亚临床甲状腺功能减退对多囊卵巢综合征患者核心生殖激素和卵巢内分泌状态的影响：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：本系统评价和荟萃分析旨在评估亚临床甲状腺功能减退症（SCH）对多囊卵巢综合征（PCOS）女性血清生殖激素水平的影响，包括卵泡刺激素（FSH）、黄体生成素（LH）、雌二醇、催乳素和总睾酮，并为临床实践提供证据。 方法：我们检索了PubMed、Web of Science、Embase和Cochrane Library，检索时间从建库至2026年4月25日，纳入相关观察性研究。采用纽卡斯尔-渥太华量表（NOS）评估研究质量。使用R软件进行统计分析，以标准化均数差（SMD）和95%置信区间（CI）作为效应量。使用I²统计量评估异质性，随后进行亚组分析、敏感性分析和发表偏倚检验。 结果：共纳入13项研究，涉及3,114名参与者（579例合并SCH的PCOS患者和2,535例甲状腺功能正常的PCOS对照组）。合并分析显示，两组间血清FSH、LH、雌二醇或总睾酮水平均无统计学显著差异。对于催乳素，在事后排除一项导致极端异质性的离群研究后，出现显著升高；这一发现属于探索性结果。FSH和LH未检测到显著发表偏倚；然而，对于其他三个结局，检验效能不足，且Begg检验提示催乳素可能存在小研究效应。根据建议评估、发展和评价分级（GRADE）框架，FSH的证据确定性为低，LH、雌二醇、催乳素和总睾酮为极低。 结论：当前证据显示，SCH与甲状腺功能正常的PCOS对照组之间FSH、LH、雌二醇或总睾酮水平无显著差异，而催乳素的证据仍不确定。高异质性和观察性研究设计妨碍了明确结论或因果推断。系统综述注册：https://www.crd.york.ac.uk/PROSPERO/view/CRD420261382326，标识符CRD420261382326。 引言 多囊卵巢综合征（PCOS）是育龄期女性最常见的内分泌代谢性疾病。其特征为排卵功能障碍、高雄激素血症和多囊卵巢形态，并增加不孕、月经紊乱和代谢综合征的风险 (, )。亚临床甲状腺功能减退症（SCH）是一种常见的甲状腺疾病，定义为血清促甲状腺激素（TSH）升高而游离甲状腺素（FT4）水平正常 (, )。近年来，由于生活方式和饮食模式的改变，育龄期女性中PCOS合并SCH的患病率显著增加 (, )。这种合并病使临床管理复杂化，并引发了关于SCH是否与PCOS患者更严重的生殖内分泌紊乱相关的广泛争论。 迄今为止，在临床观察研究和既往发表的荟萃分析中，关于亚临床甲状腺功能减退症（SCH）与多囊卵巢综合征（PCOS）核心生殖激素紊乱之间的定量关联仍存在高度争议（, ）。由Pergialiotis等人于2017年发表的里程碑式荟萃分析（），在关于PCOS患者甲状腺筛查的临床讨论中被广泛引用，其仅综合了2016年之前发表的文献，仅评估了卵泡刺激素（FSH）、黄体生成素（LH）和性激素结合球蛋白（SHBG），未对雌二醇、催乳素和总睾酮进行全面的汇总分析，并且缺乏基于临床可变TSH诊断切点或甲状腺自身免疫状态的分层分析。Ding等人于2018年发表的荟萃分析（）仅专注于估计PCOS人群中SCH的合并患病率，未包含生殖激素谱的定量合成；因此，它无法解决各原始研究之间相互矛盾的内分泌结局。Xing等人于2021年发表的另一项大规模荟萃分析（）纳入了27项研究，但严重依赖中国灰色文献。该研究既未应用推荐分级的评估、制定与评价（GRADE）框架进行确定性评级，也未进行严格的留一法分析或按偏倚风险分层的敏感性分析来探讨研究间异质性。其他于2022年发表的叙述性综述仅提供了定性总结，没有汇总的定量效应量，因此为临床内分泌评估提供的可行指导有限。 评估(). 最近两篇综述指出现有证据存在重大空白。一项2022年系统综述汇总了全球横断面数据，显示亚临床甲状腺功能减退症（SCH）与生殖激素之间的相关性在不同种族队列中存在矛盾，且没有更新的定量综合来澄清这些差异（）。一项2023年综述概述了甲状腺–卵巢轴的相互作用，并将甲状腺抗体阳性及不一致的促甲状腺激素（TSH）截断值确定为导致研究间高度异质性的两个关键混杂因素；然而，截至发表时，尚无荟萃分析对这两个因素进行分层定量分析（）。2021年至2026年间，来自多个国家的新观察性队列相继发表，均超出来自既往三项荟萃分析的检索窗口，导致当前证据基础出现显著的时间滞后。大多数原始研究采用单中心、小样本设计，没有统一的激素检测方案或标准化的混杂因素校正。这些不一致性使现有数据碎片化，阻碍临床医生为合并SCH的多囊卵巢综合征（PCOS）患者建立标准化的内分泌监测计划。 因此，开展本系统综述和荟萃分析，以整合更新的观察性证据，并检验SCH与PCOS女性血清促卵泡激素（FSH）、黄体生成素（LH）、雌二醇、催乳素和总睾酮水平之间的关联。我们旨在描述当前关于激素水平差异的证据，并确定需要进一步研究的方法学空白。 材料与方法 研究注册 本系统综述和荟萃分析按照《Cochrane干预措施系统评价手册》进行设计和实施。我们遵循了系统综述和荟萃分析的首选报告条目（PRISMA）声明（）。该综述方案于2026年5月1日在国际系统评价前瞻性注册库（PROSPERO）中正式注册（ID：CRD420261382326），可在https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD420261382326公开获取。 数据库检索于2026年4月25日完成，该阶段仅进行引文导出。所有后续的筛选、数据提取、亚组分析和敏感性分析均按照注册后的注册方案进行，在观察汇总结果后未对结局或分析计划进行事后调整。 纳入和排除标准 采用常用于观察性系统综述的人群、暴露、对照和结局（PECO）框架来确定研究问题和资格标准（）。合格的研究设计包括横断面研究、病例对照研究和回顾性队列研究。仅纳入报告上述结局的均值和标准差（SD）（或可转换为均值和标准差的数据）的原始研究。综述、信件、会议摘要、动物研究以及无可提取结局数据的研究均被排除。 用于研究资格筛选的PECO框架。 人群、干预、对照、结局和研究设计（PICOS）框架是为干预性试验开发的，包含一个不适合本观察性荟萃分析的干预成分。此处采用PECO以符合暴露－对照研究设计，PICOS的所有核心纳入维度均整合进上述四个PECO类别中。 检索策略 在PubMed、Web of Science核心合集、Embase和Cochrane Library中进行了电子检索。所有检索均于2026年4月25日通过各数据库的官方网络界面完成，并纳入自数据库建库至截止日期的所有记录。仅纳入经同行评审的英文期刊文章。 我们将分析限制在英文出版物，原因有三：高质量的相关观察性研究主要以英文发表；使用单一语言可避免翻译偏倚和不一致的数据提取；非英文的区域性论文多为方法学各异的小型单中心研究，可能引入额外的异质性。尽管如此，这一规则可能因遗漏符合条件的非英文区域性出版物而造成语言偏倚。 未进行试验注册库和灰色文献的检索，包括会议摘要、学位论文和预印本，也未进行引文追踪。未发表或未经同行评审的材料通常缺乏完整的结局数据和详细的方法，因此我们排除了这些来源以提高合并数据的可靠性。 检索语法整合了医学主题词表（MeSH）/Emtree受控词以及由布尔运算符连接的自由文本关键词。每个数据库的完整可复现检索式列于 。 文献筛选与数据提取 文献筛选和数据提取由刘云辉和赵平独立完成。筛查判断中的所有分歧首先通过讨论解决；任何剩余分歧由第三位审阅者邓婷婷裁定。使用EndNote X9整理所有检索到的引文并去除重复。两位审阅者根据预先定义的PECO纳入标准，离线分别筛选标题、摘要和全文，然后交叉核对筛选结果。所有提取的定量数据均记录在标准化的Excel表格中。 