入标准以及所涉及的主要结局，以评估潜在重叠程度。在确认存在显著重叠的情况下，我们采用分层方法：对于每个独特的患者队列，在主要荟萃分析中仅保留针对某一结局随访最全面或随访时间最长的出版物。若同一试验的多篇出版物报告了不同且不重叠的结局，则从各相关出版物中提取数据，并仔细确保在合并分析中不存在患者重叠。 该策略确保了没有潜在患者重叠的研究被合并到同一分析中，从而保持了观察的独立性和合并估计值的完整性。 质量评价 两名评审员使用Cochrane偏倚风险2（RoB-2）框架 [] 独立评估所选试验的内部有效性和方法学质量。评估针对五个主要领域：随机化过程、偏离既定方案的执行、结局数据缺失、终点指标测量的客观性以及选择性报告行为。每个领域随后被分类为低风险、存在一定疑虑或高偏倚风险，任何编码员之间的分歧由第三位作者通过协商解决。所选关键结局的证据确定性采用推荐分级的评估、制定与评价（GRADE）指南 [] 进行评估。 统计分析 Meta分析采用随机效应模型，使用限制最大似然（REML）估计量，通过R软件（版本4.5.2）进行，95%置信区间采用Hartung-Knapp校正计算。连续结局的效应量以均数差（MDs）合并。异质性采用Cochrane’s Q和I²统计量评估。当检测到异质性时，进行逐项剔除敏感性分析以检查其他结果受影响的程度。此外，对所有包含两项以上研究的结局（包括异质性较高的结局）均进行了逐项剔除敏感性分析，以评估合并估计值的稳健性并识别潜在离群研究。此外，使用漏斗图评估发表偏倚，对于包含10项及以上研究的结局，进行Egger回归检验。影响力分析采用Baujat图识别对异质性有贡献且显著影响合并效应量的研究。所有统计检验均为双侧检验，p值<0.05被认为具有统计学意义。为探讨不同机器人平台的影响，我们对每个机器人系统至少有两项研究数据的结局进行了亚组分析，以处理潜在异质性并提供对结果更深入的理解。为明确异质性来源，对三个关键结局进行了预先设定的随机效应Meta回归分析：WOMAC功能评分变化、KSS功能评分变化。 以及手术时间。我们选择这些结局指标，是因为它们具有临床重要性、观察到的异质性以及足够的纳入研究数量。对每个结局指标，我们考察了四个协变量：平均年龄（岁）、平均体重指数（kg/m²）、女性参与者比例（%）以及随访时间（月）。对于自基线变化的结局指标，当变化的标准差未直接报告时，我们使用Meta分析加速器工具[]。通过提供干预前后的均值和标准差以及相关系数，该工具可计算平均变化及变化的标准差。 采用试验序贯分析（TSA）控制随机误差，并评估手术时间证据的稳健性，使用TSA软件（丹麦哥本哈根试验中心）进行分析。采用双侧I类错误（α）为5%和统计效能（1-β）为80% [–]。采用O'Brien-Fleming α消耗函数构建试验序贯监测界值。所需信息量（RIS）使用荟萃分析中的均数差（MD）和基于模型方差的方法进行估计，以校正异质性。为得出更为保守和可靠的方差估计，我们排除了被判定为高偏倚风险的研究，并直接从其余试验中计算合并方差，从而减少潜在有缺陷的研究对RIS估计的影响。将累积Z曲线与RIS进行绘制，如果曲线在达到RIS之前越过序贯监测界值，则认为证据是确定性的。 结果 检索与筛选 文献检索于2025年12月27日完成，在四个电子数据库中共识别出5，397条记录（PubMed：1，580条；Scopus：1，906条；Web of Science：1，688条；Cochrane Library：223条）。去重后，剩余4，152条记录进行初步筛选。此外，从Google Scholar（第1–50页）识别出1000条记录进行标题和摘要筛选。经过标题和摘要筛选后，共选出44篇文章进行全文筛选。经全文评估后，20篇文章因以下原因被排除：病例报告（n = 2）、非比较研究（n = 6）、人群不相关（n = 3）、非随机研究设计（n = 9）。其余24项研究符合纳入标准，并被纳入定量分析。（图）。 PRISMA筛选流程图。 纳入研究摘要。 共24项随机对照试验符合纳入标准，包含总计3，425名患者的汇总队列。纳入的研究显示出广泛且多样化的地理分布。大量研究源自亚洲，具体包括中国、韩国、泰国和新加坡。欧洲研究中心也做出了重要贡献，提供了来自英国、比利时、俄罗斯、波兰和挪威的研究。这种广泛的地域范围增强了本综述结论在不同医疗体系和患者群体中的普适性。发表时间跨度从2011年到2025年，涵盖了该领域超过十年的技术和临床演进。样本量存在相当大的异质性，范围从20名患者的小型初步队列到纳入超过1300名患者的具有里程碑意义的大规模试验。手术的主要原因是为膝关节骨关节炎（OA）。每项研究的人群及关键特征的详细总结已在表中严谨呈现。 纳入研究的汇总及基线特征 缩写：C-TKA：传统全膝关节置换术；RA-TKA：机器人辅助全膝关节置换术；NR：未报告 介入特征揭示了现代全膝关节置换术（TKA）实践的多样化格局。机器人辅助全膝关节置换术（RA-TKA）使用了多种机器人系统，包括触觉引导系统（如MAKO）、无影像手持系统（如NAVIO）以及主动自主平台（如ROBODOC）。相比之下，传统全膝关节置换术（C-TKA）手术流程始终使用手动器械，配合股骨髓内定位器和胫骨髓外定位器。此外，在假体选择（如后稳定型 vs. 后交叉韧带保留型）、麻醉方案（全身麻醉 vs. 区域麻醉）以及止血带应用策略（持续使用、选择性使用或完全不使用）方面也观察到进一步的异质性。手术技术和围手术期管理中的这种变异性为解读研究结果提供了必要的背景信息。 偏倚风险评估 为了评估方法的严谨性，我们使用了Cochrane偏倚风险-2（RoB-2）工具。共评估了二十四项研究，结果显示方法学质量参差不齐。在24项研究中，四项被判定为总体偏倚风险低，十三项存在一些问题，七项被发现总体偏倚风险高。这些被评为低风险的研究，明确报告了随机化方法、适当的分配隐藏、充分的盲法和完整的数据报告，从而提供了较高的内部有效性。相比之下，七项研究被评定为总体偏倚风险高。这些试验中常见的偏倚来源包括随机化不充分、缺乏盲法、选择性结果报告以及数据处理不完整。其余十三项研究被判定为存在一些问题。这些问题主要源于随机化程序报告不完整、盲法不明确或部分结局数据，从而引入了中等程度的不确定性。（图） 纳入研究偏倚风险评估（ROB） 证据确定性评估 采用GRADE框架评估的各结局的证据确定性汇总于表。关于WOMAC在3个月和1年时变化以及OKS在1年时变化的证据。
全文（开放获取） · 英文原文
【Abstract】 Although robotic-assisted total knee arthroplasty (RA-TKA) is increasingly adopted for its potential to improve surgical precision, its impact on functional outcomes compared with conventional TKA (C-TKA) remains unclear. Existing syntheses are limited by pooling observational and randomized data, focusing on absolute postoperative scores without evaluating recovery trajectories, and lacking rigorous exploration of heterogeneity. To overcome these limitations, we conducted a methodologically rigorous meta-analysis of randomized controlled trials (RCTs) only, incorporating comprehensive meta-regression, platform-specific subgroup analyses, and assessment of evidence certainty using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework. PubMed, Web of Science, Scopus, the Cochrane Library, and Google Scholar (pages 1-50) were comprehensively searched from their inception up to December 27, 2025. Data were pooled using random-effects meta-analysis and Hartung-Knapp adjustments. Meta-regression of different moderators and subgroup analyses by robotic platforms were performed, when feasible, to identify sources of heterogeneity. A total of 24 RCTs encompassing 3,425 patients were included. The