混合闭环系统在年龄<7岁的1型糖尿病儿童中的有效性和安全性：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
摘要 本研究旨在评估混合闭环系统对<7岁1型糖尿病儿童血糖控制的临床疗效和安全性。通过计算机检索了PubMed、Cochrane Library、Embase和Web of Science数据库中截至2025年7月30日发表的相关研究。两名研究者独立筛选文献、提取有效数据并评估纳入研究的偏倚风险，使用RevMan 5.4软件进行荟萃分析。共纳入3项随机对照试验，涉及274名儿童。结果显示，使用混合闭环系统3个月可显著增加目标范围内时间百分比（均数差[MD] = 9.95%，95%置信区间[CI] [7.8, 12.09]，p < 0.00001）并降低糖化血红蛋白（MD = -0.51%，95% CI [-0.64, -0.37]，p < 0.00001）。然而，两组在低于目标范围时间百分比方面的差异无统计学意义（MD = 0.56%，95% CI [-0.13, 1.25]，p = 0.11）。严重不良事件比较显示，混合闭环系统组与传统治疗组在糖尿病酮症酸中毒（p = 0.50）和严重低血糖（p = 0.64）风险方面的差异均无统计学意义。混合闭环系统可有效改善7岁以下1型糖尿病儿童的血糖控制，且不增加不良事件风险。仍需进一步开展高质量、大样本的临床研究来验证和补充本结论。 全球范围内，1型糖尿病（T1D）的发病率大幅增加。根据《糖尿病地图集》第11版的数据[]，2024年全球受T1D影响的人数达到950万（与2021年相比增加13%）。具体而言，其中100万受影响者为0至14岁儿童。儿童T1D发病率持续上升，年增长率为3%–4%。这一趋势在5岁以下儿童中尤为明显，他们是新增病例中增长最快的群体[]。这种发病年龄日益年轻化的转变使幼儿的血糖管理更具挑战性[,]。 年龄<7岁的1型糖尿病（T1D）儿童与年龄较大的儿童和成人相比具有不同的临床特征：他们的胰岛素需求量较低，胰岛素敏感性更高且变异性更大，饮食和活动模式难以预测，并且对低血糖无感知[]。因此，该人群达到血糖控制目标的比例较低，且发生低血糖和糖尿病酮症酸中毒的风险较高[]。尽管常规疗法如传感器增强型胰岛素泵（SAP）治疗可部分改善血糖控制[]，但在幼儿中达到目标血糖水平仍然困难[,]，且他们发生微血管并发症的风险仍然升高[,]。混合闭环系统（HCL）通过控制算法将连续血糖监测（CGM）数据与胰岛素输注实时整合，从而实现更精确和动态的血糖调节[]。其在青少年和成人中的有效性和安全性已得到充分证实[−]。研究表明，MiniMed 670G系统在儿童和青少年中将糖化血红蛋白（HbA1c）降至7.1%，并将目标范围内时间（TIR）提高至73%[]。一项关于OmniPod系统的多中心研究也表明，6岁以上儿童达到了70.1%的TIR[]。然而，由于大多数市售HCL设备被批准用于年龄>7岁的个体，关于其在更年轻儿童中使用的证据仍然有限。仅有一项荟萃分析纳入了2–7岁的参与者，但样本量较小，且未对HbA1c等结局进行全面的系统分析。 在此年龄段中，TIR、低于目标范围时间（TBR）或高于目标范围时间（TAR）[]。 鉴于全球幼儿1型糖尿病（T1D）发病率迅速上升，以及该人群血糖管理的挑战，加上目前关于该年龄段混合闭环（HCL）治疗的证据不足，本荟萃分析旨在填补这一关键证据缺口。其重点评估HCL在<7岁儿童中的血糖有效性和安全性，以支持临床决策并最终改善幼儿的血糖结局。 材料与方法 本研究遵循预注册方案进行[PROSPERO（国际前瞻性系统评价注册平台）：CRD420251137957]，并遵循PRISMA（系统评价和荟萃分析首选报告项目）指南[]。完整的PRISMA 2020检查表见补充表S1。 检索策略 两名研究人员（Tingting Zhang和Na Wang）检索了PubMed、Cochrane Library、Embase和Web of Science数据库，检索时间从建库至2025年7月30日。我们在检索策略中结合了受控词汇表术语和自由文本关键词。核心概念包括“child”（儿童）、“hybrid closed-loop or HCL”（混合闭环或HCL）、“artificial pancreas”（人工胰腺）和“Type 1 diabetes or T1D”（1型糖尿病或T1D），以及它们的同义词和缩写。检索式使用布尔逻辑运算符连接各概念模块。检索策略见补充文件S1。 纳入和排除标准 本研究的纳入标准定义如下：（a）人群：确诊为1型糖尿病（T1D）的7岁以下儿童。1型糖尿病的诊断标准采用各原始随机对照试验中定义和报告的标准，本荟萃分析未统一设定新的诊断标准；（b）干预：干预组使用混合闭环（HCL）系统进行24小时胰岛素输注；（c）对照：任何非闭环胰岛素治疗方案；（d）结局：目标范围内时间（TIR）、糖化血红蛋白（HbA1c）、低于目标范围时间（TBR）、高于目标范围时间（TAR）以及严重不良事件（SAEs）；（e）研究设计：仅限于随机对照试验，且研究持续时间不少于12周；（f）以英文发表。本荟萃分析仅纳入英文研究，以确保数据提取的准确性，最大限度地减少因翻译错误而产生的偏倚，并聚焦于儿童1型糖尿病领域公认的高质量证据。 排除标准为会议摘要、信件、病例报告、综述、动物实验、临床指南以及研究对象相同但目的不同的研究。 研究筛选与数据提取 Tingting Zhang和Na Wang独立筛选标题和摘要，以排除无关研究。随后评估其余文章全文，以确保所有相关研究均得到彻底评估。审查者之间的任何分歧将提交给独立的第三作者（Bo Wang）裁决。Tingting Zhang和Bo Wang提取了纳入研究的数据，提取的信息如下：第一作者、发表年份、研究设计、患者人数、糖尿病病程、既往治疗、混合闭环（HCL）类型、对照组治疗、研究持续时间和血糖结局。 结局定义 主要结局是目标范围内时间（TIR），即目标传感器葡萄糖范围（70–180 mg/dL）内的时间百分比。 次要结局具体如下：(a) 糖化血红蛋白；(b) 低血糖时间百分比（传感器葡萄糖 < 54 mg/dL 和 < 70 mg/dL）；(c) 高血糖时间百分比（传感器葡萄糖 > 180 mg/dL、> 250 mg/dL 和 > 300 mg/dL）。 此外，收集了安全性结局的数据，包括：(a) 严重低血糖（定义为需要治疗的精神状态改变）；(b) 糖尿病酮症酸中毒（DKA）。 纳入研究的质量评价 方法学质量由两名研究者（Na Wang、Bo Wang）使用Cochrane RoB 2工具独立评估。评估涵盖Cochrane指南定义的五个核心领域：(a) 随机化的充分性；(b) 对干预方案的依从性；(c) 结果数据的完整性；(d) 客观结果测量；(e) 避免选择性报告；(f) 总体偏倚。每个领域被归类为低风险、中度风险或高风险偏倚。 证据确定性评估 证据的确定性采用推荐分级的评估、制定与评价（GRADE）量表进行量化[]。 数据合成与分析 统计分析采用Review Manager（RevMan）软件5.4版（丹麦哥本哈根Cochrane协作网）[]进行。对于连续型结局指标，采用均数差（MD）合并效应量；若变量不符合正态分布，则使用中位数及第1、3四分位数（Q1、Q3）进行描述。对于二分类结局指标，采用比值比（OR）作为效应量进行数据合并，并以95%置信区间（CI）反映效应量的抽样误差。所有分析均采用双侧检验，以p值<0.05为差异有统计学意义。 