文献分析报告新方案
请您从医学角度把关

约 10–15 分钟 · 您的选择会自动保存在本机，关掉再打开还在
背景：我们的软件会给医学论文自动出一份分析报告。现在的版本常常套用同一句模板，还会误判（例如把“未来需要随机试验”当成本研究做了随机分组）。
新方案不用大模型，先判断论文类型，再按规则给出判断，每条都附原文句子和页码。我们用您提供的 19 篇论文试跑了一遍（扫描版、重复的一篇和附件未列入）。

想请您回答三件事：
  1. 小模型改写可靠吗？我们试过让 0.5B 级小模型把规则句子改得更通顺。下面 6 题请先判断意思变没变，再看我们的答案。
  2. 新方案的医学判断对不对？每篇下面每条判断点“对 / 不对 / 拿不准”，不对的请写一句原因。挑您熟悉的几篇看即可。
  3. 最后两个问题：您能否接受语气上的轻微变化，以及报告里最该提示什么。
看完点右下角「复制反馈」，粘贴到微信发给我们即可。

一、小模型改写：意思变了吗？

请先选，再看我们的判断。
第 1 题 · Wang 2022 中国中老年四分型
原文

作者在 1,130 名中国患者中用 k-means基于 7 个变量聚类，得到 4 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。

小模型改写后

作者在 1,130 名中国患者中使用 K-means 方法，基于 7 个变量进行聚类，得出 4 个亚组。亚组是根据变量相似度划分的，而非按病因划分。本文沿用了既有的分型方法（复现的是方法），但未在本文之外发现独立队列验证的表述；亚组结构可能依赖于本样本。亚组与并发症/结局的差异源于随访关联，表明亚组具有预后区分度；不说明按亚组治疗能改善结局。

两段意思一样吗？
我们的判断：意思没变：只换了近义词（不是→而非、没找到→未发现）。
第 2 题 · zaharia2019 德国 GDS
原文

作者聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。作者报告在其他队列中重复得到相似亚组；重复的是亚组结构，不等于各亚组比例一致。亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。

小模型改写后

作者聚类，得到5个亚组。亚组是根据变量相似度划分的，而非按病因划分的。在其他队列中，重复得到的亚组结构与各亚组的比例不一致。亚组与并发症/结局的差异源于随访关联，表明亚组具有预后区分度；不说明按亚组治疗能改善结局。

两段意思一样吗？
我们的判断：意思变了：原文是“不等于比例一致”（不保证一致），改写后变成“比例不一致”（断言不一致）。程序校验没有拦住。
第 3 题 · Anjana 2020 INSPIRED印度INSPIRED分型
原文

作者用 k-means基于 6 个变量（age at diagnosis, BMI, waist circumference, HbA1c, serum triglycerides and serum HDL chol…）聚类，得到 4 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。

小模型改写后

作者基于6个变量（年龄诊断、BMI、腰围、HbA1c、血清甘油三酯和血清HDL胆固醇）进行聚类，得出4个亚组。亚组是根据变量相似度划分的，而非根据病因划分的。本文沿用了既有的分型方法（复现的是方法），但未在本文之外的独立队列中验证其有效性；亚组结构可能依赖于本样本。亚组与并发症/结局的差异源于随访关联，表明亚组具有预后区分度；不说明按亚组治疗能改善结局。

两段意思一样吗？
我们的判断：有错译：把英文变量 age at diagnosis 译成“年龄诊断”，还凭空加了“验证其有效性”。
第 4 题 · 2019 纪立农 Novel subgroups of patients with adult-onset diabetes in Chinese and US populations
原文

全文没找到在独立队列中重复验证的表述；亚组结构可能依赖本样本。横断面资料只能描述各亚组的特征差异，不能说明亚组之间预后不同，也不能说明按亚组选药更好。样本来自中国、美国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。

小模型改写后

全文未发现独立队列中重复验证的表述；亚组结构可能依赖本样本。横断面资料仅能描述各亚组的特征差异，无法说明亚组间预后差异，也无法说明按亚组选药是否更优。样本来自中国和美国人群；若换到其他人群，亚组比例、切点和结局差异可能均有所不同。

两段意思一样吗？
我们的判断：意思没变：只能→仅能、不能→无法，读起来更顺。
第 5 题 · 2025 UAB多族裔验证
原文

作者用 层次聚类基于 6 个变量聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。作者报告在其他队列中重复得到相似亚组；重复的是亚组结构，不等于各亚组比例一致。亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。

小模型改写后

作者基于6个变量进行层次聚类，得出5个亚组。亚组是根据变量相似度划分的，而非按病因划分的。在其他队列中，重复得到的亚组结构与各亚组比例不一致，而亚组与并发症/结局的差异主要源于随访关联，表明亚组具有预后区分度，但不说明按亚组治疗能改善结局。

两段意思一样吗？
我们的判断：意思变了：和第 2 题同样的问题，“不等于比例一致”→“比例不一致”，还多了“主要”一词。
第 6 题 · 2025 19项研究荟萃分析
原文

合并结果的异质性很高（I² 最高 100%）；合并值只能当作粗略范围，不宜当作一个确定数字引用。

小模型改写后

合并结果的异质性较高（I² 最高 100%），合并值只能作为粗略范围，不宜当作一个确定数字引用。

两段意思一样吗？
我们的判断：语气变弱：“很高”改成了“较高”，数字没变。

二、逐篇：新方案的判断对不对？

点开一篇，逐条判断。只看熟悉的几篇也很有帮助。
1979 NDDG分类分类 / 诊断标准
现在软件给出的总结

当前分析基于PDF 全文提取，主要用于定位原文中的结果、方法与证据句。原文结果部分包含统计学数据，但自动结构化提取未达到足够置信度，建议结合原文核对。 具体结论及其适用范围仍应以原…