pooled analyses demonstrated no significant differences between RA-TKA and C-TKA for change in Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) function score (MD 0.14; 95% CI - 5.88 to 6.16; p = 0.96), change in functional Knee Society Score (KSS) (MD 0.51; 95% CI - 3.43 to 4.44; p = 0.78), or change in Oxford Knee Score (OKS) (MD 2.07; 95% CI - 6.82 to 10.96; p = 0.51). Although a subgroup analysis revealed a significant improvement in OKS at 6 months only (MD 2.25; p = 0.01), this effect did not reach the minimal clinically important difference (MCID) and was not sustained at other time points. No significant differences were observed for range of motion, change in VAS pain score, KSS satisfaction, or length of hospital stay. Operative time was significantly longer for RA-TKA (MD 22.34 min; 95% CI 13.66 to 31.02; p < 0.01), though this estimate should be interpreted with caution given the very high heterogeneity (I² = 98.53%) and the likelihood that it includes cases performed during the early learning curve of robotic adoption. Trial Sequential Analysis suggested that this finding is unlikely to be attributable to random error, though the required information size was not reached. RA-TKA and C-TKA provide comparable functional improvement, pain relief, and most perioperative outcomes. The most consistent signal is a prolonged operative time with RA-TKA favoring conventional surgery, though the precise magnitude remains uncertain and may be influenced by the learning curve. The certainty of evidence is low to very low across all assessed outcomes, highlighting the need for cautious interpretation. Current evidence does not support the routine adoption of RA-TKA over C-TKA for TKA. Future adequately powered, multicenter RCTs with long-term follow-up and consistent reporting of surgeon experience are required to definitively clarify the role of RA-TKA in contemporary practice. 【Introduction】 Over one million total knee arthroplasties (TKA) are executed annually across the United States, with current projections estimating this volume to surpass three million cases by the year 2040 []. The literature supports its effectiveness; however, conventional TKA (C-TKA) still faces challenges with optimal component alignment and soft-tissue balancing, as indicated by the 20% of patients who report dissatisfaction after the procedure []. These technical challenges have prompted the development of robotic-assisted TKA (RA-TKA) that combines 3D imaging, computer navigation, and robotic assistance to facilitate alignment and potentially improve patient satisfaction [, ]. The underlying hypothesis of RA-TKA is that more accurate technical alignment of the components will result in superior patient-centered outcomes []. However, the clinical significance of this enhanced surgical precision remains controversial, with studies showing inconsistent results between radiographic improvements and functional outcomes [, ]. This disconnect illustrates the importance of focusing on patient-centered outcomes, rather than radiographic outcomes alone. Consequently, post-surgical functional recovery serves as a primary benchmark for determining procedural success. This recovery is routinely quantified using validated, standardized evaluation tools, most notably KSS, OKS, and WOMAC. These measures assess patients’ ability to execute fundamental activities of daily life, general function, and joint range of motion, all of which reflect physical independence and quality of life []. The nature and extent of functional recovery after TKA are key variables in the rehabilitation process, the level of postoperative care provided, and the patient’s long-term well-being. Postoperative pain, using a visual analog scale (VAS), constitutes another key metric and can significantly impact recovery and patient satisfaction []. Pain that is inadequately managed or uncontrolled after surgery may delay functional rehabilitation, delay