提取的项目包括： 基本研究信息：第一作者、发表年份、招募国家、研究设计类型； 参与者基线特征：研究总样本量、多囊卵巢综合征（PCOS）合并亚临床甲状腺功能减退症（SCH）的患者人数、年龄、体重指数（BMI）、是否使用鹿特丹PCOS诊断标准，以及用于定义SCH的特定促甲状腺激素（TSH）切点值； 甲状腺自身免疫数据：是否进行了甲状腺过氧化物酶抗体（TPOAb）和甲状腺球蛋白抗体（TgAb）检测，以及是否分别报告了抗体阳性与抗体阴性的SCH亚组的生殖激素结局； 测量的生殖激素指标：卵泡刺激素（FSH）、黄体生成素（LH）、雌二醇、催乳素、总睾酮； 原始定量结局数据：多囊卵巢综合征（PCOS）合并亚临床甲状腺功能减退症（SCH）的患者与甲状腺功能正常的 PCOS 亚组中每种激素的样本量、平均值和标准差； 混杂因素调整记录：关键协变量的匹配或多变量调整状态，包括年龄、体质指数（BMI）、胰岛素抵抗、PCOS 表型、甲状腺自身免疫、用药史、不孕状态、采血时的月经周期阶段以及标准化激素检测方案。 对于未直接报告标准差（SD）的研究，采用 Cochrane 手册推荐的方法，根据 p 值、t 值、95% 置信区间（CI）、标准误或 CI 限值估算 SD 值。如果研究报告了中位数和四分位距，则使用 Luo et al. () 和 Wan et al. () 的方法将数据转换为均值和 SD。 根据研究设计采用了不同的质量评估工具：横断面研究使用美国医疗保健研究与质量局（AHRQ）的9条目检查表进行评估，而病例对照研究和队列研究则使用纽卡斯尔-渥太华量表（NOS）进行评价（, ）。AHRQ检查表和NOS的原始总评分仅作为补充信息，而非呈现偏倚风险结果的主要指标。为提高临床结果的可解释性，我们进一步对所有符合条件的研究在八个关键临床领域进行了独立评估：① 研究对象选择；② 亚临床甲状腺功能减退症（SCH）的定义与测量；③ 多囊卵巢综合征（PCOS）的诊断标准；④ SCH组与甲状腺功能正常的PCOS对照组之间的可比性；⑤ 对混杂因素的调整或匹配；⑥ 激素测定；⑦ 激素检测的采血时间；⑧ 结局数据的完整性。每个领域被分类为低、中或高偏倚风险。 统计分析 所有统计分析均在R 4.3.1中使用meta、metafor和dmetar软件包进行。对于连续结局，采用标准化均数差（SMD）作为主要效应量，同时报告均数差（MD）及其95%置信区间，以统一各研究间的单位。由于观察性数据固有的临床异质性，所有分析预先指定使用DerSimonian–Laird τ²估计的随机效应模型；固定效应模型作为敏感性检验。异质性采用Cochran’s Q和I²进行评估（I² > 50% = 高度异质性）。我们进行了Hartung–Knapp校正的随机效应敏感性分析，并计算了所有结局的95%预测区间（PI），以估计未来研究中预期的效应范围，所有预测区间均在结果部分报告。预先计划了三个亚临床甲状腺功能减退症（SCH）亚组：低TSH截断值（约2.5 mIU/L）、成人标准TSH截断值（4.0–5.0 mIU/L）以及抗体阳性的SCH。还进行了额外的留一法分析和偏倚风险分层敏感性分析，以探索异质性的来源。采用Egger检验和Begg检验评估发表偏倚。采用适用于观察性证据的推荐分级评估、制定与评价（GRADE）类框架评估证据确定性。 结果 文献检索结果 共初步识别出2,886条记录。排除351条重复记录后，2535条记录进行了标题和摘要筛选。大多数不符合条件的文章因研究类型不当、研究对象不相关和研究主题无关而被排除，剩下285篇文章进行全文评估。进一步的全文筛选排除了没有SCH组或有效生殖激素指标的研究。7项无法提取数据的研究也被排除。最终，13项符合条件的研究被纳入系统评价和荟萃分析。详细的筛选过程见。具有代表性的排除研究的完整细节及排除的具体原因见。 研究选择过程的PRISMA流程图。 纳入文献的基本特征 共纳入来自9个国家的13项横断面、病例对照和回顾性队列研究（–），涉及3,114名多囊卵巢综合征（PCOS）女性（其中579名患有亚临床甲状腺功能减退症（SCH），2,535名甲状腺功能正常的PCOS对照者）；所有研究均采用鹿特丹PCOS标准，而SCH的诊断采用不同的促甲状腺激素（TSH）切点，范围为2.5至5.0 mIU/L，且甲状腺激素水平正常。所测量的生殖激素包括促卵泡激素（FSH）、黄体生成素（LH）、雌二醇、催乳素和总睾酮，大多数研究仅按年龄和体质指数（BMI）匹配受试者，未对包括胰岛素抵抗和PCOS表型在内的关键混杂因素进行多变量校正。两项研究（, ）检测了抗甲状腺过氧化物酶抗体/抗甲状腺球蛋白抗体（TPOAb/TgAb）甲状腺自身抗体，仅Gawron等人（）分别报告了抗体阳性自身免疫性SCH和孤立性非自身免疫性SCH亚组的生殖激素结果；其余11项研究未检测甲状腺自身抗体，也未提供按抗体分层的激素数据，因此基于甲状腺自身免疫状态的定量亚组荟萃分析因可用的分层证据不足而无法进行（）。文献的基本特征。 SCH，亚临床甲状腺功能减退；FT4，游离甲状腺素；T3，三碘甲状腺原氨酸；T4，甲状腺素；TSH，促甲状腺激素；FSH，促卵泡激素；LH，黄体生成素；TPOAb，甲状腺过氧化物酶抗体；TgAb，甲状腺球蛋白抗体；NA，数据不可用；IQR，四分位距；IR，胰岛素抵抗；TAI，甲状腺自身免疫；Cycle，血液样本采集的月经周期时间；Assay，标准化实验室激素检测方法。混杂控制状态的缩写：M = 匹配（SCH组与甲状腺功能正常的PCOS对照组之间进行基线匹配）；A = 调整（在统计分析中进行多变量回归调整）；NR = 未报告/未对相应变量进行匹配或调整。 文献质量评价 并基于八个预定义评价维度展示所有纳入研究的领域分层偏倚风险评估结果，将原始AHRQ总分列为补充证据。在所有研究中，D2（SCH定义和测量）、D3（PCOS诊断标准）、D6（激素测定方法）和D8（结局数据完整性）均观察到一致的低偏倚风险，因为所有文章均清晰报告了统一的诊断阈值和完整的实验室检测信息，且无核心结局数据缺失。主要方法学缺陷集中在D4（组间可比性）和D5（混杂因素调整）：多项研究未能在组间平衡基线特征，或未调整BMI和年龄等关键混杂因素，导致偏倚风险较高。中等风险常见于D1（参与者选择）和D7（激素测量时机），原因是非概率抽样以及激素采血周期的记录不充分。总体而言，七项研究的总偏倚风险为低，五项为中等，Fatima () 的研究因组间可比性和混杂控制两项均为高风险评级而被归类为高偏倚风险。 纳入研究的按维度分层的偏倚风险评估。 八个评价域的定义：D1：参与者选择；D2：SCH定义和测量；D3：PCOS诊断标准；D4：组间可比性；D5：混杂因素调整；D6：激素测定方法；D7：激素测量时机；D8：结局数据完整性。 对纳入的观察性研究进行领域特异性偏倚风险评估。八个评估领域的定义：D1：参与者选择；D2：亚临床甲状腺功能减退症（SCH）的定义与测量；D3：多囊卵巢综合征（PCOS）诊断标准；D4：组间可比性；D5：混杂因素调整；D6：激素测定方法；D7：激素测量时间；D8：结局数据的完整性。 伴与不伴亚临床甲状腺功能减退症的多囊卵巢综合征患者血清激素水平的比较 卵泡刺激素（FSH）水平 共纳入12项研究（–, –）。分析包括523例合并SCH的PCOS患者和1,986例甲状腺功能正常的PCOS对照者。随机效应模型显示两组间血清FSH水平无显著差异（SMD = 0.038，95% CI：−0.095至0.170，95% PI：−0.28至0.37，p = 0.581）（）。检测到中度异质性（I² = 36.4%，Q检验p = 0.099；研究间方差τ² = 0.018）。留一法分析表明Dai等人（）的研究是异质性的主要来源，将其排除后异质性得以解决。在逐一剔除各项研究后，合并结果保持不变（）。基于Hartung–Knapp校正随机效应模型的敏感性分析产生了合并效应量及置信区间（CI）的重叠，且总体统计推断未发生变化。根据Cohen's d标准，合并SMD为0.038代表极小效应量（|SMD| < 0.2），提示基于当前观察性证据，所观察到的FSH浓度差异对于卵巢储备评估不太可能具有临床意义。 多囊卵巢综合征（PCOS）伴或不伴亚临床甲状腺功能减退（SCH）患者卵泡刺激素（FSH）水平的Meta分析。FSH水平的留一法敏感性分析。