新流程的判断（请逐条看对不对）
这是委员会制定的分类 / 诊断标准，结论来自专家共识和当时的证据，不是一项研究的结果；使用时注意是否已被后续版本修订。
看原文依据
  1. NATIONAL DIABETES DATA GROUP第 11 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. mg/dl probably indicate a degree of abnormal glucose tolerance that has not been assessed fully as yet.第 14 页
1997 ADA病因学分类分类 / 诊断标准
现在软件给出的总结

当前分析基于PDF 全文提取，主要用于定位原文中的结果、方法与证据句。原文结果部分包含统计学数据，但自动结构化提取未达到足够置信度，建议结合原文核对。 具体结论及其适用范围仍应以原…

新流程的判断（请逐条看对不对）
这是委员会制定的分类 / 诊断标准，结论来自专家共识和当时的证据，不是一项研究的结果；使用时注意是否已被后续版本修订。
看原文依据
  1. National Diabetes Data Group: Classification and diagnosis of diabetes mellitus and other categories of glucose intolerance.第 12 页
2003 KPD Aβ分类分类 / 诊断标准
现在软件给出的总结

从当前自动化提炼结果看，这是一项以"研究类型待进一步确认"为核心设计、且 当前取得全文文本（不等于证据质量较高） 的研究。最值得保留的主结论是：原文结果部分包含统计学数据，但自动结…

新流程的判断（请逐条看对不对）
这是委员会制定的分类 / 诊断标准，结论来自专家共识和当时的证据，不是一项研究的结果；使用时注意是否已被后续版本修订。
作者原话
作者结论（原文）：
看原文依据
  1. In conclusion, we have used a heterogeneous, multiethnic第 7 页
Ahlqvist 2018 瑞典五分类数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

现在软件的误判

判为“随机分组” ← This clustering also paves the way for randomised trials targeting insulin secretion in SIDD and insulin resistance in SIRD.

判为“有盲法” ← Whereas SAID overlapped with type 1 diabetes and LADA, SIDD and SIRD represent two new, severe forms of diabetes previously masked within type 2 diabetes.

新流程的判断（请逐条看对不对）
作者在 8,980 名瑞典患者中用 k-means + 层次聚类 + 两步聚类基于 6 个变量（glutamate decarboxylase antibodies, age at diagnosis, BMI, HbA1c, and homoeostatic model …）聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Findings We identified five replicable clusters of patients with diabetes, which had significantly different patient characteristics and risk of diabetic complications.第 1 页
  2. Clusters were based on six variables (glutamate decarboxylase antibodies, age at diagnosis, BMI, HbA1c, and homoeostatic model assessment 2 estimates of β-cell function and insulin resistance), and were related to prospective data from patient records on deve…第 1 页
  3. To classify patients into novel diabetes subgroups, first we used the TwoStep clustering method in 8980 patients in the ANDIS cohort with complete data available for the clustering variables.第 4 页
  4. Methods We did data-driven cluster analysis (k-means and hierarchical clustering) in patients with newly diagnosed diabetes (n=8980) from the Swedish All New Diabetics in Scania cohort.第 1 页
分型要用到 GADA 等检测；门诊若没有这些指标，不能直接照搬这套分型。
看原文依据
  1. Clusters were based on six variables (glutamate decarboxylase antibodies, age at diagnosis, BMI, HbA1c, and homoeostatic model assessment 2 estimates of β-cell function and insulin resistance), and were related to prospective data from patient records on deve…第 1 页
作者报告在其他队列中重复得到相似亚组（文中提到的队列：ANDIS、SDR、ANDIU、DIREVA）；重复的是亚组结构，不等于各亚组比例一致。
看原文依据
  1. Replication was done in three independent cohorts: the Scania Diabetes Registry (n=1466), All New Diabetics in Uppsala (n=844), and Diabetes Registry Vaasa (n=3485).第 1 页
  2. We used three independent cohorts to replicate the clustering: SDR (n=1466), ANDIU (n=844), and DIREVA (n=3485).第 5 页
  3. Therefore, we assigned patients in replication cohorts to clusters on the basis of which cluster they were most similar to, calculated as their Euclidian第 5 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. Cox regression and logistic regression were used to compare time to medication, time to reaching the treatment goal, and risk of diabetic complications and genetic associations.第 1 页
样本来自瑞典人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Methods We did data-driven cluster analysis (k-means and hierarchical clustering) in patients with newly diagnosed diabetes (n=8980) from the Swedish All New Diabetics in Scania cohort.第 1 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. Evidence before this study National guidelines maintain information about diabetes classification, but this classification has not been much updated during the past 20 years, and very few attempts have been made to explore heterogeneity of type 2 diabetes.第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. Interpretation We stratified patients into five subgroups with differing disease progression and risk of diabetic complications.第 1 页
全文没找到作者自述的局限段落（可能是分节错误或原文确实没写）。
纠正原软件
原分析把下面这句当作“本研究采用随机化”，实为展望或引用他人试验的句子，新流程不采纳。
看原文依据
  1. This clustering also paves the way for randomised trials targeting insulin secretion in SIDD and insulin resistance in SIRD.第 undefined 页
纠正原软件
原分析把下面这句当作“盲法”，句中的 masked 是“被掩盖”的意思，与盲法无关，新流程不采纳。
看原文依据
  1. Whereas SAID overlapped with type 1 diabetes and LADA, SIDD and SIRD represent two new, severe forms of diabetes previously masked within type 2 diabetes.第 undefined 页
WHO 2019 Classification of Diabetes Mellitus分类 / 诊断标准
现在软件给出的总结

从当前自动化提炼结果看，这是一项以"研究类型待进一步确认"为核心设计、且 当前取得全文文本（不等于证据质量较高） 的研究。最值得保留的主结论是：原文结果部分包含统计学数据，但自动结…