对于纳入研究中缺失的数据，我们首先尝试通过电子邮件联系原始作者以获取补充数据。如果无法获得完整数据，则仅将具有可用数据的研究纳入meta分析，且不对缺失数据进行填补。 采用χ2检验和I2统计量评估纳入研究间的异质性。异质性分级如下：I2 < 25%为低异质性，25% ≤ I2 < 50%为中度异质性，I2 ≥ 50%为高异质性。异质性的统计学显著性定义为p < 0.10。本研究中所有结局均采用随机效应模型进行分析，无论其异质性水平如何。通过逐一剔除一项研究进行的敏感性分析用于探索异质性的潜在来源，并检验合并结果的稳健性。 为验证meta分析结果的稳定性和可靠性，对所有结局指标进行了敏感性分析。每次依次排除一项研究，并对剩余研究采用随机效应模型重新分析。评估每次排除前后合并效应量、95% CI、Z值、p值和异质性（I2）的变化。如果剔除任何单项研究后合并效应量和统计学显著性未发生实质性改变，则认为结果稳定可靠。若观察到实质性改变或统计学显著性发生逆转，则进一步探索异质性的来源。 结果 研究选择 初步筛选共获得3,951篇文章。根据纳入和排除标准，最终纳入3项研究，涉及274名受试者[−]。其识别和筛选过程如 所示。 研究选择的PRISMA流程图。 研究特征 在纳入的3项研究中，2项为随机交叉对照试验，1项采用平行组对照设计。干预组使用HCL系统进行24小时血糖管理，对照组接受常规胰岛素治疗，包括MiniMed 780G系统手动模式、CamAPS FX系统手动模式、CSII（持续皮下胰岛素输注）和MDI（每日多次注射）。随访时间约为3个月。提取了研究基本特征、受试者人口统计学特征、干预和对照措施以及基线HbA1c水平等信息，详见。 研究的基本特征。 汇总结果 主要结局（TIR） 三项纳入研究报告了目标范围内时间（TIR）。异质性分析显示，纳入研究之间无统计学显著异质性（I2 = 28%，χ2 = 2.79，df = 2，p = 0.25）。随机效应模型显示，与对照组相比，混合闭环（HCL）组的TIR显著更高。MD为9.95%（95% CI: 7.80 to 12.09），具有统计学显著性（p < 0.00001）()。通过依次排除每项纳入研究进行的敏感性分析显示，合并效应量保持一致，无明显变化（Z值范围为4.59至9.09，p < 0.00001）。 血糖结果和严重不良事件的森林图。(a) TIR。(b) HbA1c。(c) 传感器葡萄糖< 70 mg/dL的时间百分比。(d) 传感器葡萄糖< 54 mg/dL的时间百分比。(e) 严重低血糖。(f) DKA。 次要结局 HbA1c 所有三项纳入研究均提供了HbA1c（%）的数据。异质性检验显示研究间无异质性（I2 =0%，χ2 = 1.93，自由度[df] = 2，p = 0.38）。汇总结果显示，干预组的HbA1c显著低于常规胰岛素治疗组（MD –0.51%，95% CI [–0.64, –0.37]）()。这些发现在敏感性分析中稳健，其中逐个剔除单项研究并未实质性改变荟萃分析的估计值（Z值范围为4.6至7.23，p < 0.00001）。 传感器葡萄糖< 70 mg/dL的时间百分比 共有三项研究被纳入这项荟萃分析，每项研究均将TBR作为结局指标进行报告。汇总结果显示，根据随机效应模型，HCL组与常规治疗组之间TBR的差异无统计学显著性。MD为0.56%（95% CI [–0.13, 1.25]，p = 0.11），如图所示。然而，所纳入研究之间存在显著异质性（I2 = 57%，χ2 = 4.65，df = 2，p = 0.1）。 进行了留一法敏感性分析，通过逐一排除个别研究来评估汇总估计值的稳健性。在排除Battelino等[]的研究后，干预组与常规胰岛素治疗组之间TBR的差异仍无统计学显著性，MD为0.18%（95% CI [–0.44, 0.81]，p = 0.57）。异质性保持较低，I2值为0%（p = 0.53）。在排除Wadwa等[]的研究后，其余两项研究之间的异质性降低（I2 = 35%，p = 0.21）。汇总效应量变得具有统计学显著性，MD为0.84%（95% CI：0.18–1.5，p = 0.01）。当移除Ware等[]的研究时，异质性大幅增加至76%。然而，组间差异仍无显著性，MD为0.59%（95% CI [–0.48, 1.67]，p = 0.28）。 传感器葡萄糖低于54 mg/dL的时间百分比 HCL与传统胰岛素治疗之间存在临界统计学显著差异，MD为0.13%（95% CI [0.00, 0.26]，p = 0.04 ()。未观察到显著异质性（I2 = 0%，χ2 = 0.51，df = 2，p = 0.77）。 为评估合并结果的稳健性，通过依次排除每项单独研究进行敏感性分析。排除Battelino等人的研究[]后，干预组与对照组之间TBR的差异无统计学显著性（MD 0.1%，95% CI [–0.06, 0.26]，p = 0.22），异质性极小（I2 = 0%）。排除Wadwa等人的研究[]后，剩余两项研究之间的异质性保持不变（I2 = 0%，p = 0.56），但合并效应量失去统计学显著性（MD 0.15%，95% CI [–0.01, 0.32]，p = 0.07）。排除Ware等人的研究[]后，结果也变得不显著（MD 0.15%，95% CI [–0.01, 0.30]，p = 0.06），并检测到显著异质性（I2 = 76%）。 传感器葡萄糖> 180 mg/dL、>250 mg/dL和> 300 mg/dL的时间百分比 所有三项纳入研究均报告了传感器葡萄糖高于高血糖阈值的时间百分比，重点关注年龄< 7岁且患有T1D的儿童，比较使用HCL系统者与对照组。结果表明，与对照组相比，HCL组在所有高血糖阈值以上的时间百分比均较低，详见。 高血糖时间百分比。 严重不良事件 采用Mantel-Haenszel（M-H）随机效应模型合并严重低血糖和糖尿病酮症酸中毒（DKA）的风险。 严重低血糖 共纳入三项研究，包括混合闭环（HCL）组239例患者和对照组206例患者。未检测到显著异质性（I2 = 0%, χ2 = 0.3, df = 1, p = 0.58）。 关于随机化程序，所有纳入研究都概述了随机序列产生和分配隐藏的合适方法。由于HCL干预的固有特性，对参与者和工作人员实施盲法不可行，从而产生了一些关于可能偏离计划或设计干预措施的担忧。因结局数据缺失而产生的偏倚风险被评估为低风险。结局测量基于客观传感器或实验室数据，因此所有研究在该领域的风险评级均为低风险。未发现任何证据提示存在选择性结局报告。在三项纳入研究中，一项被归类为低偏倚风险[]，其余两项被归类为中偏倚风险[,]。偏倚风险评估的详细信息见补充图S1。 证据确定性 使用GRADE框架评估了HCL治疗与常规治疗在年龄<7岁的T1D儿童中关键结局的证据确定性。临床推荐强度通过综合证据确定性、获益-风险平衡和临床实践需求来确定，结果总结于。 结果总结（GRADE） 血糖控制（TIR和HbA1c） 证据确定性：中等（⊕⊕⊕○，因不精确性而降级，因为样本量有限）。 推荐强度：强烈推荐混合闭环（HCL）治疗。混合闭环（HCL）治疗显著增加目标范围内时间（TIR）并降低糖化血红蛋白（HbA1c），对血糖控制具有明确且有临床意义的益处，这
全文（开放获取） · 英文原文