全文（开放获取） · 英文原文
【Abstract】 OBJECTIVE: This systematic review and meta-analysis aimed to evaluate the effects of subclinical hypothyroidism (SCH) on serum levels of reproductive hormones, including follicle-stimulating hormone (FSH), luteinizing hormone (LH), estradiol, prolactin, and total testosterone, in women with polycystic ovary syndrome (PCOS) and to provide evidence for clinical practice. METHODS: We searched PubMed, Web of Science, Embase, and Cochrane Library from inception to 25 April 2026 for relevant observational studies. Study quality was assessed using the Newcastle-Ottawa Scale (NOS). Statistical analyses were performed using R software, with the standardized mean difference (SMD) and 95% confidence interval (CI) used as effect measures. Heterogeneity was evaluated using the I² statistic, followed by subgroup analysis, sensitivity analysis, and publication bias tests. RESULTS: Thirteen studies involving 3,114 participants (579 PCOS patients with SCH and 2,535 euthyroid PCOS controls) were included. Pooled analyses showed no statistically significant differences in serum FSH, LH, estradiol, or total testosterone levels between groups. For prolactin, a significant elevation emerged after post-hoc exclusion of an outlying study contributing to extreme heterogeneity; this finding is exploratory. No significant publication bias was detected for FSH and LH; however, tests were underpowered for the other three outcomes, and Begg's test suggested potential small-study effects for prolactin. Certainty according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework was low for FSH and very low for LH, estradiol, prolactin, and total testosterone. CONCLUSION: Current evidence shows no significant differences in FSH, LH, estradiol, or total testosterone levels between SCH and euthyroid PCOS controls, while the evidence for prolactin remains uncertain. High heterogeneity and the observational study designs preclude definitive conclusions or causal inference.Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD420261382326, identifier CRD420261382326. 【Introduction】 Polycystic ovary syndrome (PCOS) is the most common endocrine and metabolic disorder in women of reproductive age. It is characterized by ovulatory dysfunction, hyperandrogenism, and polycystic ovarian morphology, and it increases the risks of infertility, menstrual irregularity, and metabolic syndrome (, ). Subclinical hypothyroidism (SCH) is a common thyroid disorder defined by elevated serum thyroid-stimulating hormone (TSH) with normal free thyroxine (FT4) levels (). In recent years, the prevalence of comorbid PCOS and SCH has increased significantly in women of reproductive age due to changes in lifestyle and dietary patterns (, ). This comorbidity complicates clinical management and has sparked extensive debate over whether SCH is associated with more severe reproductive endocrine disturbances in patients with PCOS. To date, the quantitative association between SCH and core reproductive hormone disturbances in PCOS remains highly disputed across clinical observational studies and prior published meta-analyses (, ). The landmark 2017 meta-analysis by Pergialiotis et al. (), which