SGLT2抑制剂、GLP-1受体激动剂和吡格列酮在亚洲研究环境中对2型糖尿病和MASLD的增量效应：一项系统评价和荟萃分析。
全文（开放获取） · 中文译文
摘要 目的：我们在患有2型糖尿病（T2DM）和代谢功能障碍相关脂肪性肝病（MASLD）的亚洲成人中，评估了在可比较的背景降糖治疗基础上，加用钠-葡萄糖协同转运蛋白2（SGLT2）抑制剂、胰高血糖素样肽-1受体激动剂（GLP-1RA）或吡格列酮所带来的额外肝脏和代谢效应。 方法：我们系统检索了PubMed、Embase、Cochrane图书馆和Web of Science，检索时间截至2026年6月30日。两名评审员独立筛选研究、提取数据并评估偏倚风险。符合条件的试验评估了在随机分组之间具有可比性且要么保持稳定、要么遵循预设调整方案的背景降糖管理基础上加用研究药物。丙氨酸氨基转移酶（ALT）和糖化血红蛋白（HbA1c）自基线的变化是共同主要结局。次要结局包括天冬氨酸氨基转移酶、γ-谷氨酰转移酶、磁共振成像-质子密度脂肪分数、肝脏硬度、纤维化-4指数、空腹血糖、血脂参数、体重指数、血清脂联素和收缩压。进行了随机效应成对荟萃分析，并对两个共同主要结局进行了GRADE确定性评估。由于安全性结局的报告存在异质性且经常不完整，因此对其进行了叙述性综合。 结果：我们纳入了11项随机试验：692名受试者被随机分配至符合条件的比较组，其中642名参与了共同主要疗效分析。与单用背景治疗相比，加用SGLT2抑制剂降低ALT具有中等确定性（平均差[MD]，-7.56 U/L；95%置信区间[CI]，-11.24至-3.88）；然而，对于HbA1c，观察到很少或没有临床重要的进一步获益（MD，-0.09%；95% CI，-0.27至0.08；中等确定性）。中等确定性提示GLP-1RA可适度降低ALT（MD，-6.71 U/L；95% CI，-11.53至-1.89），而低确定性提示HbA1c降低（MD，-0.48%；95% CI，-0.83至-0.14）。存在支持加用治疗的次要探索性发现，尽管这些结果未针对多重性进行调整。比较安全性无法可靠评估。 结论：在亚洲研究环境中进行的短期随机试验中，加用SGLT2抑制剂或GLP-1RA可能改善选定的肝脏和代谢替代结局。现有证据不支持治疗类别排序，且由于随访时间短且若干比较基于稀疏证据，这些变化的持久性和临床意义仍不确定。比较安全性也仍不确定，因为不良事件报告异质且不完整。 系统综述注册：https://www.crd.york.ac.uk/PROSPERO/，标识符CRD420251231137。 引言 代谢功能障碍相关脂肪性肝病（MASLD；既往称为非酒精性脂肪性肝病）与2型糖尿病（T2DM）在病理生理学上具有相似之处，例如胰岛素抵抗、异位脂质蓄积、氧化应激和慢性低度炎症（）。2021年，全球约有5.37亿糖尿病患者，预计到2045年这一数字将增至7.83亿（）。据估计，约有三分之二的T2DM患者合并MASLD（）；因此，针对血糖控制和肝脏相关结局的治疗方法备受关注（）。 MASLD的临床表型在不同人群中存在差异。与西方人群相比，一些亚洲人群可能在较低的身体质量指数（BMI）下即发生肝脏脂肪变性和代谢异常，这种表型通常被称为“瘦人MASLD”（–）。内脏脂肪堆积、胰腺胰岛素分泌能力以及对异位脂质沉积的易感性等方面的差异，可能导致疾病表现和进展的差异。钠-葡萄糖协同转运蛋白2（SGLT2）抑制剂和胰高血糖素样肽-1受体激动剂（GLP-1RAs）在合并MASLD和T2DM的患者中显示出良好的肝脏效应（）；然而，现有证据大多来自非亚洲人群或种族异质性队列，这限制了其在亚洲研究环境中的适用性。 目前的降糖策略常被升级或调整，以满足血糖之外的目标，包括体重、心血管、肾脏和肝脏问题。因此，SGLT2抑制剂、GLP-1受体激动剂，有时还有吡格列酮，被用于现有治疗之上（而非仅在基线管理全部失败时才作为补救措施）(–)。在此背景下，附加疗法尤其适用于伴有代谢功能障碍相关脂肪性肝病（MASLD）的2型糖尿病（T2DM）患者，我们需明确在继续其他适当的背景治疗的同时，加用某一特定药物能带来何种额外获益。背景降糖管理在各随机分组之间必须保持一致，且在试验期间保持不变或遵循预设的调整方案——否则，改变伴随治疗可能混淆与随机化策略相关的结果。因此，我们具有临床意义的目标估计量是，所加药物/治疗策略与可比较的背景管理相比的增量效应，而非其对异质性“常规治疗”方案的疗效。 然而，在相似背景治疗基础上增加的叠加效应缺乏明确定义。在试验和荟萃分析中，入选标准、疾病定义、基线代谢特征、治疗时长、结局指标以及比较设计均存在显著差异（–）；许多证据综合将安慰剂、生活方式干预、常规治疗和主动降糖对照合并分析，同时未平衡随机分组间的背景治疗（–），这使得将特定的肝脏和代谢效应归因于所添加的药物颇具挑战性。其普适性尚不清楚，因为大多数综述涉及地理和种族异质人群，且未对亚洲研究环境进行单独评估。直接比较（尤其是涉及SGLT2抑制剂与GLP-1受体激动剂的比较）很少；因此，任何类别差异主要来自间接证据，而非确立比较优效性（–）。 因此，我们对在亚洲国家开展的随机对照试验进行了系统评价和成对荟萃分析，这些试验评估了在成人2型糖尿病（T2DM）合并代谢相关脂肪性肝病（MASLD）患者中，在可比背景降糖管理基础上加用SGLT2抑制剂、GLP-1受体激动剂或吡格列酮的增量肝脏和代谢效应。我们检查了肝酶、肝脏脂肪变性和纤维化标志物、血糖控制、血脂参数以及人体测量学结局。 方法 原始方案已在国际前瞻性系统评价注册系统（PROSPERO）中注册（CRD420251231137），拟对三种药物治疗T2DM合并非酒精性脂肪性肝病患者进行网络荟萃分析。采用设计-治疗交互作用模型评估全局不一致性，采用节点拆分法评估局部不一致性。p值<0.05被认为表明存在统计学显著的不一致性。在直接与间接证据之间发现统计学显著的不一致性（p<0.05）后，原计划的网络荟萃分析未继续进行。因此，分析改为基于直接比较的常规成对荟萃分析。本分析的方案已在PROSPERO中回溯性注册（CRD420261457952）。 检索策略 我们检索了PubMed、Embase、Cochrane图书馆和Web of Science，检索时间截至2026年6月30日，查找SGLT2抑制剂、GLP-1受体激动剂或吡格列酮在T2DM合并MASLD患者中的随机对照试验。检索结合了医学主题词（MeSH）和自由文本词汇，涵盖T2DM、MASLD或非酒精性脂肪性肝病、三类治疗药物以及随机试验，并使用布尔运算符。完整检索策略见附录。两名 reviewers 独立筛选标题、摘要和全文。分歧通过讨论和共识解决。 入选与资格标准 研究符合入选标准需满足以下全部条件：（1）受试者为亚洲研究环境中的成年T2DM合并MASLD或非酒精性脂肪性肝病患者；（2）研究为前瞻性随机对照试验，随访时间至少12周；（3）随机各组在基线时的背景降糖管理具有可比性，且在随访期间保持不变或按照预设方案进行管理；（4）随机对照评估的是在可比背景管理基础上加用SGLT2抑制剂、GLP-1RA、吡格列酮或活性对照药的效果，伴随治疗方案的任何规定变更均被记录，而非作为计划外联合干预处理；（5）至少报告了一项预设的主要或次要结局。 我们排除了以下研究：非随机研究、入组人群为非亚洲人群的研究、随机各组间背景治疗不可比的研究、允许计划外更改背景药物的研究、使用可能影响结局的伴随降糖和/或保肝治疗的研究、报告重复/重叠数据集的研究，以及结局数据不充分的研究。 质量评估 两位评审员使用Cochrane偏倚风险2（RoB 2）工具独立评估偏倚风险。由于RoB 2的判断是针对特定结局的，因此对构成共同主要结局的丙氨酸氨基转移酶（ALT）和糖化血红蛋白（HbA1c）相对于基线的变化，分别对每项随机化比较进行了评估。目标估计量为分配到干预组的效果。评估涵盖了RoB 2的五个标准领域，并依据已发表的报告以及在可获得时的研究方案、统计分析计划和试验注册信息进行。根据RoB 2的信号问题和决策算法，领域层面及总体判断被分类为低偏倚风险、存在一定问题或高偏倚风险。 