the return to usual activities, and be a source of patient dissatisfaction and revision surgery [, ]. The relationship between surgical technique and pain reporting has been demonstrated but remains incompletely understood. However, enhanced accuracy may theoretically optimize joint biomechanics and reduce aberrant stress distribution. Patient satisfaction is the paramount patient-centered endpoint that encompasses several domains, including pain relief, improvement in function, realization of pre-operative expectations, and overall perception of surgical success. Given that one-fifth of TKA patients are dissatisfied despite arguably objective radiographic success, determining strategies to maximize this critical endpoint during surgery is a research priority [, ]. Despite the growing number of meta-analyses which compare the two techniques in arthroplasty practices, several critical gaps remain [–]. First, many prior syntheses have pooled randomized and observational studies, potentially introducing confounding and obscuring true treatment effects derived from high-level evidence. However, this study is restricted to randomized controlled trials (RCTs), thereby ensuring more robust causal inference and minimizing the confounding inherent in observational data. Second, while previous meta-analyses have largely focused on absolute postoperative outcomes, such measures do not adequately capture the magnitude of patient improvement. In contrast, our study uniquely concentrates on change‑from‑baseline scores, which more accurately reflect the degree of recovery and account for potential differences in baseline patient characteristics across studies. This distinction is clinically meaningful, as two patients with identical postoperative scores may have experienced vastly different magnitudes of improvement depending on their preoperative status. Third, we carried out comprehensive meta-regression analyses to discover the cause of heterogeneity across multiple clinically relevant covariates, an approach that has been limited or absent in prior work. This allows for a more rigorous exploration of potential effect modifiers and sources of between-study variability. Fourth, we conducted subgroup analyses by specific robotic platform, when feasible, allowing identification of potential system-specific effects rather than treating robotic-assisted TKA as a homogeneous intervention. Fifth, we performed Trial Sequential Analysis (TSA) for operative time, an outcome characterized by very high heterogeneity, to assess whether the observed difference is robust against random error, thereby providing a more definitive evaluation of this important outcome. Finally, we applied the GRADE framework to provide a transparent and clinically interpretable evaluation of the strength of the findings. Collectively, these methodological innovations provide a more rigorous and clinically meaningful synthesis of the current evidence base and address key limitations of previous meta-analyses. 【Methods】 【Search strategy】 This review strictly adhered to the PRISMA 2020 recommendations [] and was registered on PROSPERO (CRD420261301318). Two authors independently carried out a thorough search of five major biomedical databases, covering all eligible publications available through December 27, 2025: PubMed, Web of Science, Scopus, the Cochrane Library, and Google Scholar (pages 1–50). In addition, we supplemented our search by hand-searching the reference lists of included studies and relevant reviews to search for any additional eligible records not captured by the electronic database searches. The detailed search strategy is available in supplementary material as Supplementary Table 1. 