【Abstract】 This study aimed to evaluate the clinical efficacy and safety of hybrid closed-loop systems for glycemic control in children aged < 7 years with Type 1 diabetes. Relevant studies published up to July 30, 2025 were electronically retrieved from the PubMed, Cochrane Library, Embase, and Web of Science databases. Two investigators independently screened the literature, extracted valid data, and assessed the risk of bias of included studies, with a meta-analysis conducted via RevMan 5.4 software. Three randomized controlled trials, with 274 children enrolled, were incorporated into the analysis. The results indicated that 3 months of using the hybrid closed-loop system significantly increased the percentage of time in range (mean difference [MD] = 9.95%, 95% confidence interval [CI] [7.8, 12.09], p < 0.00001) and reduced glycated hemoglobin (MD = -0.51%, 95% CI [-0.64, -0.37], p < 0.00001). However, no significant difference was noted in the percentage of time below range between the two groups (MD = 0.56%, 95% CI [-0.13, 1.25], p = 0.11). A comparison of serious adverse events showed no statistically significant differences in the risk of diabetic ketoacidosis (p = 0.50) and severe hypoglycemia (p = 0.64) between the hybrid closed-loop system group and the conventional therapy group. The hybrid closed-loop system can effectively improve glycemic control in children under 7 years old with Type 1 diabetes without increasing the risk of adverse events. Further high-quality, large-sample clinical studies are still required to verify and supplement the present conclusions. 【Introduction】 Worldwide, there has been a substantial increase in the incidence of Type 1 diabetes (T1D). According to data from the 11th edition of the Diabetes Atlas[], in 2024, the global prevalence count of individuals affected by T1D reached 9.5 million (a 13% increase compared with 2021). Specifically, 1 million of those affected were children aged between 0 and 14 years. The incidence of T1D in children continues to rise, with an annual growth rate of 3%–4%. This trend is particularly pronounced among children under 5 years old, who represent the fastest-growing group of new cases[]. This shift toward an increasingly younger age of onset makes glycemic management in young children more challenging[,]. Children aged < 7 years with T1D present distinct clinical characteristics compared with older children and adults: they have lower insulin requirements, higher and more variable insulin sensitivity, unpredictable eating and activity patterns, and nonawareness of hypoglycemia[]. As a result, this population achieves target glycemic control less frequently and is at higher risk of hypoglycemia and diabetic ketoacidosis[]. Although conventional therapies such as sensor-augmented pump (SAP) therapy can partially improve glycemic control[], achieving the target glucose levels remains difficult in young children[,], and they remain at an elevated risk of microvascular complications[,]. The hybrid closed-loop system (HCL) integrates continuous glucose monitoring (CGM) data with insulin delivery in real time via a control algorithm, enabling more precise and dynamic glycemic regulation[]. Its efficacy and safety have been well established in adolescents and adults[−]. Studies have shown that the MiniMed 670G system reduced