我们并未仅因开放标签设计而认定其表明偏倚风险增加。相反，我们评估了知晓治疗分组是否可能合理影响对既定干预的偏离或结局测量。关于缺失结局数据和选择性报告，ALT和HbA1c分别进行了评估。分歧通过讨论和共识解决。完整的信号问题应答、领域层面判断以及各研究特定的理由说明在附录中提供。 数据提取 两名独立评审员使用标准的Microsoft Excel表格提取研究设计、参与者人口学特征、诊断标准、基线临床特征、干预与对照方案、各组背景降糖治疗（药物或类别、明确报告的剂量、随机化前稳定性要求及方案规定的调整规则）、随访时长和预设结局。分歧经讨论并通过共识解决。安全性数据包括不良事件、严重/重度不良事件、因不良事件停药以及其他退出原因。当注意到未报告各组背景药物/剂量时，我们将其记录为未报告（NR），而非根据现有信息进行推断。详细的分组研究药物方案和背景治疗管理可见于；总结了研究设计、随访、入选标准、参与者人群以及对我们荟萃分析有贡献的结局。我们分别提取了随机化参与者、明确报告的治疗/安全性人群、可获得时完成随访者以及按结局分析者的人群计数。若因原始研究未单独报告而无法推断治疗分母，则相应予以反映。 连续结局数据优先以均数和标准差（SD）的形式提取。当研究报告的是中位数及范围（range）和/或四分位数间距（interquartile range），而非均数和SD时，则采用经验证的转换方法估算相应的均数和SD（–）。对于自基线变化（change-from-baseline）的结局，若未报告变化分数的SD，但可获得基线和治疗后SD，则根据Cochrane手册计算变化分数的SD：SD_change = √(SD_baseline² + SD_post² − 2r × SD_baseline × SD_post)，其中r表示基线与治疗后测量值之间的受试者内相关性（）。这些计算假设受试者内相关系数r = 0.8。当结局数据以标准误（SE）报告时，则使用SD = SE × √n（）将其转换为SD。 共同主要终点是丙氨酸氨基转移酶（ALT）（主要肝脏生化终点）和糖化血红蛋白（HbA1c）（主要血糖终点）自基线的变化。次要肝脏终点为天冬氨酸氨基转移酶（AST）、γ-谷氨酰转移酶、磁共振成像质子密度脂肪分数（MRI-PDFF）、肝脏硬度测量值和纤维化-4指数。次要代谢终点为空腹血糖、甘油三酯、高密度脂蛋白胆固醇、低密度脂蛋白胆固醇、BMI、血清脂联素和收缩压。 数据合成与分析 统计分析使用R版本4.0.3中的meta包和Stata/MP版本17进行。连续结局以均数差（MD）及95%置信区间（CI）汇总。配对荟萃分析采用逆方差随机效应模型，研究间方差（τ2）使用DerSimonian–Laird方法估计。当τ2可估计时，我们报告合并分析的τ2；对于单项研究比较，则未报告τ2。 对于多臂试验，仅纳入与预设分析相关的配对比较，且同一荟萃分析中任何参与者组均未被重复纳入，从而避免共享组的重复计数。每个森林图中显示的比较类别代表预设的配对治疗对比，而非临床或方法学子组。未进行临床或方法学子组分析，也未进行亚组效应的正式交互作用检验，软件生成的各显示比较类别间差异的检验未被用于推断性目的。 统计学异质性采用Cochran’s Q检验和I²统计量进行评估（分别以p < 0.10和I² ≥ 50%为界）。森林图标示获益方向：对于数值较低为有利的结局指标，如ALT、AST、MRI-PDFF、肝脏硬度、HbA1c、空腹血糖、甘油三酯、低密度脂蛋白胆固醇、收缩压和体重，无效线左侧的数值支持试验组或加用治疗组。对于数值较高为有利的结局指标，如高密度脂蛋白胆固醇和脂联素，无效线右侧的数值支持试验组或加用治疗组。 采用逐项剔除分析评估单项研究对合并估计值的影响。对所有结局指标进行敏感性分析，以评估合并估计值的稳健性。发表偏倚的评估计划仅在纳入研究数量足够时通过漏斗图目视检查进行。由于大多数合并分析纳入的研究少于10项，漏斗图仅视为探索性分析，不将其解释为排除发表偏倚的证据。 对两项共同主要结局的分析予以优先处理。次要结局视为探索性分析；报告的名义p值未进行多重性正式校正，解读时需谨慎。 由于不良事件的定义和报告格式存在异质性，且分臂数据常无法获得，安全性结局采用叙述性综合而非定量合并的方式进行汇总。 证据的确定性 对于每个预设的成对比较中两个共同主要结局（ALT和HbA1c）的证据确定性，使用推荐分级评估、制定与评价（GRADE）方法进行评估。随机证据起始为高确定性，并在适当时因偏倚风险、不一致性、间接性、不精确性和发表偏倚而进行降级。对于偏倚风险领域，GRADE判断基于相应共同主要结局的最终结局特异性RoB 2评估。次要结局为探索性结局，未进行正式分级。不精确性主要通过将95%置信区间与预设的临床重要阈值进行比较，并评估最优信息样本量（OIS）来评估，而非仅依据样本量或置信区间是否跨越无效线。对于HbA1c，采用0.5个百分点的实用性组间阈值。对于ALT，根据美国肝病研究学会（AASLD）指南，采用17 U/L的降低作为情境性组织学相关应答阈值。由于该值并非经过验证的组间最小重要差异，且可能具有机制特异性，故仅用于定义GRADE不精确性边界，而不用于推断组织学获益。OIS按两组比较计算，采用双侧α = 0.05、80%把握度、相应阈值以及来自参与试验臂的评分变化值的合并组内标准差。确定性评为高、中、低或极低。关于SGLT2抑制剂降低ALT和HbA1c的详细GRADE证据概况及降级理由见附录。 与背景治疗相比，GLP-1受体激动剂与背景治疗相比，SGLT2抑制剂与噻唑烷二酮类相比，以及GLP-1受体激动剂与噻唑烷二酮类相比，分别。 结果 纳入研究的描述 我们识别出327条记录。经过去重以及依次进行的标题、摘要和全文筛选，11项随机对照试验（RCT）被纳入定性和定量综合（）。ALT和HbA1c为共同主要结局；所有其他终点均作为探索性次要结局进行分析。 研究识别与筛选的PRISMA流程图。 在每个比较组中，共有692名参与者被随机分配，其中642名参与了共同主要疗效分析。在所有纳入的试验中，参与者年龄范围为46.6至62.3岁。总数差异反映了原始试验中报告的试验特异性失访和分析人群；若明确报告了治疗/安全人群则予以记录，否则编码为NR。干预措施包括SGLT2抑制剂、GLP-1受体激动剂和吡格列酮。比较组包括继续匹配的背景治疗、匹配背景治疗基础上的安慰剂，以及包括吡格列酮在内的活性对照药。各试验的背景降糖方案有所不同。各试验药物组剂量、背景治疗详情以及方案规定的稳定性/调整规则见附录；背景药物剂量在明确可获得时予以报告，否则显示为NR。符合条件的试验要么维持背景治疗不变，要么采用预设的管理规则，使我们能够将这些对比解释为增量随机治疗策略与计划外合并用药效果之间的比较。中位随访时间为24周（范围：12–28周）；试验在中国（n = 3）、日本（n = 4）、印度（n = 2）、伊朗（n = 1）和泰国（n = 1）进行。关于研究和参与者的详细信息见附录和附录。 纳入的随机对照试验的研究特征、参与者人群、治疗方案和随访期。 R，随机化人群；T，治疗/安全性人群；A，分析人群；ITT，意向性治疗；NR，未报告。给出了符合条件的比较组数量。“*”表示在 Guo 等人的研究中，共有三个组：此处呈现的随机化人数对应于进入预设两两比较的利拉鲁肽组和安慰剂组。“†”表示在 Horibe 等人的研究中，共有 52 名随机化参与者；由于未提供组水平的随机化分母，分析人群被用作疗效分析的分母，该分母可能与随机化人群不同。各组的背景治疗详情、报告的背景药物剂量以及方案规定的稳定性/调整规则总结于；未报告剂量未作推断。
全文（开放获取） · 英文原文