【Inclusion and exclusion criteria】 This review was limited to RCTs to avoid the confounding present in observational data and to give a rigorous causal assessment of the treatment impact. The population included adult patients scheduled to undergo total knee arthroplasty for any indication. The intervention was robotic-assisted TKA, with any available robotic system. Comparators were conventional TKA, performed using manual or conventional techniques without robotic assistance. To warrant inclusion, studies were required to have data on one or more of these clinical endpoints: Pain and Function Measures: Postoperative pain severity via change in VAS; standardized functional metrics, including changes in OKS, KSS, and WOMAC function subscales. Surgical and Perioperative Outcomes: Joint mobility evaluated through range of motion (ROM); length of stay (LOS); operative time (minutes); estimated intraoperative or cumulative blood loss (mL); post-surgical wound drainage volume (mL); perioperative hemoglobin fluctuations; systemic inflammatory indicators; and validated patient satisfaction scores. Exclusion criteria were non-RCTs, case series, study protocols, or abstracts; studies not comparing RA-TKA to C-TKA or lacking a control group; studies not reporting relevant clinical outcomes. Eligibility assessment was carried out independently by two investigators using Rayyan []. 【Data extraction】 Independently, data were extracted by two investigators by a pre‑piloted Microsoft Excel spreadsheet. To comprehensively capture functional recovery, we extracted change‑from‑baseline scores for key functional outcomes (e.g., WOMAC, KSS, OKS), as these measures reflect the magnitude of improvement over time and account for potential differences in baseline patient characteristics across studies. Unlike absolute postoperative scores, which merely indicate the final functional level achieved, change‑from‑baseline scores are more sensitive to differences in the rate and magnitude of early recovery, providing a more meaningful assessment of whether a technique accelerates improvement and/or alters the ultimate functional outcome. To address potential unit-of-analysis issues, outcomes reported per knee were carefully evaluated. One included RCT (Song et al., ) involved simultaneous bilateral procedures and reported data at the knee level. Patient-level data were extracted when available; otherwise, knee-level data were included under an assumption of independence, which may underestimate variance and overestimate precision. To evaluate this approach, leave-one-out sensitivity analyses were undertaken for all outcomes addressed in this study. Additionally, to systematically assess and prevent double‑counting of participants from the same RCT cohort across multiple publications (e.g., studies with different follow‑up durations or shared authorship), we constructed supplementary tables (Supplementary Tables S2–S6) for studies with potential patient overlap. Each table presents key information for related publications, including study ID, total sample size with breakdown by intervention group (RA‑TKA/C‑TKA), setting or hospital, patient recruitment period, brief inclusion criteria, and key outcomes addressed, to evaluate the extent of potential overlap. Where substantial overlap was identified, we applied a hierarchical approach: for each unique patient cohort, we retained only the publication with the most comprehensive or most extended follow‑up for a given outcome in the primary meta‑analysis. In cases where multiple publications from the same trial reported distinct, non‑overlapping outcomes, data were extracted from each relevant publication, carefully ensuring no patient overlap in the pooled analysis. This strategy ensured that no studies with potential patient overlap were pooled together in the same analysis, thereby maintaining the independence of observations and the integrity of the pooled estimates. 