glycated hemoglobin (HbA1c) to 7.1% and increased time in range (TIR) to 73% in children and adolescents[]. A multicenter study on the OmniPod system also demonstrated that children over 6 years old achieved a TIR of 70.1%[]. However, as most commercially available HCL devices are approved for use in individuals aged > 7 years, evidence regarding their use in younger children remains limited. Only one meta-analysis included participants aged 2–7 years, but the sample size was small, and no comprehensive systematic analysis was conducted on outcomes such as HbA1c, TIR, time below range (TBR), or time above range (TAR) in this age group[]. Given the rapidly increasing global incidence of T1D in young children and the challenges in glycemic management within this population, coupled with the current insufficient evidence regarding HCL therapy in this age group, this meta-analysis aims to fill a critical evidence gap. It focuses specifically on evaluating the glycemic efficacy and safety of HCL in children aged < 7 years, with the objective of supporting clinical decision-making and ultimately enhancing glycemic outcomes in young children. 【Materials and methods】 This study was performed in line with a pre-registered protocol [PROSPERO (International Prospective Register of Systematic Reviews): CRD420251137957] and followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines[]. The completed PRISMA 2020 checklist is provided in Supplementary Table S1. 【Search strategy】 Two investigators (Tingting Zhang and Na Wang) searched the PubMed, Cochrane Library, Embase, and Web of Science databases from their inception to July 30, 2025. We combined both controlled vocabulary terms and free-text keywords in our search strategy. The key concepts encompassed "child", "hybrid closed-loop or HCL", "artificial pancreas" and "Type 1 diabetes or T1D", along with their synonyms and abbreviations. The search queries were constructed using Boolean logic operators to link the conceptual blocks.The search strategies are provided in Supplemental File S1. 【Inclusion and exclusion criteria】 The inclusion criteria for this study were defined as follows: (a) Population: children under 7 years old with a confirmed diagnosis of T1D. The diagnostic criteria for T1D were those defined and reported in each original randomized controlled trial, and no uniform diagnostic standard was newly set in this meta-analysis; (b) Intervention: the intervention group used HCL for 24-h insulin delivery; (c) Comparison: any non-closed-loop insulin therapy regimen; (d) Outcomes: TIR, HbA1c, TBR, TAR, and severe adverse events (SAEs); (e) Study design: limited to randomized controlled trials and a study duration of no less than 12 weeks; (f) Published in English. This meta-analysis restricted inclusion to English-language studies to ensure the accuracy of data extraction, minimize potential bias arising from translation errors, and focus on well-established high-quality evidence in pediatric T1D. Exclusion criteria were conference abstracts, letters, case reports, reviews, animal experiments, clinical guidelines, and studies with the same participants but different purposes. 