Back to Details

design-study

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Design-Study Skill

Design-Study Skill

Purpose

目的

This skill pressure-tests whether a study is answerable, interpretable, and defensible before large amounts of drafting or analysis work accumulate.
Use it when:
  • a study question is known but the analysis plan is still fluid
  • the user wants a methods sanity check
  • a manuscript feels vulnerable to reviewer criticism
  • a peer review requires explicit methodological diagnosis
该Skill用于在大量起草或分析工作开展前，检验研究是否具备可回答性、可解释性和可辩护性。
适用于以下场景：
  • 研究问题已明确但分析方案仍未确定
  • 用户希望对研究方法进行合理性检查
  • 手稿可能容易受到审稿人批评
  • 同行评审需要明确的方法学诊断

Communication Rules

沟通规则

  • Communicate with the user in their preferred language.
  • Use English for statistical, radiologic, and reporting-guideline terminology.
  • Be direct about validity risks, but always propose the smallest feasible fix first.
  • 使用用户偏好的语言沟通。
  • 统计、放射学和报告指南术语使用英文。
  • 直接指出有效性风险，但始终先提出最小可行的修复方案。

Core Review Questions

核心审查问题

Always inspect these dimensions:
  1. What is the exact research question?
  2. What is the analysis unit: patient, lesion, exam, study, phase, report?
  3. What is the index date or decision point?
  4. How are inclusion and exclusion criteria applied?
  5. Is there any information leakage?
  6. What is the reference standard or endpoint definition?
  7. What comparator is clinically meaningful?
  8. What validation strategy is used?
  9. What uncertainty reporting is required?
  10. Which reporting guideline best fits?
  11. Are exposure/outcome/covariate definitions literature-grounded, or invented ad-hoc from the data dictionary? If ad-hoc, defer to 
    /define-variables
    before drafting Methods.
需始终检查以下维度：
  1. 具体的研究问题是什么？
  2. 分析单元是什么：患者、病灶、检查、研究、阶段、报告？
  3. 索引日期或决策点是什么？
  4. 纳入与排除标准如何应用？
  5. 是否存在信息泄露？
  6. 参考标准或终点定义是什么？
  7. 具有临床意义的对照是什么？
  8. 使用的验证策略是什么？
  9. 需要报告哪些不确定性？
  10. 最适配的报告指南是哪一个？
  11. 暴露/结局/协变量的定义是否基于文献，还是从数据字典中临时拟定？如果是临时拟定，在起草方法部分前先调用
    /define-variables
    。

Standard Output

标准输出

text
undefined
text
undefined

Study Design Review

Study Design Review

Question: ... Study type: ... Analysis unit: ... Index date / prediction timepoint: ...
Question: ... Study type: ... Analysis unit: ... Index date / prediction timepoint: ...

Strengths

Strengths

  • ...
  • ...

Major validity risks

Major validity risks

  1. ...
  2. ...
  1. ...
  2. ...

Minimal fixes

Minimal fixes

  • ...
  • ...

Reporting fit

Reporting fit

  • Recommended guideline: ...
  • Recommended guideline: ...

Decision

Decision

  • Ready for analysis / Needs redesign / Drafting can proceed with limitations

---
  • Ready for analysis / Needs redesign / Drafting can proceed with limitations

---

Workflow

工作流程

Phase 1: Reconstruct the study

阶段1：重构研究

Extract from protocol, draft, slides, tables, or notes:
  • clinical problem
  • intended use case
  • population
  • inputs
  • outputs
  • outcome definition
  • timing of variable availability
Gate: Present the reconstructed study summary (question, analysis unit, intended use) to the user. Confirm before proceeding — if the reconstruction is wrong, the entire validity review will be misdirected.
从方案、草稿、幻灯片、表格或笔记中提取：
  • 临床问题
  • 预期用例
  • 研究人群
  • 输入数据
  • 输出结果
  • 结局定义
  • 变量可用时间点
关卡： 向用户展示重构后的研究摘要（问题、分析单元、预期用途），确认无误后再继续——如果重构错误，整个有效性审查都会偏离方向。