is widely cited in clinical discussions of thyroid screening for PCOS patients, synthesized only literature published before 2016, evaluated only follicle-stimulating hormone (FSH), luteinizing hormone (LH), and sex hormone-binding globulin (SHBG) without a comprehensive pooled analysis of estradiol, prolactin, and total testosterone, and lacked stratified analysis based on clinically variable TSH diagnostic cutoffs or thyroid autoimmunity status. A 2018 meta-analysis by Ding et al. () focused exclusively on estimating the pooled prevalence of SCH among PCOS populations and contained no quantitative synthesis of reproductive hormone profiles; therefore, it could not address conflicting endocrine outcomes across primary studies. Another large-scale meta-analysis published by Xing et al. in 2021 () included 27 studies but relied heavily on Chinese gray literature. The study neither applied the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework for certainty rating nor conducted rigorous leave-one-out analysis or risk-of-bias-stratified sensitivity analysis to explore between-study heterogeneity. Other narrative reviews published in 2022 provided only qualitative summaries without pooled quantitative effect sizes, thus offering limited actionable guidance for clinical endocrine evaluation (). Two recent reviews identified major gaps in the existing evidence. A 2022 systematic review pooled global cross-sectional data and showed conflicting correlations between SCH and reproductive hormones across ethnic cohorts, with no updated quantitative synthesis to resolve these discrepancies (). A 2023 review outlined thyroid–ovarian axis crosstalk and identified thyroid autoantibody positivity and inconsistent TSH cutoffs as two key confounders driving high between-study heterogeneity; however, no meta-analysis had performed stratified quantitative analyses for these two factors at the time of publication (). Multiple new observational cohorts from various countries have been published between 2021 and 2026, all falling outside the search windows of the three prior meta-analyses, creating a substantial time lag in the current evidence base. Most original studies rely on single-center, small-sample designs without unified hormone testing protocols or standardized confounder adjustment. These inconsistencies fragment the available data and hinder clinicians from establishing standardized endocrine monitoring plans for PCOS patients with SCH. This systematic review and meta-analysis was therefore performed to integrate updated observational evidence and examine the associations between SCH and serum levels of FSH, LH, estradiol, prolactin, and total testosterone in women with PCOS. We aim to describe the current evidence on differences in hormone levels and identify methodological gaps requiring further investigation. 【Materials and methods】 【Study registration】 This systematic review and meta-analysis was designed and performed in accordance with the Cochrane Handbook for Systematic Reviews of Interventions. We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (). The review protocol was formally registered on 1 May 2026 in the International Prospective Register of Systematic Reviews (PROSPERO) (ID: CRD420261382326), publicly available at https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD420261382326. Database searching was completed on 25 April 2026, with only citation export performed at that stage. All subsequent screening, data extraction, subgroup analyses, and sensitivity analyses were performed according to the registered protocol after registration, with no post-hoc adjustments to outcomes or analytical plans after observing the pooled results. 