【Study selection and data extraction】 Tingting Zhang and Na Wang independently screened the titles and abstracts to exclude irrelevant studies. The full texts of the remaining articles were then assessed to ensure that all relevant studies were thoroughly evaluated. Any disagreements between the reviewers would be referred to an independent third author (Bo Wang) for resolution. Data from the included studies was extracted by Tingting Zhang and Bo Wang, and the extracted information was as follows: First author, year of publication, study design, number of patients, duration of diabetes, prior therapy, type of HCL, control group treatment, study duration, and glycemic outcomes. 【Outcome definition】 The primary outcome was the TIR, namely the percentage of time spent in the target sensor glucose range (70–180 mg/dL). The secondary outcomes are specified below: (a) Glycated hemoglobin; (b) Percentage of hypoglycemic time (sensor glucose < 54 mg/dL and < 70 mg/dL); (c) Percentage of hyperglycemic time (sensor glucose > 180 mg/dL, > 250 mg/dL, and > 300 mg/dL). Additionally, data on safety outcomes were gathered, which included: (a) Severe hypoglycemia (defined as requiring treatment for altered mental status); (b) Diabetic ketoacidosis (DKA). 【Quality assessment of the included studies】 Methodological quality was appraised independently by two investigators (Na Wang and Bo Wang) using the Cochrane RoB 2 tool. The assessment covered five core domains defined by the Cochrane guidelines: (a) Adequacy of randomization; (b) Adherence to the intervention protocols; (c) Completeness of outcome data; (d) Objective outcome measurement; (e) Selective reporting avoidance; (f) Overall bias. Each domain was categorized as low, with a moderate risk or high risk of bias. 【Certainty of evidence assessment】 The certainty of the evidence was quantified using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) scale[]. 【Data synthesis and analysis】 Statistical analyses were conducted utilizing Review Manager (RevMan) software version 5.4 (The Cochrane Collaboration, Copenhagen, Denmark)[]. For continuous outcome measures, the mean difference (MD) was used to pool the effect sizes; if the variables were not normally distributed, the median Quantiles 1 and 3 (Q1 and Q3) was used for description. For dichotomous outcome measures, the odds ratio (OR) was used as the effect size for data pooling, with the 95% confidence interval (CI) to reflect the sampling error of the effect sizes. Two-tailed tests were used for all analyses, with statistical significance defined as a p-value < 0.05. For data missing from the included studies, we first attempted to contact the original authors via email to obtain supplementary data. If complete data could not be obtained, only studies with available data were included in the meta-analysis, and no imputation for missing data was performed. Heterogeneity among included studies was evaluated using the χ2 test and the I2 statistic. Heterogeneity was graded as follows: low heterogeneity for I2 < 25%, moderate heterogeneity for 25% ≤ I2 < 50%, and high heterogeneity for I2 ≥ 50%. The statistical significance of heterogeneity was defined as p < 0.10. All outcomes in this study were analyzed using the