【Quality assessment】 The internal validity and methodological quality of the selected trials were appraised independently by two reviewers using the Cochrane Risk of Bias 2 (RoB-2) framework []. Evaluations targeted five primary domains: the randomization process, deviations from assigned protocols, missing outcome data, objective measurement of endpoints, and selective reporting behavior. Each domain was subsequently classified as presenting low risk, some concerns, or a high risk of bias, with any coder discrepancies settled by a third author via consensus. The certainty of evidence for selected critical outcomes was evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) guidelines []. 【Statistical analysis】 Meta-analyses, using random‑effects models with the Restricted Maximum‑Likelihood (REML) estimator, were carried out using R (version 4.5.2), and 95% CIs were calculated using the Hartung‑Knapp adjustment. Effect sizes for continuous outcomes were pooled as mean differences (MDs). Heterogeneity was evaluated using Cochrane’s Q and I² statistics. When heterogeneity was detected, leave‑one‑out sensitivity analyses were undertaken to check the extent to which other results would be affected. In addition, we performed leave‑one‑out sensitivity analyses for all outcomes with more than two studies, including those with high heterogeneity, to evaluate the robustness of the pooled estimates and identify potential outlier studies. Moreover, publication bias was assessed using funnel plots, and for outcomes with 10 or more studies, Egger’s regression test was performed. Influence analyses were evaluated using Baujat plots to identify studies that contributed to heterogeneity and significantly influenced the pooled effect sizes. All statistical tests were two‑tailed, and p‑values < 0.05 were considered statistically significant. To explore the impact of different robotic platforms, we performed subgroup analyses for outcomes for which data were available from at least two studies for each robotic system. This was done to address potential heterogeneity and provide a more granular understanding of the results. To understand sources of heterogeneity, pre‑specified random‑effects meta‑regressions were conducted on three key outcomes: change in WOMAC function score, change in KSS function, and operative time. We chose these outcomes for their clinical importance, observed heterogeneity, and enough contributing studies. Four covariates were examined for each outcome: mean age (years), mean BMI (kg/m²), proportion of female participants (%), and follow‑up duration (months). For change‑from‑baseline outcomes, when the standard deviation (SD) of the change was not directly reported, we used the Meta‑Analysis Accelerator tool []. By providing the means and SDs before and after the intervention, along with the correlation coefficient, the tool calculates the mean change and standard deviation of the change. A Trial Sequential Analysis (TSA) was undertaken to control for random errors and evaluate the robustness of the evidence for operative time, using TSA software (Copenhagen Trial Unit, Denmark). A two‑sided type I error (α) of 5% and a statistical power (1‑β) of 80% were applied [–]. The O’Brien‑Fleming alpha‑spending function was used to construct the trial sequential monitoring boundaries. The required information size (RIS) was estimated using the mean difference (MD) from the meta‑analysis and the model-variance-based method to correct for heterogeneity. To derive a more conservative and reliable variance estimate, we excluded studies judged to have a high risk of bias and calculated the pooled variance directly from the remaining trials, thereby reducing the influence of potentially flawed studies on the RIS estimate. The cumulative Z‑curve was plotted against the RIS, and the evidence was considered conclusive if the curve crossed the sequential monitoring boundaries before it reached the RIS. 