Phase 2: Check structural validity

阶段2：检查结构有效性

A. Analysis unit

A. 分析单元

Look for mismatches such as:
  • patient-level claim from lesion-level analysis
  • exam-level split with patient overlap
  • phase-level samples treated as independent
寻找不匹配情况，例如：
  • 基于病灶层面分析得出患者层面结论
  • 检查层面拆分但存在患者重叠
  • 阶段层面样本被视为独立样本

B. Leakage

B. 信息泄露

Look for:
  • postoperative features used for preoperative prediction
  • normalization or thresholding performed before data split
  • repeated exams across train/test
  • reader annotations derived from outcome information
  • input-text contamination for NLP/LLM extraction tasks: if the model input includes report sections such as clinical history, indication, impression, prior diagnosis, or referral text, confirm that those fields do not literally name or strongly imply the target label. If the target is already present in the supplied text, the task is information retrieval under label leakage, not phenotype inference; redesign the input mask, report a sensitivity analysis excluding leaky fields, or reframe the claim.
  • construct dependence (a predictor that is a definitional component of the outcome). Two cases: (i) mathematical definition — an input that computes the outcome (when the outcome is HOMA-IR = f(fasting insulin, fasting glucose), those two inputs are not independent predictors); (ii) near-tautological composite — a ratio or score built from the outcome's defining components, which shows an inflated, near-circular association. Test: "could this predictor be derived, in whole or part, from the outcome's definition or the same measurement?" If yes, exclude it, or retain it only as a labeled calibration probe rather than a reported discovery.
寻找以下情况：
  • 将术后特征用于术前预测
  • 在数据拆分前进行归一化或阈值处理
  • 训练/测试集中存在重复检查数据
  • 阅片者注释源自结局信息
  • NLP/LLM提取任务的输入文本污染：如果模型输入包含临床病史、检查指征、印象、既往诊断或转诊文本等报告章节，需确认这些字段未直接提及或强烈暗示目标标签。如果目标标签已存在于提供的文本中，该任务属于标签泄露下的信息检索，而非表型推断；需重新设计输入掩码、报告排除泄露字段的敏感性分析，或重新表述研究结论。
  • 结构依赖（预测因子是结局定义的组成部分）。分为两种情况：(i) 数学定义——用于计算结局的输入（例如结局为HOMA-IR = f(空腹胰岛素, 空腹血糖)，则这两个输入并非独立预测因子）；(ii) 近乎同义的复合指标——由结局定义组件构建的比率或评分，会显示出夸大的近乎循环的关联。测试方法：“该预测因子是否可以全部或部分从结局定义或相同测量值中推导出来？”如果是，则排除该预测因子，或仅将其作为标记的校准探针而非已报道的发现保留。

F. Time origin & survivorship (incident / transition models)

F. 时间起点与生存情况（事件/转换模型）

For any time-to-event or incident/transition design, check before drafting:
  • Time origin per model. Each incident model starts its at-risk clock at the correct origin. Watch for immortal-time bias (a span in which the event cannot occur, misattributed to one group) and left-truncation / delayed entry (subjects entering the risk set after the origin).
  • Mediator-ascertainment-window survivorship. A "progressor" / transition label that is conditional on surviving to a later ascertainment (a second scan, a follow-up visit) is survivorship-biased; plan a landmark time or an explicit intermediate-state (multistate / illness-death) model.
  • Primary-analysis-set selection. If the primary will not be the full cohort (e.g., complete-case while a large fraction is missing), pre-specify the selection justification and a MAR rationale; do not let the complete-case model become primary because it is the significant one (an outcome-dependent choice).
  • A design that cannot yet answer these should say so honestly — but note that at review time a Methods/Limitations admission that the issue was "not formally assessed" is escalated to a MAJOR by the survival probe (S1), not waved through as a limitation.
对于任何时间-事件或事件/转换设计，起草前需检查：
  • 每个模型的时间起点。每个事件模型需从正确起点启动风险时钟。注意不朽时间偏倚（事件无法发生的时间段被错误归因于某一组）和左截断/延迟入组（研究对象在起点后才进入风险集）。
  • 中介变量确定窗口的生存偏倚。以“存活至后续确定时间点（第二次扫描、随访就诊）”为条件的“进展者”/转换标签存在生存偏倚；需规划里程碑时间或明确的中间状态（多状态/疾病-死亡）模型。
  • 主要分析集选择。如果主要分析集不是完整队列（例如，大部分数据缺失时使用完整病例分析），需预先说明选择理由和MAR（随机缺失）假设；不能仅因为完整病例模型有统计学意义就将其作为主要分析集（这是依赖结局的选择）。
  • 目前无法回答这些问题的设计需如实说明——但需注意，在评审时，方法/局限性部分承认该问题“未正式评估”会被生存探针（S1）升级为重大问题，而非作为局限性被忽略。