random effects model regardless of their heterogeneity level. Sensitivity analyses conducted by sequentially omitting one study at a time were performed to explore potential sources of heterogeneity and to test the robustness of the pooled results. To verify the stability and reliability of the meta-analysis results, sensitivity analyses were performed for all outcome measures. One study at a time was sequentially excluded, and the remaining studies were re-analyzed using the random effects model. Changes in the pooled effect size, 95% CI, Z-value, p-value, and heterogeneity (I2) were evaluated before and after each exclusion. The result was considered to be stable and reliable if the pooled effect size and statistical significance did not change substantially after the exclusion of any single study. If substantial changes or reversed statistical significance were observed, the source of heterogeneity was explored further. 【Results】 【Study selection】 In total, 3,951 articles were obtained through preliminary screening. According to the inclusion and exclusion criteria, three studies involving 274 participants were finally selected for inclusion[−]. The process of their identifcation and selection is shown in . PRISMA flowchart for study selection. 【Study characteristics】 Across the three included studies, two were randomized crossover controlled trials and one used a parallel-group controlled design. The intervention groups utilized a HCL system for 24-h glycemic management, whereas the control groups received conventional insulin therapy, including the manual mode of the MiniMed 780G system, the manual mode of the CamAPS FX system, CSII (Continuous Subcutaneous Insulin Infusion), and MDI (Multiple Daily Injections). The follow-up period was approximately 3 months. Information such as the basic study characteristics, the participants' demographics, intervention and control measures, and baseline HbA1c levels were extracted, as detailed in . 【Basic characteristics of the studies.】 【Results of the synthesis】 【Primary outcome (TIR)】 Three included studies reported the time TIR. The heterogeneity analysis revealed no statistically significant heterogeneity across the included studies (I2 = 28%, χ2 = 2.79, df = 2, p = 0.25). The random effects model indicated that TIR was significantly higher in the HCL group compared with the controls. The MD was 9.95% (95% CI: 7.80 to 12.09), which was statistically significant (p < 0.00001) (). Sensitivity analysis, performed by sequentially excluding each included study, showed that the pooled effect size remained consistent without substantial change (Z-values ranged from 4.59 to 9.09, p < 0.00001). Forest plots of glycemic outcomes and serious adverse events. (a) TIR. (b) HbA1c. (c) Percentage of time with sensor glucose < 70 mg/dL. (d) Percentage of time with sensor glucose < 54 mg/dL. (e) Severe hypoglycemia. (f) DKA. 【Secondary outcomes】 【HbA1c】 All three included studies provided data on HbA1c (%). Heterogeneity testing indicated no heterogeneity across the studies (I2 =0%, χ2 = 1.93, degrees of freedom [df] = 2, p = 0.38). Pooled findings derived showed that the HbA1c was significantly lower in the intervention group than in the conventional insulin therapy group (MD –0.51%, 95% CI [–0.64, –0.37]) (). The findings were robust across sensitivity analyses, in which the iterative removal of individual studies did not substantially change the estimates of the meta-analysis (Z-values ranged from 4.6 to 7.23, p < 0.00001). 