【Abstract】 OBJECTIVE: We evaluated the incremental hepatic and metabolic effects of adding sodium-glucose cotransporter 2 (SGLT2) inhibitors, glucagon-like peptide-1 receptor agonists (GLP-1RAs), or pioglitazone to comparable background glucose-lowering therapy in Asian adults with type 2 diabetes mellitus (T2DM) and metabolic dysfunction-associated steatotic liver disease (MASLD). METHODS: We systematically searched PubMed, Embase, the Cochrane Library, and Web of Science through 30 June 2026. Two reviewers independently screened studies, extracted data, and assessed risk of bias. Eligible trials evaluated the addition of a study drug to background glucose-lowering management that was comparable between randomized groups and either remained stable or followed a prespecified adjustment protocol. Changes from baseline in alanine aminotransferase (ALT) and glycated hemoglobin (HbA1c) were coprimary outcomes. Secondary outcomes included aspartate aminotransferase, gamma-glutamyl transferase, magnetic resonance imaging-proton density fat fraction, liver stiffness, fibrosis-4 index, fasting plasma glucose, lipid parameters, body mass index, serum adiponectin, and systolic blood pressure. Random-effects pairwise meta-analyses were performed, and GRADE certainty was assessed for the two coprimary outcomes. Safety outcomes were synthesized narratively because reporting was heterogeneous and frequently incomplete. RESULTS: We included eleven randomized trials: 692 participants were randomized into eligible comparisons, of which 642 participated in the coprimary efficacy analyses. Moderate certainty was found for a reduction in ALT with the addition of an SGLT2 inhibitor compared with background therapy alone (mean difference [MD], - 7.56 U/L; 95% confidence interval [CI], - 11.24 to - 3.88); however, there was little or no clinically important further benefit observed for HbA1c (MD, - 0.09%; 95% CI, - 0.27 to 0.08; moderate certainty). Modest reductions in ALT were suggested by moderate certainty for GLP-1RAs (MD, - 6.71 U/L; 95% CI, - 11.53 to - 1.89) while low certainty indicated an HbA1c reduction (MD, - 0.48%; 95% CI, - 0.83 to - 0.14). Secondary exploratory findings favoring add-on therapy existed, although they are unadjusted for multiplicity. The comparative safety could not be assessed reliably. CONCLUSION: In short-term randomized trials conducted in Asian study settings, adding SGLT2 inhibitors or GLP-1RAs may improve selected hepatic and metabolic surrogate outcomes. The available evidence does not support a treatment-class ranking, and the durability and clinical significance of these changes remain uncertain because follow-up was short and several comparisons were based on sparse evidence. Comparative safety also remains uncertain because adverse-event reporting was heterogeneous and incomplete. SYSTEMATIC REVIEW REGISTRATION: https://www.crd.york.ac.uk/PROSPERO/, identifier CRD420251231137. 【Introduction】 Metabolic dysfunction-associated steatotic liver disease (MASLD; previously known as nonalcoholic fatty liver disease) shares pathophysiologic similarities such as insulin resistance, ectopic lipid accumulation, oxidative stress, and chronic low-grade inflammation with type 2 diabetes mellitus (T2DM) (). There were an estimated 537 million people worldwide who lived with diabetes in 2021, a number that is projected to increase to 783 million by 2045 (). Approximately two-thirds of those with T2DM are estimated to have MASLD (); therefore, there is great interest in therapies targeting both glycemic control and liver-related outcomes (). Clinical phenotypes of