新流程的判断（请逐条看对不对）
这是委员会制定的分类 / 诊断标准，结论来自专家共识和当时的证据，不是一项研究的结果；使用时注意是否已被后续版本修订。
看原文依据
  1. With this in mind, the Expert group considered it best to define a classification system that prioritizes第 13 页
文中给出 3 处诊断切点（见依据）；切点是标准的一部分，不同版本之间可能不同。
看原文依据
  1. Diagnostic criteria for diabetes: fasting plasma glucose ≥ 7.0 mmol/L or 2-hour post-load plasma glucose ≥ 11.1 mmol/L or Hba1c ≥ 48 mmol/mol Diagnostic criteria for gestational diabetes: fasting plasma glucose 5.1–6.9 mmol/L or 1-hour post-load plasma glucos…第 6 页
  2. plasma glucose values of ≥ 7.0 mmol/L (126 mg/dl), 2-h post-load plasma glucose ≥ 11.1 mmol/L第 8 页
  3. (200 mg/dl) (5), HbA1c ≥ 6.5% (48 mmol/mol); or a random blood glucose ≥ 11.1 mmol/L (200 mg/第 8 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. type 1 and type 2 diabetes and dropped MRDM because of lack of evidence to support its existence第 12 页
2020单基因糖尿病综述
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
这是综述，不产生新数据；文中的比例和结论来自所引用的研究，引用时应回到原始研究核对。
全文没找到作者自述的局限段落（可能是分节错误或原文确实没写）。
Anjana 2020 INSPIRED印度INSPIRED分型数据驱动分型
现在软件给出的总结

从当前自动化提炼结果看，这是一项以"研究类型待进一步确认"为核心设计、且 当前取得全文文本（不等于证据质量较高） 的研究。最值得保留的主结论是：原文结果部分包含统计学数据，但自动结…