【Percentage of time with sensor glucose < 70 mg/dL】 Three studies were included in the meta-analysis, each reporting TBR as an outcome measure. The pooled results indicated that the difference in TBR between the HCL and conventional therapy groups was not statistically significant, according to a random-effects model. The MD was 0.56% (95% CI [–0.13, 1.25], p = 0.11), as shown in . However, considerable heterogeneity was detected among the included studies (I2 = 57%, χ2 = 4.65, df = 2, p = 0.1). Leave-one-out sensitivity analysis was performed, systematically excluding individual studies to assess the robustness of the pooled estimates. After excluding the study by Battelino et al.[], the difference in TBR between the intervention and conventional insulin therapy groups remained statistically non-significant, with an MD of 0.18% (95% CI [–0.44, 0.81], p = 0.57). Heterogeneity remained low, with an I2 value of 0% (p = 0.53). Upon exclusion of the study by Wadwa et al.[], heterogeneity among the remaining two studies decreased (I2 = 35%, p = 0.21). The pooled effect size became statistically significant, showing an MD of 0.84% (95% CI: 0.18–1.5, p = 0.01). When the study by Ware et al. was removed[], heterogeneity increased substantially to 76%. However, the difference between groups remained nonsignificant, with an MD of 0.59% (95% CI [–0.48, 1.67], p = 0.28). 【Percentage of time with sensor glucose < 54 mg/dL】 There was a borderline statistical significant difference between the HCL and conventional insulin therapy, with an MD of 0.13% (95% CI [0.00, 0.26], p = 0.04 (). No significant heterogeneity was observed (I2 = 0%, χ2 = 0.51, df = 2, p = 0.77). To assess the robustness of the pooled results, sensitivity analysis was conducted by excluding each individual study in sequence. After excluding the study by Battelino et al.[], the difference in TBR between the intervention and control groups was not statistically significant (MD 0.1%, 95% CI [–0.06, 0.26], p = 0.22), with minimal heterogeneity (I2 = 0%). When the study by Wadwa et al. was removed[], the heterogeneity among the remaining two studies remained unchanged (I2 = 0%, p = 0.56), whereas the pooled effect size lost its statistical significance (MD 0.15%, 95% CI [–0.01, 0.32], p = 0.07). After excluding the study by Ware et al.[], the result also became nonsignificant (MD 0.15%, 95% CI [–0.01, 0.30], p = 0.06), with substantial heterogeneity detected (I2 = 76%). 【Percentage of time with sensor glucose > 180 mg/dL, >250 mg/dL, and > 300 mg/dL】 The percentage of time spent with sensor glucose above the hyperglycemia thresholds was reported in all three included studies, focusing on children aged < 7 years with T1D. comparing those using a HCL system with a control group. The results demonstrated that the HCL group spent a lower percentage of time above all hyperglycemia thresholds compared with the control group, as detailed in . 