MASLD vary across populations. Compared with Western populations, some Asian populations may develop hepatic steatosis and metabolic abnormalities at lower body mass index (BMI), a phenotype often described as “lean MASLD” (–). Differences in visceral adiposity, pancreatic insulin secretory capacity, and susceptibility to ectopic lipid deposition may contribute to differences in disease presentation and progression. Sodium–glucose cotransporter 2 (SGLT2) inhibitors and glucagon-like peptide-1 receptor agonists (GLP-1RAs) have shown favorable hepatic effects in patients with coexisting MASLD and T2DM (); however, much of the available evidence comes from non-Asian or ethnically heterogeneous cohorts, limiting its generalizability to Asian study settings. Today’s glucose-lowering strategies are often escalated/reformatted to meet goals beyond glycemia, including weight, cardiovascular, renal, and liver concerns. As such, SGLT2 inhibitors, GLP-1RAs, and sometimes pioglitazone are used atop existing therapy (rather than only as rescue when all baseline management fails) (–). In this context, an add-on approach applies particularly well to patients with T2DM and MASLD where we ask what additional benefit there is from adding one particular agent while continuing other appropriate background therapy. Background glucose-lowering management must be similar across randomized groups and either unchanged or subject to a prespecified adjustment protocol during trials—otherwise, changing concomitant therapies could confound results associated with our randomized strategy. Our clinically meaningful estimand is thus the incremental effect of the added drug/treatment strategy compared to otherwise comparable background management rather than its efficacy against a heterogeneous “usual-care” regimen. However, incremental add-on effects with similar background therapy are poorly defined. Eligibility criteria, disease definition, baseline metabolic characteristics, length of treatment, outcome measures, and design of comparison differ substantially among trials and meta-analyses (–); many evidence syntheses pool placebo, lifestyle intervention, usual care, and active glucose-lowering comparators while not balancing background therapy across randomized groups (–), which makes attributing specific hepatic and metabolic effects to an added agent challenging. The generalizability is unclear, as most reviews involve a geographically and ethnically heterogeneous population without separate evaluation for Asian study settings. There are few direct comparisons (especially those involving SGLT2 inhibitors vs. GLP-1RAs); thus, any class difference appears primarily from indirect evidence rather than establishing comparative superiority (–). Therefore, we conducted a systematic review and pairwise meta-analyses of randomized controlled trials conducted in Asian countries that evaluated the incremental hepatic and metabolic effects of adding an SGLT2 inhibitor, a GLP-1RA, or pioglitazone to comparable background glucose-lowering management in adults with T2DM and MASLD. We examined liver enzymes, markers of hepatic steatosis and fibrosis, glycemic control, lipid parameters, and anthropometric outcomes. 