【Inclusion and exclusion criteria】 The population, exposure, comparator, and outcomes (PECO) framework, which is commonly used for observational systematic reviews, was used to define the research question and eligibility criteria (). Eligible study designs included cross-sectional, case-control, and retrospective cohort studies. Only original studies reporting means and standard deviations (SDs) (or data that could be converted to means and SDs) for the above outcomes were included. Reviews, letters, conference abstracts, animal studies, and studies without extractable outcome data were excluded. PECO framework for study eligibility screening. The population, intervention, comparison, outcomes, and study design (PICOS) framework was developed for interventional trials and contains an intervention component unsuitable for this observational meta-analysis. PECO was applied herein to align with exposure-comparison study design, and all core eligibility dimensions of PICOS are integrated into the four PECO categories above. 【Search strategy】 Electronic searches were conducted in PubMed, Web of Science Core Collection, Embase, and Cochrane Library. All retrievals were completed on 25 April 2026 via each database’s official web interface and included all records from database inception to the cutoff date. Only peer-reviewed English-language journal articles were included. We restricted our analysis to English-language publications for three reasons: high-quality relevant observational studies are predominantly published in English; the use of a single language avoids translation bias and inconsistent data extraction; and non-English regional papers are mostly small, single-center studies with variable methodology that may introduce additional heterogeneity. Nevertheless, this rule may create language bias by omitting eligible non-English regional publications. Searches of trial registries and gray literature, including conference abstracts, dissertations, and preprints, and citation tracking were not performed. Unpublished or non-peer-reviewed materials usually lack complete outcome data and detailed methods, so we excluded these sources to improve the reliability of the pooled data. The search syntax integrated Medical Subject Headings (MeSH)/Emtree controlled terms and free-text keywords connected by Boolean operators. Full replicable search strings for every database are listed in . 【Literature screening and data extraction】 Literature screening and data extraction were independently performed by Yunhui Liu and Ping Zhao. All disagreements in screening judgments were first resolved through discussion; any remaining disagreements were adjudicated by the third reviewer, Tingting Deng. EndNote X9 was used to organize all retrieved citations and remove duplicates. The two reviewers separately screened titles, abstracts, and full texts offline according to predefined PECO eligibility criteria and then cross-checked their screening