新流程的判断（请逐条看对不对）
作者用 k-means基于 6 个变量（age at diagnosis, BMI, waist circumference, HbA1c, serum triglycerides and serum HDL chol…）聚类，得到 4 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Results We identified four clusters of patients, differing in phenotypic characteristics as well as disease outcomes: cluster 1 (Severe Insulin Deficient Diabetes, SIDD), cluster 2 (Insulin Resistant Obese Diabetes, IROD), cluster 3 (Combined Insulin Resistan…第 2 页
  2. For purposes of this validation study, we selected 3851 individuals with type 2 diabetes from the INDIAB study population in 15 states of India and clustering was performed using the following six variables: age at diagnosis, BMI, waist circumference, HbA1c, …第 4 页
  3. Research design and methods From a network of 50 diabetes centers across nine states of India, we selected 19 084 individuals with type 2 diabetes (aged 10–97 years) with diabetes duration of less than 5 years at the time of first clinic visit and performed k…第 2 页
本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。
看原文依据
  1. We applied the k-means clustering (k=4) on the Asian Indian population (N=19 084) based on these five variables and analyzed the optimal cluster number based on the silhouette width method.第 4 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. Cox proportional hazards showed that SIDD had the highest hazards for developing retinopathy, followed by CIRDD, while CIRDD had the highest hazards for kidney disease.第 2 页
样本来自印度人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Research design and methods From a network of 50 diabetes centers across nine states of India, we selected 19 084 individuals with type 2 diabetes (aged 10–97 years) with diabetes duration of less than 5 years at the time of first clinic visit and performed k…第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. Conclusions Compared with previously reported clustering, we show two novel subgroups of type 2 diabetes in the Asian Indian population with important implications for prognosis and management.第 2 页
全文没找到作者自述的局限段落（可能是分节错误或原文确实没写）。
lin Sun 2021 中国两步七分型数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
作者用 层次聚类 + 两步聚类基于 6 个变量（GADA, age at diagnosis, BMI, HbA1c, HOMA2-B and HOMA2-IR）聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. When other variables were added to the cluster analysis, seven subgroups were identified, including five clusters of the European classification and two novel subgroups, namely, uric acid-related diabetes and inheritance-related diabetes.第 1 页
  2. Validating the applicability of the European classification in Chinese populations with type 2 diabetes Two-step cluster analysis was carried out in patients with type 2 diabetes, including 3,087 men and 2,327 women, with a focus on six variables (GADA, age a…第 3 页
  3. It identifies clusters by two processes: first, preclustering, followed by hierarchical clustering.第 3 页
分型要用到 GADA、HOMA 等检测；门诊若没有这些指标，不能直接照搬这套分型。
看原文依据
  1. Validating the applicability of the European classification in Chinese populations with type 2 diabetes Two-step cluster analysis was carried out in patients with type 2 diabetes, including 3,087 men and 2,327 women, with a focus on six variables (GADA, age a…第 3 页
本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。
看原文依据
  1. We also used TC instead of TG to do the clustering analysis.第 9 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. Furthermore, triglycerides and uric acid were added to refine the cluster analysis, and Cox regression was used to evaluate the risk of diabetic complications.第 1 页
样本来自中国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Aims/Introduction: The aim of this study was to determine whether distinct subphenotypes of patients with type 2 diabetes in the European classification exist in Chinese populations, and to further establish novel subphenotypes more suitable for Chinese popul…第 1 页
文中有 6 处 HR/OR/RR 类估计，均来自观察性比较，只能读作关联，不能读作“某亚组导致了某结局”。
看原文依据
  1. Risk evaluation of diabetic complications in seven clusters Compared with the patients in cluster 7 (IRD), the patients in cluster 1 (SIRD) showed a higher risk of DR (HR 1.3390, 95% confidence interval [CI] 0.919–1.95, P = 0.128; Figure 4a, Table S3), DPN (H…第 4 页
只看到相对风险（HR/OR），没找到绝对发生率或每千人年事件数；相对风险大不等于实际多出很多病例。
看原文依据
  1. Risk evaluation of diabetic complications in seven clusters Compared with the patients in cluster 7 (IRD), the patients in cluster 1 (SIRD) showed a higher risk of DR (HR 1.3390, 95% confidence interval [CI] 0.919–1.95, P = 0.128; Figure 4a, Table S3), DPN (H…第 4 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. Due to differences in ethnicities, regions, genetic backgrounds, lifestyles and natural environment, it is unclear whether the European subclassification is applicable to Chinese populations.第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. Conclusions: Seven subgroups of type 2 diabetes were identified in Chinese populations, with distinct characteristics and disparate clinical outcomes.第 1 页
作者原话
作者自述局限 6 句，主题：外推性、病程混杂、偏倚。
看原文依据
  1. Several limitations need to be mentioned.第 11 页
  2. First, patients with a long duration of diabetes were included in the present study, and seven clusters should also be shown in newly diagnosed patients and heterogeneous populations.第 11 页
  3. Second, genetic associations in this sample were not shown.第 11 页
  4. Furthermore, the duration of diabetes was determined from the natural course of disease, and prospective studies need to be carried out in the future.第 11 页
  5. Furthermore, there was a measurement bias of the duration of diabetes mellitus that many patients develop clinical manifestations of diabetes mellitus before diagnosis of diabetes mellitus.第 11 页
  6. Additionally, parametric survival models might be more appropriate than Cox regression to examine the associations.第 11 页
Wang 2022 中国中老年四分型数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
作者在 1,130 名中国患者中用 k-means基于 7 个变量聚类，得到 4 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Participants were classified into 4 diabetes subgroups based on 7 variables measured at baseline (Figure 1 and Supplementary Material).第 4 页
  2. In our study, K-means cluster analysis was performed with 7 variables including BMI, age at diagnosis of diabetes, FPG, 2hPG, HbA1c, HOMA-b, and HOMA-IR at baseline.第 3 页
  3. 688 in 1130 participants with complete cluster variable data underwent the follow-up interview.第 4 页
  4. In our study, K-means cluster analysis was performed with 7 variables including BMI, age at diagnosis of diabetes, FPG, 2hPG, HbA1c, HOMA-b, and HOMA-IR at baseline.第 3 页
分型要用到 HOMA 等检测；门诊若没有这些指标，不能直接照搬这套分型。
看原文依据
  1. In our study, K-means cluster analysis was performed with 7 variables including BMI, age at diagnosis of diabetes, FPG, 2hPG, HbA1c, HOMA-b, and HOMA-IR at baseline.第 3 页
作者说明部分患者缺少自身抗体（GADA）结果，自身免疫型可能混入其他亚组。
看原文依据
  1. Firstly, parameters such as the glutamic acid decarboxylase antibodies (GADA) were not measured or included in the cluster analysis in the current study.第 8 页
本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。
看原文依据