【Methods】 The original protocol was registered in International Prospective Register of Systematic Reviews (PROSPERO) (CRD420251231137) for a network meta-analysis of three pharmacological treatments in patients with T2DM and nonalcoholic fatty liver disease. Global inconsistency was assessed using the design-by-treatment interaction model, while local inconsistency was evaluated using the node-splitting approach. A p-value < 0.05 was considered to indicate statistically significant inconsistency. The planned network meta-analysis was not pursued after statistically significant inconsistency was identified between direct and indirect evidence (p < 0.05). The analysis was therefore revised to conventional pairwise meta-analysis based on direct comparisons. The protocol for the present analysis was retrospectively registered in PROSPERO (CRD420261457952). 【Search strategy】 We searched PubMed, Embase, the Cochrane Library, and Web of Science through 30 June 2026 for randomized controlled trials of SGLT2 inhibitors, GLP-1RAs, or pioglitazone in patients with T2DM and MASLD. The search combined Medical Subject Headings (MeSH) and free-text terms for T2DM, MASLD or nonalcoholic fatty liver disease, the three therapeutic classes, and randomized trials using Boolean operators. The full search strategies are provided in . Two reviewers independently screened titles, abstracts, and full texts. Disagreements were resolved by discussion and consensus. 【Criteria for selection and eligibility】 Studies were eligible if they met all of the following criteria: (1) participants were adults in Asian study settings with T2DM and MASLD or nonalcoholic fatty liver disease; (2) the study was a prospective randomized controlled trial with at least 12 weeks of follow-up; (3) background glucose-lowering management was comparable between randomized groups at baseline and was either maintained during follow-up or managed according to a prespecified protocol; (4) the randomized contrast evaluated the addition of an SGLT2 inhibitor, a GLP-1RA, pioglitazone, or an active comparator on otherwise comparable background management, with any protocol-specified changes in concomitant therapy documented rather than treated as unplanned cointerventions; and (5) at least one prespecified primary or secondary outcome was reported. We excluded studies that were nonrandomized, had an enrolled non-Asian study population, used noncomparable background treatment among randomization groups, allowed for unplanned changes of background medication, used concomitant glucose-lowering and/or hepatoprotective therapies potentially affecting the outcomes, reported duplicate/overlapping datasets, and provided inadequate outcome data. 【Quality assessment】 Two reviewers independently assessed risk of bias using the Cochrane Risk of Bias 2 (RoB 2) tool. As RoB 2 judgments are outcome-specific, separate assessments were performed for each randomized comparison contributing to the coprimary outcomes of change from baseline in alanine aminotransferase (ALT) and glycated hemoglobin (HbA1c). The target estimand was the effect of assignment to intervention. Assessments covered the five standard RoB 2 domains and were informed by published reports and, when available, , protocols, statistical analysis plans, and trial registries. Domain-level and overall judgments were classified as low risk of bias, some concerns, or high risk of bias according to the RoB 2 signaling questions and decision algorithms. We did not consider an open-label design alone to indicate increased risk of bias. Instead, we evaluated whether knowledge of treatment allocation could plausibly influence deviations from intended interventions or outcome measurement. ALT and HbA1c were assessed separately with respect to missing outcome data and selective reporting. Disagreements were resolved by discussion and consensus. Full signaling-question responses, domain-level judgments, and study-specific justifications are provided in , . 