C. Reference standard

C. 参考标准

Check:
  • who established ground truth
  • when it was established
  • whether blinding was possible
  • whether only a subset had gold standard verification
  • Construct ↔ nominal-definition match. Does the exposure/finding construct stay inside its stated definition, or does it quietly exceed it? An "incidentaloma" defined as an indeterminate finding must not include frank malignancy reads; a label that overshoots its definition inflates the apparent cohort and breaks the κ. For each construct, restate the nominal definition and confirm every included case satisfies it.
  • Per-flag reference-standard concordance. When the index finding is flagged against a reference standard, report the concordance per flag category (not just overall). A construct where a large fraction of flags do not match the reference standard (e.g., ~86% non-match) is measuring something other than the named construct.
  • Manuscript definition ↔ 
    variable_operationalization.md
    .     The variable definitions written in Methods must match the operationalization table verbatim (dictionary-first). A blinded re-classification form must quote the analytic protocol's definition verbatim — paraphrase / "common-sense extension" in the form (but not the Methods) is the documented cause of a low κ that is a definition mismatch, not real disagreement. Cross-check with 
    /define-variables
    output before drafting.
检查：
  • 谁确立了金标准
  • 何时确立
  • 是否可能实施盲法
  • 是否仅部分样本接受了金标准验证
  • 结构↔名义定义匹配。暴露/发现的结构是否符合其既定定义，还是超出了定义范围？例如，定义为不确定发现的“偶发瘤”不得包含明确恶性病变的阅片结果；超出定义范围的标签会扩大表观队列规模并破坏κ值。对于每个结构，重述名义定义并确认所有纳入病例均符合该定义。
  • 每类标记的参考标准一致性。当索引发现与参考标准进行标记比对时，需按标记类别报告一致性（而非仅整体一致性）。如果大部分标记与参考标准不匹配（例如约86%不匹配），则该结构测量的并非命名的目标结构。
  • 手稿定义↔
    variable_operationalization.md
    。方法部分中撰写的变量定义必须与操作化表完全一致（以字典优先）。盲法重新分类表必须逐字引用分析方案中的定义——表中的意译/“常识扩展”（而非方法部分）是κ值低的记录原因，这属于定义不匹配，而非真实分歧。起草前需与
    /define-variables
    的输出进行交叉核对。

D. Validation

D. 验证

Classify:
  • apparent only
  • internal split
  • cross-validation
  • temporal validation
  • external validation
  • multi-center external validation
分类：
  • 仅表观验证
  • 内部拆分验证
  • 交叉验证
  • 时间验证
  • 外部验证
  • 多中心外部验证