  1. We replicated the clustering approach used by Ahlqvist et al.第 3 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. In the current study, we aimed to examine the novel subgroups of diabetes in a cohort of middle-aged and elderly Chinese using variables reported in previous studies plus FPG and 2hPG at baseline (6), to evaluate the changes of specific metabolic markers duri…第 2 页
样本来自中国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Using data from a prospective cohort study of Chinese community residents aged ≥40 years, we were able to identify four subgroups of T2DM based on the 7 variables measured at baseline and used in the cluster analysis, including age at diagnosis, BMI, FPG, 2hP…第 4 页
文中有 4 处 HR/OR/RR 类估计，均来自观察性比较，只能读作关联，不能读作“某亚组导致了某结局”。
看原文依据
  1. The risk of developing abnormal ABI was higher in cluster SOIRD (OR 3.46; 95% CI 1.54-8.04) compared with MARD.第 4 页
只看到相对风险（HR/OR），没找到绝对发生率或每千人年事件数；相对风险大不等于实际多出很多病例。
看原文依据
  1. The risk of developing abnormal ABI was higher in cluster SOIRD (OR 3.46; 95% CI 1.54-8.04) compared with MARD.第 4 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. Furthermore, most reports were cross-sectional (7, 14) and few studies have examined such classification method among Chinese T2DM patients.第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. In conclusion, the middle-aged and elderly Chinese adults with newly-diagnosed diabetes can be allocated to specific clusters.第 8 页
作者原话
作者自述局限 6 句，主题：自身抗体缺失或不全、外推性、样本量有限、随访不足。
看原文依据
  1. Our study has several limitations.第 8 页
  2. Firstly, parameters such as the glutamic acid decarboxylase antibodies (GADA) were not measured or included in the cluster analysis in the current study.第 8 页
  3. However, the prevalence of being GADA-positive could be less than 5.9% in population-based screening of adult-onset T2DM in China (7).第 8 页
  4. Secondly, the follow-up duration was short and the limited numbers of incident macrovascular and microvascular diseases might not be able to provide sufficient statistical power to detect a true difference, therefore a further examination of第 8 页
  5. these diabetic complications were not conducted.第 8 页
  6. Thirdly, our study was conducted in community residents in Shanghai, China, which had limited extensibility to other populations.第 8 页
2019 纪立农 Novel subgroups of patients with adult-onset diabetes in Chinese and US populations数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
全文没找到在独立队列中重复验证的表述；亚组结构可能依赖本样本。
横断面资料只能描述各亚组的特征差异，不能说明亚组之间预后不同，也不能说明按亚组选药更好。
看原文依据
  1. Frequency, immunogenetics, and clinical characteristics of latent autoimmune diabetes in China (LADA China study): a nationwide, multicenter, clinic-based cross-sectional study.第 3 页
样本来自中国、美国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. The characteristics of participants in each cluster were similar across all ethnicities, except for the participants in the severe insulin-resistant diabetes cluster, who were young in the Chinese dataset but old in the American and European datasets.第 1 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. However, since all cohorts tested were from registry-based northern European studies, the utility of this new classification in Chinese and US populations is unclear.第 1 页
作者原话
作者自述局限 7 句，主题：自身抗体缺失或不全、外推性。
看原文依据
  1. Inclusion of this population is important for diabetes control especially in middle-income countries such as China, where nearly two thirds of patients with diabetes are unware of their condition unless screened.2 A limitation of our research was that we were…第 1 页
  2. Another limitation was that we had to choose mean plasma glucose, HOMA2-Bʹ, and HOMA2-IRʹ as surrogates for HbA1c, HOMA2-B, and HOMA2-IR because第 1 页
  3. For more on NHANES III see https://wwwn.cdc.gov/nchs/ nhanes/nhanes3/Default.aspx ## Correspondence第 1 页
  4. www.thelancet.com/diabetes-endocrinology Vol 7 January 2019第 2 页
  5. Figure: Participant distribution and cluster第 2 页
  6. characteristics (A) Participant distribution and cluster characteristics in the China National Diabetes and Metabolic Disorders Study第 2 页
  7. (B) Participant distribution and cluster characteristics in the 1988–94 National Health第 2 页
2020 周智广 中国15,772例数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
作者在 4,351 名瑞典、中国患者中用 k-means + 两步聚类基于 3 个变量聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Design: Patients were classified into 5 subgroups by K-means and Two-Step methods according to 6 clinical parameters.第 1 页
  2. Kahkoska et al stratified 20 274 patients with long-term T2DM, recruited from 3 cohorts in the United States, into 4 subgroups using 3 variables (age, BMI, and第 2 页
  3. Dennis et al repeated the same data-driven cluster analysis in 4351 patients with diabetes enrolled in the A Diabetes Outcome Progression Trial (ADOPT) study in the United Kingdom.第 2 页
  4. Design: Patients were classified into 5 subgroups by K-means and Two-Step methods according to 6 clinical parameters.第 1 页
作者说明部分患者缺少自身抗体（GADA）结果，自身免疫型可能混入其他亚组。
看原文依据
  1. First, 5 continuous variables, ie, BMI, age at onset of diabetes, HbA1c, HOMA2-B and HOMA2-IR, and a binary variable denoting the presence or absence of GADA were chosen as model variables.第 3 页
本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。
看原文依据
  1. We replicated the 2 clustering methods utilized by Ahlqvist et al (9).第 3 页
横断面资料只能描述各亚组的特征差异，不能说明亚组之间预后不同，也不能说明按亚组选药更好。
看原文依据
  1. Our study repeated the same data-driven cluster analysis in a nationwide, multicenter, cross-sectional study of Chinese newly diagnosed adult-onset diabetic patients to validate its usefulness in Chinese adults with newly diagnosed diabetes.第 3 页
样本来自瑞典、中国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Objective: This study aimed to validate the practicality of the Swedish diabetes regrouping scheme in Chinese adults with newly diagnosed diabetes.第 1 页
全文约 37 个 P 值，没找到多重比较校正的说明；个别“P<0.05”可能是偶然结果。
作者原话
作者结论（原文）：
看原文依据
  1. Conclusion: The Swedish diabetes regrouping scheme is applicable to adult-onset diabetes in China, with a high proportion of patients with the severe insulin deficient diabetes.第 2 页
作者原话
作者自述局限 9 句，主题：外推性、自身抗体缺失或不全、随访不足、病程混杂。
看原文依据
  1. Our study has some strengths and limitations.第 8 页
  2. First, a strength of our survey was that it was a large, multicenter survey covering the major regions of China, with urbanization being taken into account.第 8 页
  3. Therefore, our survey had better representation of Chinese adults with diabetes than the previous study (13).第 8 页
  4. Second, GADA was measured in all patients in our survey, so that a full-scale validation was feasible.第 8 页
  5. Our study has several limitations.第 8 页
  6. First, our study did not follow up these patients, and diabetes complications were not available for the current analysis.第 8 页
  7. Preliminary data of CVD risk estimated using the Framingham CVD risk score suggested that the Swedish regrouping strategy might correctly stratify the risk of CVD among Chinese adults with newly diagnosed diabetes.第 8 页
  8. Second, we did not detect the expression levels at genetic loci in patients; therefore, we were unable to analyze genetic differences among the 5 clusters.第 8 页
  9. Third, other antibodies were not measured and only GADA was used to identify autoimmune diabetic patients, which may lead to underestimation of the proportion of patients belonging to cluster 1 (11).第 8 页
2024 母义明 中国社区数据驱动分型
现在软件给出的总结