【Data extraction】 Two independent reviewers used a standard Microsoft Excel form to extract study design, participant demographics, diagnostic criteria, baseline clinical characteristics, intervention and comparator regimens, arm-level background glucose-lowering therapy (drug or class, dose if explicitly reported, prerandomization stability requirements, and protocol-specified adjustment rules), follow-up duration, and prespecified outcomes. Disagreements were discussed and settled through consensus. Safety data comprised adverse events, serious/severe adverse events, discontinuation because of adverse events, and other reason(s) for withdrawal. When unreported arm-level background agents/doses were noted, we recorded them as not reported (NR) instead of inferring from the information that is available. Detailed arm-level study-drug regimens and background-therapy management can be found in ; summarizes study design, follow-up, eligibility, participant populations, and outcomes contributing to our meta-analysis. We extracted population counts separately for randomized participants, treated/safety populations where explicitly reported, those who completed follow-up when available, and those analyzed per outcome. Where no treated denominator could be inferred because it was not separately reported in the original study, this was reflected accordingly. Continuous outcome data were extracted preferentially as means and standard deviations (SDs). When studies reported medians with ranges and/or interquartile ranges rather than means and SDs, corresponding means and SDs were estimated using validated conversion methods (–). For change-from-baseline outcomes, when the SD of the change score was not reported, but baseline and posttreatment SDs were available, the change-score SD was calculated according to the Cochrane Handbook as SD_change = √(SD_baseline2 + SD_post2 − 2r × SD_baseline × SD_post), where r denotes the within-participant correlation between baseline and posttreatment measurements (). A within-participant correlation coefficient of r = 0.8 was assumed for these calculations. When outcome data were reported as standard errors (SEs), they were converted to SDs using SD = SE × √n (). The coprimary endpoints were changes from baseline in ALT (the principal hepatic biochemical endpoint) and HbA1c (the principal glycemic endpoint). Secondary hepatic endpoints were aspartate aminotransferase (AST), gamma-glutamyl transferase, magnetic resonance imaging–proton density fat fraction (MRI-PDFF), liver stiffness measurement, and fibrosis-4 index. Secondary metabolic endpoints were fasting plasma glucose, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, BMI, serum adiponectin, and systolic blood pressure. 