E. Reader / expert-elicitation studies (load on demand)

E. 阅片者/专家征询研究（按需加载）

When the study elicits expert ratings — a reader study, an annotation panel, an AI-output evaluation — the design decisions that matter are made before data collection, and the acceptance ceiling of a perceptual / reader AI study is fixed at design time: no quality of execution lifts a ceiling baked into the comparator, the estimand, or the reader cohort.
For an AI-system-versus-human-expert benchmark specifically, route to 
/design-ai-benchmarking
, which extends this subsection with arm definition, LLM-as-judge versus human-as-judge adjudication, and a structured export schema.
FileRead it whenCost if read blindly
references/reader_elicitation_design.md
the design has a human-rater or expert-elicitation arm — rubric axes, calibration probes, operational rigor, human-as-operator, and the six ceiling decisions~2,400 tokens, none of which applies to a design with no reader arm
references/dag_adjustment.md
confounding control needs an explicit adjustment set—
references/target_trial_emulation.md
the design emulates a target trial—
references/venue_accept_recipe.md
it is a clinical DL / AI-validation study and the question is which venue tier the achievable design can be accepted at, and the one design move that reaches the tier above (the design→acceptance-tier ladder + the five acceptance levers, reverse-engineered from accepted OA papers; the bridge into 
/find-journal
)		~1,800 tokens; skip for a design with no publication-tier decision
references/combine_models_ablation_design.md
the model is built by combining / adapting / fine-tuning existing models (nnU-Net, TotalSegmentator, SAM/MedSAM, a pretrained backbone) — how to design the comparator as an ablation that proves the combination earns its complexity (beat the un-adapted base + the best single component + direct-train), reverse-engineered from accepted OA papers~1,600 tokens; skip if the model is trained de novo with no reused component
references/multi_model_comparison_design.md
the study's contribution is comparing several models / architectures head-to-head (CNN vs Transformer vs foundation backbone; N segmentation networks) — how to make the comparison fair: one frozen split + one preprocessing through every model, a strong self-configuring baseline (nnU-Net) not a hobbled one, matched training/HPO/compute budget (the #1 threat — "new≠better, just tuned harder") or disclosed, variance-over-single-run, a pre-specified primary metric + a paired delta test, honest ranking. Reverse-engineered from accepted OA papers~1,700 tokens; skip for a single-model study (use 