results. All extracted quantitative data were recorded in a standardized Excel form. Extracted items included: Basic study information: first author, publication year, country of recruitment, and study design type; Participant baseline characteristics: total study sample size, number of PCOS patients with SCH, age, body mass index (BMI), use of the Rotterdam PCOS diagnostic criteria, and specific TSH cutoff thresholds used to define SCH; Thyroid autoimmunity data: whether thyroid peroxidase antibody (TPOAb) and thyroglobulin antibody (TgAb) testing was performed, and whether reproductive hormone outcomes were separately reported for antibody-positive vs. antibody-negative SCH subgroups; Measured reproductive hormone indicators: FSH, LH, estradiol, prolactin, total testosterone; Raw quantitative outcome data: sample size, mean value, and standard deviation of each hormone for PCOS patients with SCH and euthyroid PCOS subgroups; Confounder adjustment records: matching or multivariable adjustment status for key covariates including age, BMI, insulin resistance, PCOS phenotype, thyroid autoimmunity, medication history, infertility status, menstrual phase of blood collection, and standardized hormone testing protocols. For studies that did not report standard deviation (SD) directly, values were estimated using methods recommended by the Cochrane Handbook from p-values, t-values, 95% confidence intervals (CIs), standard errors, or CI limits. If studies reported medians and interquartile ranges, data were converted to means and SDs using the methods of Luo et al. () and Wan et al. (). 【Risk of bias assessment】 Different quality assessment instruments were applied according to study design: cross-sectional studies were evaluated using the 9-item Agency for Healthcare Research and Quality (AHRQ) checklist, whereas case–control and cohort studies were appraised with the Newcastle–Ottawa Scale (NOS) (, ). The raw total scores from the AHRQ checklist and NOS served only as supplementary information rather than as the primary metric for presenting risk-of-bias results. To improve the interpretability of the clinical findings, we further conducted independent evaluations of all eligible studies across eight key clinical domains: ① Participant selection; ② Definition and measurement of subclinical hypothyroidism (SCH);③ Diagnostic criteria for polycystic ovary syndrome (PCOS); ④ Comparability between the SCH group and the euthyroid PCOS control group; ⑤ Adjustment or matching for confounding factors; ⑥ Hormone assays; ⑦ Timing of blood sampling for hormone measurement; ⑧ Completeness of outcome data. Each domain was categorized as having a low, moderate, or high risk of bias. 【Statistical analysis】 All statistical analyses were performed in R 4.3.1 with the meta, metafor, and dmetar packages. For continuous outcomes, the standardized mean difference (SMD) was adopted as the primary effect size alongside the mean difference (MD), with 95% CIs, to unify units across studies. A random-effects model with DerSimonian–Laird τ² estimation was prespecified for all analyses because of the inherent clinical heterogeneity of observational data; fixed-effects models served as sensitivity checks. Heterogeneity was assessed using Cochran’s Q and I² (I² > 50% = substantial). We conducted Hartung–Knapp-adjusted random-effects