【Data synthesis and analysis】 Statistical analyses were performed using the meta package in R version 4.0.3 and Stata/MP version 17. Continuous outcomes were summarized as mean differences (MDs) with 95% confidence intervals (CIs). Pairwise meta-analyses used inverse-variance random-effects models, with between-study variance (τ2) estimated using the DerSimonian–Laird method. We report τ2 for pooled analyses when it was estimable; τ2 was not reported for single-study comparisons. For multiarm trials, only pairwise comparisons relevant to the prespecified analyses were included, and no participant group was included more than once in the same meta-analysis, thereby avoiding double-counting of shared groups. The comparison categories displayed in each forest plot represented distinct prespecified pairwise treatment contrasts rather than clinical or methodological subgroups. No clinical or methodological subgroup analyses or formal interaction tests for subgroup effects were performed, and software-generated tests for differences among the displayed comparison categories were not used for inferential purposes. Statistical heterogeneity was assessed using Cochran’s Q test and I2 statistic (p < 0.10 and I2 ≥ 50%, respectively). The forest plot indicates the direction of benefit: for outcomes for which lower values are favorable, such as ALT, AST, MRI-PDFF, liver stiffness, HbA1c, fasting plasma glucose, triglycerides, low-density lipoprotein cholesterol, systolic blood pressure, and body weight, values to the left side of the null line favor the experimental or add-on therapy. For outcomes for which higher values are favorable, such as high-density lipoprotein cholesterol and adiponectin, values to the right side of the null line favor the experimental or add-on therapy. Leave-one-out analyses were conducted to assess the influence of individual studies on the pooled estimates. Sensitivity analyses were performed for all outcomes to evaluate the robustness of the pooled estimates. Assessment of publication bias was planned using visual inspection of funnel plots only when sufficient studies were available. As most pooled analyses included fewer than 10 studies, funnel plots were considered exploratory and were not interpreted as evidence to exclude publication bias. Analyses of the two coprimary outcomes were prioritized. Secondary outcomes were considered exploratory; nominal p-values were reported without formal adjustment for multiplicity and were interpreted cautiously. As adverse-event definitions and reporting formats were heterogeneous and arm-level data were frequently unavailable, safety outcomes were synthesized narratively rather than pooled quantitatively. 【Certainty of evidence】 Certainty of evidence for the two coprimary outcomes (ALT and HbA1c) within each prespecified pairwise comparison was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (). Randomized evidence started at high certainty and was downgraded, when appropriate, for risk of bias, inconsistency, indirectness, imprecision, and publication bias. For the risk-of-bias domain, GRADE judgments were based on the final outcome-specific RoB 2 assessments for the corresponding coprimary outcome. Secondary outcomes were exploratory and were not formally graded. Imprecision was assessed primarily by comparing the 95% CI with a prespecified clinically important threshold and by evaluating the optimal information size (OIS), rather than by sample size or whether the CI crossed the null alone. A pragmatic between-group threshold of 0.5 percentage points was used for HbA1c. For ALT, a reduction of 17 U/L was used as a contextual histology-associated response threshold based on AASLD guidance (). As this is not a validated between-group