combine_models_ablation_design.md
for an ablation of one model, 
/design-ai-benchmarking
for AI-vs-human)
references/segmentation_failure_characterization_design.md
the claim is that a segmentation model is clinically usable, not that it scores well — a pre-specified failure taxonomy (boundary drift / missed / hallucinated / catastrophic), an acceptability endpoint with a stated scale, named judges and an adjudication rule, the tail (per-case distribution, worst percentile, catastrophic count) beside the mean, edit effort paired against manual-from-scratch and disaggregated per structure and site, and failures stratified by what predicts them. Reverse-engineered from accepted OA papers~1,700 tokens; skip when the endpoint is benchmark accuracy with no usability claim (metric choice → 
/model-evaluation
; abstention / risk–coverage → 
/uncertainty-imaging
)
当研究需要征询专家评分时——如阅片研究、注释小组、AI输出评估——关键的设计决策是在数据收集前做出的，感知/阅片AI研究的接受上限在设计时就已确定：执行质量无法突破由对照、估计量或阅片者队列决定的上限。
对于AI系统与人类专家的基准测试，需转至
/design-ai-benchmarking
，该模块扩展了本节内容，包括试验组定义、LLM作为评判者与人类作为评判者的裁决，以及结构化导出模式。
文件阅读时机盲目阅读的成本
references/reader_elicitation_design.md
设计包含人类评分者或专家征询组——涉及评分轴、校准探针、操作严谨性、人类操作者，以及六项上限决策约2400个token，不适用于无阅片组的设计
references/dag_adjustment.md
需要明确调整集以控制混杂—
references/target_trial_emulation.md
设计模拟目标试验—
references/venue_accept_recipe.md
这是临床DL/AI验证研究，且问题是可实现的设计能被哪个层级的期刊接受，以及能提升一个层级的设计调整方案（设计→接受层级阶梯+五项接受杠杆，从已接受的OA论文逆向工程而来；衔接
/find-journal
）		约1800个token；无需做期刊层级决策的设计可跳过
references/combine_models_ablation_design.md
模型是通过组合/适配/微调现有模型（nnU-Net、TotalSegmentator、SAM/MedSAM、预训练骨干网络）构建的——如何将对照设计为消融试验，以证明组合模型的复杂性具有价值（击败未适配的基础模型+最佳单一组件+直接训练模型），从已接受的OA论文逆向工程而来约1600个token；从头训练且无复用组件的模型可跳过
references/multi_model_comparison_design.md
研究的贡献是直接比较多个模型/架构（CNN vs Transformer vs 基础骨干网络；N个分割网络）——如何使比较公平：统一固定拆分+统一预处理应用于所有模型，使用强大的自配置基线（nnU-Net）而非受限基线，匹配训练/HPO/计算资源预算（头号威胁——“新≠更好，只是调参更充分”）或披露预算，单次运行的方差，预先指定的主要指标+配对差异检验，如实排名。从已接受的OA论文逆向工程而来约1700个token；单模型研究可跳过（单个模型的消融试验使用
combine_models_ablation_design.md
，AI与人类对比使用
/design-ai-benchmarking
）
references/segmentation_failure_characterization_design.md
研究声称分割模型具有临床可用性，而非评分优异——需预先指定失败分类法（边界偏移/遗漏/幻觉/灾难性错误）、明确量表的可接受性终点、指定评判者和裁决规则，除均值外还需关注尾部情况（逐病例分布、最差百分位数、灾难性错误数量），将编辑工作量与从头手动标注进行对比，并按结构和站点分类，同时按预测因素分层分析失败情况。从已接受的OA论文逆向工程而来约1700个token；终点为基准准确率且无可用性声称的情况可跳过（指标选择→
/model-evaluation
；弃权/风险-覆盖→
/uncertainty-imaging
）