从当前自动化提炼结果看，这是一项以"研究类型待进一步确认"为核心设计、且 当前取得全文文本（不等于证据质量较高） 的研究。最值得保留的主结论是：原文结果部分包含统计学数据，但自动结…

新流程的判断（请逐条看对不对）
作者在 13,364 名中国患者中用 k-means基于 5 个变量（Figure 1 and Table 1）聚类，得到 4 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. The subjects were classified into four diabetes subgroups based on five variables (Figure 1 and Table 1).第 5 页
  2. The subjects were classified into four diabetes subgroups based on five variables (Figure 1 and Table 1).第 5 页
  3. First, we included a total of 13 364 people who met the American Diabetes Association's (ADA) criteria for diabetes (FBG ≥7.0 mmol/L or PBG ≥11.1 mmol/L or第 3 页
  4. Methods: We used K-means cluster analysis in 6369 newly diagnosed diabetic patients from eight centers of the REACTION (Risk Evaluation of cAncers in第 2 页
作者说明部分患者缺少自身抗体（GADA）结果，自身免疫型可能混入其他亚组。
看原文依据
  1. First, GADA were not measured or included in the cluster analysis in the present study.第 12 页
全文没找到在独立队列中重复验证的表述；亚组结构可能依赖本样本。
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. The K-means cluster analysis was performed with the same clustering parameters (age, BMI, HbA1c, HOMA- IR, and HOMA-β) as the study by Ahlqvist et al.9 Box– Cox transformations were performed for variables not in line with normal distribution.26,27 All values…第 4 页
样本来自中国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. At present, China has become the country with the largest number of diabetic populations.6 It was reported that China's diabetes populations accounted for 25% of the world's total diabetes populations19 Moreover, Chinese diabetic patients have distinctive cha…第 3 页
文中有 1 处 HR/OR/RR 类估计，均来自观察性比较，只能读作关联，不能读作“某亚组导致了某结局”。
看原文依据
  1. In model 3, after adjusting for the confounding factors, populations in SOIRD (OR, 1.573; 95% CI, 1.105– 2.139; p = 0.011) and MARD (OR, 1.564; 95% CI, 1.120–2.185; p = 0.009) subgroups showed higher第 6 页
只看到相对风险（HR/OR），没找到绝对发生率或每千人年事件数；相对风险大不等于实际多出很多病例。
看原文依据
  1. In model 3, after adjusting for the confounding factors, populations in SOIRD (OR, 1.573; 95% CI, 1.105– 2.139; p = 0.011) and MARD (OR, 1.564; 95% CI, 1.120–2.185; p = 0.009) subgroups showed higher第 6 页
作者原话
作者结论（原文）：
看原文依据
  1. Conclusion: The data-driven approach to differentiating the status of newonset diabetes in the Chinese community was feasible.第 3 页
全文没找到作者自述的局限段落（可能是分节错误或原文确实没写）。
zaharia2019 德国 GDS数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
作者聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Findings 1105 patients were classified at baseline into five clusters, with 386 (35%) assigned to mild age-related diabetes (MARD), 323 (29%) to mild obesity-related diabetes (MOD), 247 (22%) to severe autoimmune diabetes (SAID), 121 (11%) to severe insulin-r…第 1 页
作者报告在其他队列中重复得到相似亚组；重复的是亚组结构，不等于各亚组比例一致。
看原文依据
  1. Added value of this study In our study, we did comprehensive phenotyping of an independent cohort of patients at the time of diagnosis, with reassessment after 5 years.第 2 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. Patients suggested to be insulin-resistant were at higher risk of diabetic nephropathy and—to some extent—cardiovascular diseases.第 2 页
样本来自德国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. patients with newly diagnosed diabetes in the German Diabetes Study cohort were included in our study at baseline.第 3 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. Although some evidence already exists for an association between cluster assignment and risk for nephropathy and cardiovascular diseases,1 risk stratification for diabetic neuropathy and non-alcoholic fatty liver disease (NAFLD) using appropriate measurements…第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. Interpretation Cluster analysis can characterise cohorts with different degrees of whole-body and adipose-tissue insulin resistance.第 1 页
全文没找到作者自述的局限段落（可能是分节错误或原文确实没写）。
2024 挪威 HUNT数据驱动分型
现在软件给出的总结

从当前自动化提炼结果看，这是一项以"研究类型待进一步确认"为核心设计、且 当前取得全文文本（不等于证据质量较高） 的研究。最值得保留的主结论是：原文结果部分包含统计学数据，但自动结…

新流程的判断（请逐条看对不对）
作者说明部分患者缺少自身抗体（GADA）结果，自身免疫型可能混入其他亚组。
看原文依据
  1. Statistical analysis The cluster analysis was performed as in the study by Ahlqvist et al.1 First, participants with elevated GAD antibodies were assigned to the severe autoimmune diabetes (SAID) subgroup and were not included in the cluster analysis; this su…第 3 页
本文沿用了既有的分型方法（复现的是方法），但没找到在本文之外的独立队列中验证的表述；亚组结构可能依赖本样本。
看原文依据
  1. We used the elbow method, the gap statistic, and the mean silhouette width to assess the optimal number of clusters in our data.17–19 We did not find evidence of a clear optimal number of clusters (online supplemental figure 2), and thus instructed the cluste…第 3 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. We used Cox proportional hazards models to investigate associations between clusters and risks of vascular complications and mortality.第 1 页
样本来自挪威人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Research design and methods Participants from the second (HUNT2, 1995–1997) and third (HUNT3, 2006– 2008) surveys of the Norwegian population-based Trøndelag Health Study (HUNT Study) with adult-onset diabetes were included (n=1899).第 1 页
文中有 1 处 HR/OR/RR 类估计，均来自观察性比较，只能读作关联，不能读作“某亚组导致了某结局”。
看原文依据
  1. The incidence of retinopathy was higher in the SAID (HR 1.76 (95% CI 1.28 to 2.42)) and SIDD (HR 1.53 (95% CI 1.17 to 2.01)) subgroups, compared with the MARD subgroup (figure 2 and online supplemental table 2).第 4 页
只看到相对风险（HR/OR），没找到绝对发生率或每千人年事件数；相对风险大不等于实际多出很多病例。
看原文依据
  1. The incidence of retinopathy was higher in the SAID (HR 1.76 (95% CI 1.28 to 2.42)) and SIDD (HR 1.53 (95% CI 1.17 to 2.01)) subgroups, compared with the MARD subgroup (figure 2 and online supplemental table 2).第 4 页
作者原话
作者指出的研究空白（原文）：
看原文依据
  1. Although the original study mainly focused on populations of newly diagnosed diabetes, they found similar clusters in a cohort of participants with longer durations of diabetes, and suggested that the subgroups could be stable over time.1 It has been postulat…第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. Conclusions Cluster assignment did not provide better prediction of vascular complications or all-cause mortality compared with established risk factors.第 1 页
作者原话
作者自述局限 6 句，主题：胰岛功能指标不全、治疗混杂、随访不足、外推性。
看原文依据
  1. The HOMA2 model to estimate insulin resistance and beta cell function has several limitations and has not been validated for use in persons on glucose-lowering medications.29 Also, direct comparison of C-peptide levels and thus HOMA2 estimates between differe…第 9 页
  2. are common limitations to studies investigating diabetes subgroups based on HOMA2 estimates.第 9 页
  3. At baseline, 27% of participants reported not using glucose-lowering medication, likely indicating a mild diabetes, and thus our follow-up time may have been too short to detect late complications in these.第 9 页
  4. This proportion was, however, similar to findings from a population-based study from Norway with data from 2005 and 2014,30 indicating that our sample may be representative for a general Norwegian diabetes population.第 9 页
  5. The use of insulin, oral glucose-lowering medications and antihypertensive medication differed between subgroups at baseline.第 9 页
  6. We did not have information on the use of lipid-lowering medication, nor in changes in the use of medication during follow-up.第 9 页
2025 UAB多族裔验证数据驱动分型
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