sensitivity analyses and calculated 95% prediction intervals (PIs) for all outcomes to estimate expected effect ranges in future studies, with all PIs reported in the Results. Three predefined SCH subgroups were planned: a low TSH cutoff (~2.5 mIU/L), an adult standard TSH cutoff (4.0–5.0 mIU/L), and antibody-positive SCH. Additional leave-one-out and risk-of-bias-stratified sensitivity analyses were performed to explore sources of heterogeneity. Egger’s and Begg’s tests were used to evaluate publication bias. Certainty was evaluated using a Grading of Recommendations Assessment, Development and Evaluation (GRADE)-like framework adapted to observational evidence. 【Results】 【Literature search results】 A total of 2,886 records were initially identified. After excluding 351 duplicates, 2535 records underwent title and abstract screening. Most ineligible articles were excluded because of inappropriate study types, irrelevant research subjects, and unrelated research themes, leaving 285 articles for full-text assessment. Further full-text screening excluded studies without SCH groups or valid reproductive hormone indicators. Seven studies with unextractable data were also excluded. Finally, 13 eligible studies were included in the systematic review and meta-analysis. The detailed screening process is shown in . Full details of representative excluded studies and specific reasons for exclusion are provided in . PRISMA flow diagram of the study selection process. 【Basic characteristics of the included literature】 A total of 13 cross-sectional, case-control, and retrospective cohort studies (–) from nine countries involving 3,114 women with PCOS (579 with SCH and 2,535 euthyroid PCOS controls) were included; all studies used the Rotterdam PCOS criteria, whereas SCH was diagnosed using variable TSH cutoffs ranging from 2.5 to 5.0 mIU/L with normal thyroid hormone levels. The reproductive hormones measured were FSH, LH, estradiol, prolactin, and total testosterone, and most studies matched participants only by age and BMI without multivariable adjustment for key confounders, including insulin resistance and PCOS phenotype. Two studies (, ) tested TPOAb/TgAb thyroid autoantibodies, and merely Gawron et al. () separately reported reproductive hormone results for antibody-positive autoimmune SCH and isolated non-autoimmune SCH subgroups; the other 11 studies did not detect thyroid autoantibodies or provide antibody-stratified hormonal data, so quantitative subgroup meta-analysis based on thyroid autoimmunity status could not be carried out due to insufficient available stratified evidence (). Basic characteristics of the literature. SCH, subclinical hypothyroidism; FT4, free thyroxine; T3, triiodothyronine; T4, thyroxine; TSH, thyroid-stimulating hormone; FSH, follicle-stimulating hormone; LH, luteinizing hormone; TPOAb, thyroid peroxidase antibody; TgAb, thyroglobulin antibody; NA, data not available; IQR, interquartile range; IR, insulin resistance; TAI, thyroid autoimmunity; Cycle, menstrual-cycle timing of blood sampling; Assay, standardized laboratory hormone assay methods. Abbreviations for confounding control status: M = Matched (baseline matching between SCH and euthyroid PCOS control groups); A = Adjusted (multivariable regression adjustment in statistical analyses); NR = Not reported/no matching or adjustment performed for the corresponding variable. 