Phase 3: Clinical framing

阶段3：临床框架

Ask whether the comparator and endpoint support the stated claim:
  • is the model better than current practice or just another model?
  • is the endpoint clinically meaningful?
  • does performance translate to action?
  • incremental value: if the study frames the model/marker as adding value beyond / on top of / incremental to an existing tool (a clinical score, a routine test, a baseline model), the design must pre-specify the baseline comparator built from the in-routine-use predictors and an incremental-value metric — ΔC-index / ΔAUC (with a paired CI, e.g. DeLong), categorical or continuous NRI, IDI, or decision-curve net benefit. A standalone discrimination number ("our model's AUC was 0.84") does not support a "beyond X" claim; without the nested-model comparison the finding may be real but redundant. Plan this at design time — it cannot be added post hoc without the baseline model.
  • fine-tuning contribution baseline: if an NLP/LLM study claims that fine-tuning, LoRA, prompt engineering, or a multi-agent wrapper improves extraction/classification, pre-specify a same-backbone zero-shot or few-shot comparator on the identical input, output schema, and test split. A comparison only against a weaker or unrelated baseline cannot establish that the proposed adaptation adds value. For an imaging model built by combining / adapting / fine-tuning existing models (nnU-Net, a foundation model, a pretrained backbone), design the full ablation ladder — un-adapted base, best single component, direct-train vs transfer — per 
    references/combine_models_ablation_design.md
    . When the contribution is instead a head-to-head comparison of several models (which architecture wins), the decisive design question is comparison fairness — one frozen split/preprocessing through every model, a strong fairly-tuned baseline, a matched (or disclosed) compute budget, and a paired delta test — per 
    references/multi_model_comparison_design.md
    . When the claim is not that a segmentation model scores well but that it is clinically usable, the design must carry a pre-specified failure taxonomy, an acceptability endpoint with a named judge and adjudication rule, the tail beside the mean, and edit effort paired against manual-from-scratch — per 
    references/segmentation_failure_characterization_design.md
    ; a mean DSC cannot be converted into a usability claim after the fact.
  • endpoint↔conclusion scope: decide up front what kind of conclusion the design can support, so the manuscript does not overreach. A cross-sectional / single-visit / prevalence design cannot support a prognostic or surveillance claim (rescreen interval, disease progression) — that needs longitudinal follow-up. A binary surrogate endpoint (present/absent, >0, dichotomized) is risk stratification, not a patient-care directive (defer/withhold/initiate therapy). At review time 
    /self-review
    §D + 
    check_scope_coherence.py
    flag 
    CROSS_SECTIONAL_PROGNOSTIC
    / 
    SURROGATE_CARE_DIRECTIVE
    against the conclusion.
询问对照和终点是否支持既定结论：
  • 模型是优于当前实践还是仅优于其他模型？
  • 终点是否具有临床意义？
  • 性能是否能转化为临床行动？
  • 增量价值：如果研究声称模型/标志物在现有工具（临床评分、常规检查、基线模型）之外/之上/增量增加价值，设计必须预先指定由常规使用的预测因子构建的基线对照以及增量价值指标——ΔC指数/ΔAUC（带配对置信区间，如DeLong法）、分类或连续NRI、IDI，或决策曲线净获益。单独的鉴别数值（“我们的模型AUC为0.84”）无法支持“优于X”的结论；若无嵌套模型对比，发现可能真实但冗余。需在设计阶段规划——事后无法添加基线模型。
  • 微调贡献基线：如果NLP/LLM研究声称微调、LoRA、提示工程或多Agent包装器改进了提取/分类效果，需预先指定在相同输入、输出模式和测试集上的同骨干网络零样本或少样本对照。仅与更弱或无关基线对比无法证明所提出的适配方案增加了价值。对于通过组合/适配/微调现有模型（nnU-Net、基础模型、预训练骨干网络）构建的影像模型，需按照
    references/combine_models_ablation_design.md
    设计完整的消融阶梯——未适配基础模型、最佳单一组件、直接训练vs迁移训练。如果研究贡献是直接比较多个模型（哪种架构胜出），决定性的设计问题是对比的公平性——统一固定拆分/预处理应用于所有模型、强大且公平调优的基线、匹配（或披露）的计算资源预算、配对差异检验，需遵循
    references/multi_model_comparison_design.md
    。如果研究声称分割模型并非评分优异而是具有临床可用性，设计必须包含预先指定的失败分类法、带指定评判者和裁决规则的可接受性终点、均值之外的尾部情况、与从头手动标注对比的编辑工作量——需遵循
    references/segmentation_failure_characterization_design.md
    ；事后无法将平均DSC转化为可用性声称。
  • 终点↔结论范围：预先确定设计能支持的结论类型，避免手稿过度延伸。横断面/单次就诊/患病率设计无法支持预后或监测结论（复查间隔、疾病进展）——这需要纵向随访。二元替代终点（存在/不存在、>0、二分类）属于风险分层，而非患者护理指令（推迟/暂停/启动治疗）。评审时
    /self-review
    §D + 
    check_scope_coherence.py
    会标记
    CROSS_SECTIONAL_PROGNOSTIC
    /
    SURROGATE_CARE_DIRECTIVE
    与结论不一致的情况。

Phase 4: Reporting fit

阶段4：报告适配性

Recommend one primary guideline:
  • TRIPOD-AI
  • CLAIM
  • STARD
  • STROBE
  • PRISMA
  • CARE
  • ARRIVE
  • journal-specific additions if needed
推荐一个主要指南：
  • TRIPOD-AI
  • CLAIM
  • STARD
  • STROBE
  • PRISMA
  • CARE
  • ARRIVE
  • 必要时添加期刊特定要求

Frequent Failure Modes

常见失败模式

Diagnostic AI

诊断AI

  • no clinically relevant comparator
  • exam-level split instead of patient-level split
  • unclear reference standard
  • AUROC-only reporting without threshold metrics
  • 无临床相关对照
  • 检查层面拆分而非患者层面拆分
  • 参考标准不明确
  • 仅报告AUROC而无阈值指标