新流程的判断（请逐条看对不对）
作者用 层次聚类基于 6 个变量聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Indeed, it yielded similar 5 subgroups and a very similar pattern of the variables used for clustering.第 6 页
  2. Using this cohort, we performed cluster ana lysis based on the same 6 variables as Ahlqvist et al, resulting in 5 comparable clusters.第 2 页
  3. Five clusters were then assigned by hierarchical clustering with Ward’s linkage using the AgglomerativeClustering function from the scikit-learn Python machine learning library.第 2 页
作者说明部分患者缺少自身抗体（GADA）结果，自身免疫型可能混入其他亚组。
看原文依据
  1. In analogy to the Ahlqvist analysis and because of the lack of sufficient data on other autoantibodies, only subjects with GAD-autoantibody results were included in the clustering,第 7 页
作者报告在其他队列中重复得到相似亚组；重复的是亚组结构，不等于各亚组比例一致。
看原文依据
  1. In addition, some of the observed differen ces in racial distribution have been independently validated in other US populations with more Hispanics/ Latinos.第 8 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. 4I), which is in contrast to the relative similar or lower incidence of cardiovascular disease observed between Black/African Americans and Whites with diabetes in other studies (11, 14-16).第 4 页
样本来自美国、非裔美国人人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. Patients: Adult patients with International Classification of Diseases diabetes codes were selected based on available data for 6 established clustering parameters (glutamic acid decarboxylase autoantibody; hemoglobin A1c; body mass index; diagnosis age; HOMA…第 1 页
文中有 7 处 HR/OR/RR 类估计，均来自观察性比较，只能读作关联，不能读作“某亚组导致了某结局”。
看原文依据
  1. Black/African Americans were more likely to have severe insulin-deficient diabetes (OR, 1.83; 95% CI, 1.36-2.45; P < .001), associated with more serious metabolic perturbations and a higher risk for complications (OR, 1.42; 95% CI, 1.06-1.90; P = .020).第 1 页
只看到相对风险（HR/OR），没找到绝对发生率或每千人年事件数；相对风险大不等于实际多出很多病例。
看原文依据
  1. Black/African Americans were more likely to have severe insulin-deficient diabetes (OR, 1.83; 95% CI, 1.36-2.45; P < .001), associated with more serious metabolic perturbations and a higher risk for complications (OR, 1.42; 95% CI, 1.06-1.90; P = .020).第 1 页
全文约 33 个 P 值，没找到多重比较校正的说明；个别“P<0.05”可能是偶然结果。
作者原话
作者结论（原文）：
看原文依据
  1. Conclusion: Racial background greatly influences diabetes cluster distribution and Black/African Americans are more frequently and more severely affected by severe insulin-deficient diabetes.第 1 页
作者原话
作者自述局限 6 句，主题：外推性。
看原文依据
  1. However, a limitation of our study is the fact that the findings have not (yet) been validated in Black populations outside of the United States.第 8 页
  2. Indeed, a recent study looking just at a population from Ghana found a completely different cluster distribution (35).第 8 页
  3. Because it is very likely that geographic dif ferences might exist, future larger studies in different parts of the world will be necessary in order to validate and obtain a global picture of these type 2 diabetes clusters and their implications.第 8 页
  4. This work was supported by National Institutes of Health grants R01DK137506 to A.S.第 8 页
  5. and the UM1TR004771 UAB Center for Clinical and Translational Science (CCTS).第 8 页
  6. The authors have nothing to disclose.第 8 页
2019 纪立农 中国老年新诊断T2DM观察性研究
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

现在软件的误判

判为“随机分组” ← Randomized, prospective trials that have examined glycemic control and complications focusing on the older patient (>65 years) are needed to obtain the optimal glycemic target.

新流程的判断（请逐条看对不对）
样本来自中国人群；换到其他人群，亚组比例、切点和结局差异都可能不同。
看原文依据
  1. treatment paradigm for newly diagnosed type 2 diabetes patients in Chin: A nationwide prospective cohort study,” Journal of Diabetes Investigation, 2019.第 8 页
全文没找到作者自述的局限段落（可能是分节错误或原文确实没写）。
纠正原软件
原分析把下面这句当作“本研究采用随机化”，实为展望或引用他人试验的句子，新流程不采纳。
看原文依据
  1. Randomized, prospective trials that have examined glycemic control and complications focusing on the older patient (>65 years) are needed to obtain the optimal glycemic target.第 undefined 页
2025 19项研究荟萃分析荟萃分析
现在软件给出的总结

从当前自动化提炼结果看，这是一项以"研究类型待进一步确认"为核心设计、且 当前取得全文文本（不等于证据质量较高） 的研究。最值得保留的主结论是：原文结果部分包含统计学数据，但自动结…

新流程的判断（请逐条看对不对）
合并结果的异质性很高（I² 最高 100%）；合并值只能当作粗略范围，不宜当作一个确定数字引用。
看原文依据
  1. In all included cases, the age was 50.14 (95%CI, 48.64 to 51.65; I2 = 100.0%, P < 0.001) years and the BMI was 27.54 (95%CI, 26.62 to 28.46; I2 = 100.0%, P < 0.001).第 3 页
纳入研究数见原文：“19 studies”。各研究使用的分型变量不同时，合并比例的含义会被稀释。
看原文依据
  1. Ultimately, 19 studies met al.l inclusion criteria and were selected for inclusion in the current meta-analysis7–11,19–32.第 2 页
作者原话
作者结论（原文）：
看原文依据
  1. In conclusion, our systematic review and meta-analysis is the first to summarize the clinical characteristics of the novel diabetes subgroups.第 7 页
作者原话
作者自述局限 6 句，主题：外推性、数据缺失、偏倚。
看原文依据
  1. Our study has several limitations.第 7 页
  2. Firstly, important demographic information such as race and ethnicity was missing from the retrieved articles, limiting our ability to perform detailed subgroup analyses based on these demographics.第 7 页
  3. Secondly, the included studies were predominantly conducted in two countries (China, n = 4 and Germany, n = 3), potentially leading to over-representative results of these regions.第 7 页
  4. Additionally, there was high heterogeneity (I2 value of 75 to 100%) in the gender ratio across different studies.第 7 页
  5. We also only reviewed studies published in English, which could introduce biases in our analysis.第 7 页
  6. Furthermore, factors such as environmental and genetic influences on diabetes development and progression were not considered in this study, which might affect the comprehensiveness and representativeness of our findings.第 7 页
2019 多国 ADOPT RECORD试验事后分析
现在软件给出的总结