【Quality evaluation of literature】 and present the domain-stratified risk-of-bias results of all included studies based on eight predefined evaluation dimensions, with original AHRQ total scores listed as supplementary evidence. Uniform low risk of bias was observed across all studies for D2 (SCH definition and measurement), D3 (PCOS diagnostic criteria), D6 (Hormone assay methods), and D8 (Completeness of outcome data), as all articles clearly reported unified diagnostic thresholds and complete laboratory testing information without missing core outcome data. The major methodological flaws were concentrated in D4 (Group comparability) and D5 (Confounder adjustment): several studies failed to balance baseline characteristics between groups or adjust for key confounding factors such as BMI and age, resulting in a high risk of bias. Moderate risk frequently appeared in D1 (Participant selection) and D7 (Timing of hormone measurement), caused by non-probability sampling and inadequate records of hormone blood collection cycles. Overall, seven studies had low overall bias risk, five had moderate risk, and the study by Fatima () was classified as high risk owing to dual high-risk ratings in group comparability and confounding control. Stratified risk-of-bias assessment by dimension of included studies. Definitions of eight evaluation domains: D1: Participant selection; D2: SCH definition and measurement; D3: PCOS diagnostic criteria; D4: Group comparability; D5: Confounder adjustment; D6: Hormone assay methods; D7: Timing of hormone measurement; D8: Completeness of outcome data. Domain-specific risk-of-bias assessment of the included observational studies. Definitions of eight evaluation domains: D1: Participant selection; D2: SCH definition and measurement; D3: PCOS diagnostic criteria; D4: Group comparability; D5: Confounder adjustment; D6: Hormone assay methods; D7: Timing of hormone measurement; D8: Completeness of outcome data. 【Comparison of serum hormone levels between PCOS patients with and without SCH】 【FSH levels】 A total of 12 studies (–, –) were included. The analysis involved 523 PCOS patients with SCH and 1,986 euthyroid PCOS controls. The random-effects model revealed no significant difference in serum FSH levels between groups (SMD = 0.038, 95% CI: −0.095 to 0.170, 95% PI: −0.28 to 0.37, p = 0.581) (). Moderate heterogeneity was detected (I² = 36.4%, Q-test p = 0.099; between-study variance τ² = 0.018). Leave-one-out analysis indicated the study by Dai et al. () as the main contributor to heterogeneity, which was resolved after its exclusion. The pooled outcome remained unchanged after sequential omission of individual studies (). Sensitivity analysis based on the Hartung–Knapp-adjusted random-effects model generated consistent pooled effect sizes and overlapping CIs, with no change to the overall statistical inference. According to Cohen’s d criteria, the pooled SMD of 0.038 represents a negligible effect size (|SMD| < 0.2), suggesting that any observed difference in FSH concentrations is unlikely to be clinically meaningful for ovarian reserve evaluation based on current observational evidence. Meta-analysis of FSH levels in patients with PCOS with and without SCH. Leave-one-out sensitivity analysis of FSH levels. 【