Prognostic modeling

预后建模

  • unclear time zero
  • immortal time bias
  • feature timing mismatch
  • no calibration
  • 时间零点不明确
  • 不朽时间偏倚
  • 特征时间不匹配
  • 未进行校准

Retrospective cohort / screening database

回顾性队列/筛查数据库

  • time zero misalignment: cohort entry ≠ follow-up start → immortal time bias
  • interval-censored outcomes treated as exact → underestimation of event times
  • healthy volunteer bias unacknowledged → inflated external validity claims
  • surveillance bias from unequal follow-up frequency between groups
  • 3 bias classification (Hernan/Robins): selection bias (who enters), information bias (how measured), confounding (what else differs) — explicitly map each threat
  • comparative / causal question → emulate a target trial. For a treatment-vs-treatment, screening-vs-no-screening, or drug-A-vs-drug-B question on routinely-collected data, specify the seven target-trial components (eligibility, strategies, assignment, time zero, outcome, causal contrast, analysis plan) before extraction — this is what prevents the immortal-time / prevalent-user / confounding-by-indication trio above and turns an association into a defensible causal contrast. New-user + active-comparator design, grace-period clone-censor-weight, and negative controls are in 
    references/target_trial_emulation.md
    .
  • confounding completeness: pre-specify the adjustment set from a DAG (not a Table-1 p < 0.05 rule), and plan to report whether any measured covariate that turns out imbalanced by exposure but outside the adjustment set leaves the primary estimate robust (an extended-adjustment sensitivity model). Build the DAG and pre-screen the proposed covariates with 
    scripts/adjustment_set_helper.py
    (flags mediator / collider / descendant adjustment and omitted confounders, and proposes a candidate backdoor set), then derive the minimal sufficient set with dagitty — see 
    references/dag_adjustment.md
    . At review time 
    /self-review
    Phase 2.5e + the O1–O12 probes in 
    observational_confounding.md
    check this against Table 1 (including O7 over-adjustment, O10 overlapping-subset-gradient discipline, for complex-survey data O11 design-based weighting and O12 data-driven-threshold mining, O13 — a cross-sectional mediation claim cannot order X→M→Y, and O14 — a synergy/joint-effect claim needs the additive interaction scale (RERI/AP/S), not a multiplicative-only test).
  • 时间零点错位：队列入组≠随访开始→不朽时间偏倚
  • 区间删失结局被视为精确结局→事件时间被低估
  • 未承认健康志愿者偏倚→外部有效性主张被夸大
  • 组间随访频率不均导致监测偏倚
  • 3类偏倚分类（Hernan/Robins）：选择偏倚（谁入组）、信息偏倚（如何测量）、混杂偏倚（其他差异因素）——需明确映射每个风险
  • 比较/因果问题→模拟目标试验。对于常规收集数据的治疗vs治疗、筛查vs无筛查、药物Avs药物B问题，提取前需指定目标试验的七个组件（ eligibility、策略、分配、时间零点、结局、因果对比、分析方案）——这可防止上述不朽时间/现患用户/指示混杂三联偏倚，并将关联转化为可辩护的因果对比。新用户+活性对照设计、宽限期克隆删失权重、阴性对照详见
    references/target_trial_emulation.md
    。
  • 混杂控制完整性：从DAG预先指定调整集（而非基于表1的p<0.05规则），并计划报告任何暴露组间不平衡但在调整集之外的测量协变量是否会使主要估计值保持稳健（扩展调整敏感性模型）。使用
    scripts/adjustment_set_helper.py
    构建DAG并预先筛选拟议协变量（标记中介/碰撞体/后代调整和遗漏混杂因素，并提出候选后门集），然后使用dagitty推导最小充分集——详见
    references/dag_adjustment.md
    。评审时
    /self-review
    阶段2.5e + 
    observational_confounding.md
    中的O1–O12探针会对照表1检查这一点（包括O7过度调整、O10重叠子集梯度原则、复杂调查数据的O11基于设计的加权和O12数据驱动阈值挖掘、O13——横断面中介主张无法确定X→M→Y的顺序、O14——协同/联合效应主张需要加性交互尺度（RERI/AP/S），而非仅乘性检验）。