尚未取得可直接支持两组角色的完整原句候选，比较对象待核实。 以下为有定位的结果与安全候选，尚需人工复核。统计学显著不代表临床获益或药物等效性。本报告不从实体顺序、通用阈值或候选片段…

现在软件的误判

判为“随机分组” ← All individuals within a trial were pooled to assess progression, regardless of randomised therapy.

新流程的判断（请逐条看对不对）
作者用 k-means聚类，得到 5 个亚组。亚组是算法按变量相似度划出来的，不是按病因划分的。
看原文依据
  1. Ahlqvist and colleagues4 identified five replicable clusters of individuals with diabetes in the All New Diabetics in Scania (ANDIS) cohort.第 1 页
  2. We applied K-means clustering specifying four clusters to the GADA-negative subset of individuals because K-means clustering does not incorporate binary variables; all GADA-positive individuals were manually assigned to a separate cluster.4 The same R command…第 3 页
作者说明部分患者缺少自身抗体（GADA）结果，自身免疫型可能混入其他亚组。
看原文依据
  1. Four type 2-like clusters were then characterised by the absence of GADA positivity and varying degrees of differences in age at diagnosis, and baseline measures of BMI, HbA1c, and homoeostatic model assessment (HOMA) 2 measured insulin resistance and β-cell …第 1 页
作者报告在其他队列中重复得到相似亚组（文中提到的队列：ADOPT、RECORD、ANDIS、UKPDS）；重复的是亚组结构，不等于各亚组比例一致。
看原文依据
  1. In the independent validation cohort (RECORD) we found clinical features outperformed the clusters for treatment selection.第 7 页
  2. (B) RECORD validation cohort(n=4057), clusters strategy (left panel) and clinical features strategy (right panel).第 8 页
亚组与并发症 / 结局的差异来自随访关联，说明亚组有预后区分度；不说明按亚组治疗能改善结局。
看原文依据
  1. We found differences in incidence of chronic kidney disease between clusters; however, estimated glomerular filtration rate at baseline was a better predictor of time to chronic kidney disease.第 1 页
本文是对既有试验（ADOPT、RECORD）数据的事后分析；亚组之间的疗效差异属于提出假设，没有被原试验的随机化和预设检验保护。
看原文依据
  1. Methods We identified five clusters in the ADOPT trial (n=4351) using the same data-driven cluster analysis as reported by Ahlqvist and colleagues.第 1 页
作者原话
作者结论（原文）：
看原文依据
  1. Interpretation The proposed data-driven clusters differ in diabetes progression and treatment response, but models that are based on simple continuous clinical features are more useful to stratify patients.第 1 页
作者原话
作者自述局限 2 句，主题：外推性、治疗混杂。
看原文依据
  1. Approaches to assess treatment selection strategies are not well developed and are the subject of ongoing methodological research.18 A limitation of our study is the potential nonrepresentativeness of participants due to the original trial exclusion criteria.第 9 页
  2. Both ADOPT and RECORD had exclusion criteria based on blood glucose levels and age (and BMI in RECORD); these clinical variables informed the cluster analysis.第 9 页
纠正原软件
原分析把下面这句当作“本研究采用随机化”，实为展望或引用他人试验的句子，新流程不采纳。
看原文依据
  1. All individuals within a trial were pooled to assess progression, regardless of randomised therapy.第 undefined 页
ADA 2026 Older Adults指南
现在软件给出的总结

当前分析基于PDF 全文提取，主要用于定位原文中的结果、方法与证据句。原文结果部分包含统计学数据，但自动结构化提取未达到足够置信度，建议结合原文核对。 具体结论及其适用范围仍应以原…

新流程的判断（请逐条看对不对）
本文共 8 条推荐带证据等级：A 3、B 3、C 1、E 1；其中 A 级（有高质量 RCT 支持）占 38%，E 级（专家共识）占 13%。
看原文依据
  1. <70 mg/dL [3.9 mmol/L] of <1%) for those with significant cognitive and/or functional limitations, frailty, severe comorbidities, and a less fa vorable risk-to-benefit ratio of diabe tes medications. C 13.7c Older adults with very complex or poor health recei…第 null 页
  2. Recommendations 13.16 Recommend diabetes education/ training (including that for CGM devi ces, insulin pumps, and advanced in sulin delivery systems) for the staff of post-acute and long-term care settings to improve the management of older adults with diabet…第 null 页
  3. Recommendations 13.1 Assess the medical, psychological, functional (self-management abilities), and social domains in older adults with diabetes using a comprehensive ap proach to determine goals and therapeutic approaches for diabetes manage ment. B 13.2 Scr…第 null 页
这是指南，不是一项研究；某条推荐有多可靠，看它的等级和所引用的原始研究，不看本文本身。
作者原话
作者结论（原文）：
看原文依据
  1. In conclusion, the management of dia betes in older adults at the end of life necessitates a person-centered approach that prioritizes attainment of individual ized treatment goals, comfort, symptom management, quality of life, and the preservation of dignity.第 16 页

三、最后两个问题

1. 把“很高”改成“较高”、“不能”改成“无法”这类语气或措辞变化，您能接受吗？
2. 看完后，您觉得这类报告最该提醒读者什么？现在漏了什么？

说明：新方案的规则是根据这批论文调出来的，换一批论文还需要再校。小模型改写部分只是评估用，默认不会出现在软件里。

尚未作答