【Results】 【Search and screening】 The literature search, completed on December 27, 2025, identified a total of 5,397 records across the four electronic databases (PubMed: 1,580; Scopus: 1,906; Web of Science: 1,688; Cochrane Library: 223). Following de‑duplication, 4,152 records remained for initial screening. In addition, 1000 records were identified from Google Scholar (pages 1–50) for title and abstract screening. After title and abstract screening, 44 articles were selected for full‑text screening. Following full‑text assessment, 20 articles were excluded for the following reasons: case reports (n = 2), not comparative studies (n = 6), irrelevant population (n = 3), non‑randomized study design (n = 9). The remaining 24 studies met the inclusion criteria and were included in the quantitative analysis. (Fig. ). PRISMA flowchart of the selection process. 【Summary of studies included】 Twenty-four RCTs met the inclusion criteria, comprising an aggregate cohort of 3,425 patients. The included research demonstrated a broad and diverse geographical distribution. A significant concentration of studies originated from Asia, specifically China, South Korea, Thailand, and Singapore. European centers also made substantial contributions, with studies from the United Kingdom, Belgium, Russia, Poland, and Norway. This broad geographical scope enhances the generalizability of the review’s conclusions on different healthcare systems and patient populations. The publication dates ranged from 2011 to 2025, capturing over a decade of technological and clinical evolution in the field. There was considerable heterogeneity in the sample sizes, ranging from small pilot cohorts of 20 patients to landmark, large-scale trials enrolling over 1300 patients. The primary cause for surgery was osteoarthritis (OA) of the knee joint. A detailed summary of each study’s population and key characteristics is meticulously presented in Table . Summary and baseline characteristics of included studies Abbreviations: C-TKA: Conventional TKA;RA-TKA: Robotic-Assisted TKA;NR: Not Reported The interventional characteristics revealed a diverse landscape of modern TKA practice. A broad spectrum of robotic systems was utilized for RA-TKA, including haptic-guided systems (e.g., MAKO), imageless handheld systems (e.g., NAVIO), and active autonomous platforms (e.g., ROBODOC). In contrast, C-TKA procedures consistently used manual instrumentation with intramedullary femoral and extramedullary tibial guides. Further heterogeneity was noted in the choice of prosthesis (e.g., posterior-stabilized vs. cruciate-retaining), anesthesia protocols (general vs. regional), and tourniquet application strategies (continuous use, selective use, or no use at all). This variability in surgical technique and perioperative management provides essential context for interpreting the findings. 【Risk of bias assessment】 To evaluate methodological rigor, we used the Cochrane Risk of Bias-2 (RoB-2) tool. A total of twenty-four studies were assessed, revealing variable methodological quality. Among the 24 studies, four were judged to have a low overall risk of bias, thirteen showed some concerns, and seven were found to have a high overall risk of bias. These trials, rated as low risk, demonstrated clearly reported randomization methods, proper allocation concealment, adequate blinding, and complete data reporting, thereby providing high internal validity. In contrast, seven studies were rated as having a high overall risk of bias. Familiar sources of bias among these trials included inadequate randomization, lack of blinding, selective outcome reporting, and incomplete data handling. The remaining thirteen studies were judged to have some concerns. These concerns primarily resulted from incomplete reporting of randomization procedures, unclear blinding, or partial outcome data, which introduced a moderate degree of uncertainty. (Fig. ) Assessment of the included studies' risk of bias (ROB) 【Assessment of evidence certainty】 The certainty of the evidence for the evaluated outcomes, as assessed using the GRADE framework, is summarized in Table . The evidence for change in WOMAC at 3‑months and 1‑year, as well as change in OKS at 1‑year