Multimodal LLM / report generation

多模态LLM/报告生成

  • no clear rubric for clinical correctness
  • benchmark labels derived from noisy reports without adjudication
  • unsupported claims about safety or workflow benefit
  • input text contains the target label or diagnosis being predicted
  • no same-backbone zero-shot/few-shot baseline for a fine-tuning or prompt-engineering claim
  • 无明确的临床正确性评分标准
  • 基准标签源自未经裁决的噪声报告
  • 关于安全性或工作流获益的主张无依据
  • 输入文本包含待预测的目标标签或诊断
  • 微调或提示工程主张无同骨干网络零样本/少样本基线

Imaging meta-analysis

影像荟萃分析

  • overlapping cohorts
  • paired modalities analyzed as independent
  • heterogeneity metrics missing
  • zero-cell handling unspecified
  • 队列重叠
  • 配对模态被视为独立分析
  • 异质性指标缺失
  • 零单元格处理方式未明确

Minimal-Fix Principle

最小修复原则

Whenever possible, recommend the smallest feasible repair first:
  • clarify the claim
  • narrow the target population
  • add a limitation statement
  • add a clinically relevant baseline
  • re-run one key sensitivity analysis
  • redefine the endpoint more explicitly
Escalate to redesign only when the central claim is not defensible otherwise.
尽可能先推荐最小可行的修复方案：
  • 明确研究主张
  • 缩小目标人群范围
  • 添加局限性说明
  • 添加临床相关基线
  • 重新运行一项关键敏感性分析
  • 更明确地重新定义终点
仅当核心主张无法通过其他方式辩护时，才升级为重新设计。

Handoff Rules

移交规则

  • route to 
    analyze-stats
    when the design is basically sound but analysis details need refinement
  • route to 
    check-reporting
    after the design is locked
  • route to 
    self-review
    when the user wants a pre-submission quality check on their own manuscript
  • route back to 
    write-paper
    only after the main validity risks are documented
  • 当设计基本合理但需细化分析细节时，转至
    analyze-stats
  • 设计锁定后，转至
    check-reporting
  • 当用户希望对手稿进行提交前质量检查时，转至
    self-review
  • 仅在记录主要有效性风险后，才转回
    write-paper

What This Skill Does NOT Do

该Skill不执行的操作

  • It does not compute statistics directly
  • It does not draft full manuscript prose
  • It does not resolve raw data engineering issues
  • It does not replace a full peer review when journal-facing tone is required
  • 不直接计算统计量
  • 不起草完整的手稿正文
  • 不解决原始数据工程问题
  • 当需要符合期刊语气时，无法替代完整的同行评审

Anti-Hallucination

反幻觉规则

  • Never fabricate references. All citations must be verified via 
    /search-lit
    with confirmed DOI or PMID. Mark unverified references as 
    [UNVERIFIED - NEEDS MANUAL CHECK]
    .
  • Never invent clinical definitions, diagnostic criteria, or guideline recommendations. If uncertain, flag with 
    [VERIFY]
    and ask the user.
  • Never fabricate numerical results — compliance percentages, scores, effect sizes, or sample sizes must come from actual data or analysis output.
  • If a reporting guideline item, journal policy, or clinical standard is uncertain, state the uncertainty rather than guessing.
  • 切勿编造参考文献。所有引用必须通过
    /search-lit
    验证DOI或PMID。未验证的参考文献标记为
    [UNVERIFIED - NEEDS MANUAL CHECK]
    。
  • 切勿编造临床定义、诊断标准或指南建议。如有不确定，标记为
    [VERIFY]
    并询问用户。
  • 切勿编造数值结果——合规百分比、评分、效应量或样本量必须来自实际数据或分析输出。
  • 如果对报告指南条目、期刊政策或临床标准不确定，需说明不确定性而非猜测。