【Percentage of time with hyperglycemia.】 【Serious adverse events】 The risk of severe hypoglycemia and DKA was pooled using the Mantel-Haenszel (M-H) random-effects model. 【Severe hypoglycemia】 Three studies involving 239 patients in the HCL group and 206 in the control group were included. No significant heterogeneity was detected (I2 = 0%, χ2 = 0.3, df = 1, p = 0.58). The pooled results demonstrated no statistically significant difference in the risk of severe hypoglycemia between the two groups (odds ratio [OR] = 1.51, 95% CI: 0.22–10.51, p = 0.68) (). The effect size for the Ware et al.[] study was not estimable because there were zero events in both groups. 【Diabetic ketoacidosis】 Data from the same three studies were analyzed. Heterogeneity analysis indicated no between-study heterogeneity (I2 = 0%, χ2 = 0.09, df = 1, p = 0.77). The meta-analysis showed no significant difference in the risk of DKA between the HCL group and the control group (OR = 2.16, 95% CI: 0.22–21.04, p = 0.51) (). The effect of the Battelino et al.'s study[] was not estimable as no DKA events occurred in either arm. For both of the abovementioned serious adverse events, a sensitivity analysis was conducted by using the leave-one-out method, whereby the pooled effect size was recalculated after sequentially excluding each individual study. The results demonstrated no substantial changes in the pooled effect estimates or their statistical significance, confirming the robustness of our findings. 【Results of assessing the risk of bias】 In relation to the randomization procedure, all included studies outlined suitable approaches for both the generation of random sequences and the concealment of allocation. Because of the inherent nature of the HCL intervention, blinding of the participants and personnel was not feasible, resulting in some worries regarding possible departures from the interventions that were planned or designed. The risk of bias arising from missing outcome data was evaluated as low. Measurements of the outcomes were based on objective sensor or laboratory data, leading to a low risk rating in this domain for all studies. No evidence was found to suggest the presence of selective outcome reporting. Of the three included studies, one was categorized as having a low risk of bias[], whereas the remaining two were categorized as having a moderate risk of bias[,]. Details of the bias risk assessment are presented in Supplementary Fig. S1. 【Certainty of the evidence】 The certainty of the evidence for key outcomes of HCL therapy versus conventional therapy in children aged < 7 years with T1D was assessed using the GRADE framework. The clinical recommendation strength was determined by synthesizing the certainty of evidence, the benefit–risk balance, and the needs of clinical practice, with the results summarized in . 【Summary of the findings (GRADE)】 【Glycemic control (TIR and HbA1c)】 Certainty of evidence: Moderate (⊕⊕⊕○, downgraded for imprecision because of the limited sample size). Recommendation strength: Strong recommendation for HCL therapy. HCL therapy significantly increased TIR and reduced HbA1c, with clear and clinically meaningful benefits for glycemic control, which