minimally important difference and may be mechanism-specific, it was used only to define the GRADE imprecision boundary and not to infer histological benefit. OIS was calculated for a two-group comparison using a two-sided α = 0.05, 80% power, the corresponding threshold, and the pooled within-group SD of change scores from contributing trial arms. Certainty was rated as high, moderate, low, or very low. Detailed GRADE evidence profiles and reasons for downgrading ALT and HbA1c are provided in - for SGLT2 inhibitors versus background therapy, GLP-1RAs versus background therapy, SGLT2 inhibitors versus thiazolidinediones, and GLP-1RAs versus thiazolidinediones, respectively. 【Results】 【Description of included studies】 We identified 327 records. After deduplication and sequential title, abstract, and full-text screening, 11 randomized controlled trials (RCTs) were included in the qualitative and quantitative syntheses (). ALT and HbA1c were the coprimary outcomes; all other endpoints were analyzed as exploratory secondary outcomes. PRISMA flow diagram of study identification and selection. In each comparison arm, a total of 692 participants were randomized, with 642 contributing to the coprimary efficacy analyses. Participant ages ranged from 46.6 to 62.3 years among all included trials. Differences in totals reflect trial-specific attrition and analysis populations reported in the original trials; treated/safety populations were captured if explicitly reported, and otherwise coded as NR. SGLT2 inhibitors, GLP-1RAs, and pioglitazone interventions were included. Comparison arms comprised continued matched background therapy, placebo on matched background therapy, and active comparators including pioglitazone. Background glucose-lowering regimens differed across trials. Doses per arm of study drugs along with background-therapy details and protocol-stated stability/adjustment rules are provided in ; background drug doses are reported where explicitly available and otherwise shown as NR. Eligible trials either preserved background therapy or applied prespecified management rules that allowed us to interpret these contrasts as incremental randomized treatment strategies versus effects of unplanned comedication. Median follow-up duration was 24 weeks (range: 12–28 weeks); trials took place in China (n = 3), Japan (n = 4), India (n = 2), Iran (n = 1), and Thailand (n = 1). Details regarding studies and participants are presented in and . Study characteristics, participant populations, treatment regimens, and follow-up periods of included randomized controlled trials. R, randomized population; T, treated/safety population; A, analyzed population; ITT, intention-to-treat; NR, not reported. The count of eligible comparison arms is given. "*" indicates, in Guo et al., there were three arms: the randomized count presented here corresponds to the liraglutide and placebo arms that entered into the prespecified pairwise comparison. †" indicates that, in Horibe et al., there were 52 randomized participants overall; because an arm-level randomized denominator was not available, the analyzed population was used as the denominator for the efficacy analysis, which may differ from the randomized population. Background therapy details at each arm level, reported background drug dose(s), and protocol-stated stability/adjustment rules are summarized in ; unreported doses were not inferred. 【Risk