请作为心内科研究生导师，检查一下manuscript的result部分表达的逻辑和内容是否符合学术期...

下面以心内科研究生导师视角，对你稿件的 Results（结果）部分及相关图表/补充材料的引用进行系统性审阅与校正建议。所有问题与修改点均基于你上传的主文稿与补充材料逐条核对而来。(manuscript.docx) (supplementary files.docx)

---

一、总体评价（结论先行）

• 整体逻辑：结果部分基本遵循“研究人群 → 基线特征 → 主要结局 → 次要结局/延长随访 → 敏感性分析/亚组 → 机器学习探究”的结构，核心发现（BAR 与多时间点死亡风险的独立、近线性关联）与表格数值一致，方向清晰。(manuscript.docx)
• 可发表性：主要统计指标（HR、95%CI、P 值）和关键图表（KM 曲线、RCS、亚组森林图）已覆盖主结论，具备投稿基础。
• 需要重点整改之处：存在图题与内容不匹配、分组边界表述不规范、段落重复、编号/缩写/单位不一致、补充材料编号错配等，会直接影响学术规范性与审稿体验，建议优先修正（详见下文“必须修改”）。(manuscript.docx) (supplementary files.docx)

---

二、必须修改（Major issues）

1. BAR 三分位分组区间不连贯

   • 现状：正文写“T1 (4–7)、T2 (9–12)、T3 (17–27)，各 337 例”，区间 出现空档（8、13–16 缺失），与“等例数三分位”表述逻辑冲突；表 1 的 BAR 分布（总体 11[7–17]；各组中位与 IQR）也提示应为连续分段而非“跳跃段”。(manuscript.docx)
   • 建议：给出确切的三分位切点并统一全文/表/图（示例：T1 ≤7 mg·g⁻¹；T2 7–12 mg·g⁻¹；T3 ≥12 mg·g⁻¹；以实际分位点为准），并在表 1 的栏头与图注中同时标注单位与切点。(manuscript.docx)

2. 图 2 的题名与内容不一致

   • 现状：图题写“Cox Proportional‑Hazards Models…”，但图注与正文描述的是Kaplan–Meier 曲线（A–D 为 30/60/120/365 天生存曲线）。(manuscript.docx)
   • 建议：将图题更正为“Kaplan–Meier survival by BAR tertiles at 30/60/120/365 days”，如另有 Cox 森林图，请新设独立图编号（或移至补充材料），避免题图不符。(manuscript.docx)

3. 3.3 节内容重复

   • 现状：3.3 内部出现两段几乎同义的文字（一次以“在多变量 Cox 中…”起笔，一次以“Kaplan–Meier analysis revealed…”起笔），且 RCS 与亚组结果在上下文重复描述。(manuscript.docx)
   • 建议：保留一次完整、紧凑的表述；删除重复句群，使“KM → 多变量 Cox（连续与分组）→ RCS → 亚组/交互”的叙述各述一次即可。(manuscript.docx)

4. 小节编号冲突

   • 现状：出现两个“3.5”（“Sensitivity, Robustness…” 与 “Machine Learning Validation”）且随后还有“3.6 Prespecified Subgroups…”。(manuscript.docx)
   • 建议：顺序建议为 3.5 Subgroups → 3.6 Sensitivity/Diagnostics → 3.7 ML（Exploratory），或将 ML 整体置于补充材料，仅在主文结果简短提要。(manuscript.docx)

5. 统计方法细节在文内不一致（MICE 次数）

   • 现状：方法学段落先写“m=20 次插补”，后文又出现“m=5”。(manuscript.docx)
   • 建议：统一为一个数值，并在结果或补充方法附“插补收敛与敏感性一致性”简述。(manuscript.docx)

6. 缩写不统一（IMV/IVW）

   • 现状：正文/表格混用“IMV/IVW”，方法脚注标注“IMV = invasive mechanical ventilation”。(manuscript.docx)
   • 建议：全稿统一用 IMV，表 1 与表 2、图注、补充材料同时修正。(manuscript.docx)

7. 中英文混排与不确定语气

   • 现状：3.2 基线特征段落含中文与疑问号（如“体温？未达显著”），而表 1 中体温 P<0.001 为显著差异。(manuscript.docx)
   • 建议：全英文并与表 1 对齐（体温为显著差异变量），删除疑问式语句。(manuscript.docx)

8. 单位/精度与术语

   • 现状：表 2 行题“Per 1 Unit increase”建议明确为“Per 1 mg·g⁻¹ increase”；pH 在表 1 写“7 (7–7)”不合常规精度；动脉血气符号（PaO₂/PaCO₂）与 RBC 单位记法不统一。(manuscript.docx)
   • 建议：
     • 统一写法：BAR（mg·g⁻¹）；PaO₂/PaCO₂（文内可用 PO₂/PCO₂，但图表尽量用规范学术写法）；RBC（×10⁶/µL）；pH 保留 2 位小数。
     • HR 与 95%CI 建议各保留 2–3 位小数，P 值统一格式 P=0.001 或 P<0.001。(manuscript.docx)

9. 补充材料编号错配

   • 现状：正文多处提到“Supplementary Table S5（Boruta）”，而补充材料实际为“Supplementary Table 6：Boruta 特征筛选结果”；正文称“S1 为变量定义与单位”，但补充 S1 实为“插补前连续变量与缺失概况”，并含正态性检验 S2、死亡分层基线 S3、单因素 Cox S4 等。(manuscript.docx) (supplementary files.docx)
   • 建议：统一编号（例如：S1 变量定义与单位；S2 缺失与分布；S3 生存与死亡基线；S4 单因素 Cox；S5–S6 机器学习/DCA/PR/SHAP 与 Boruta），并在正文相应处逐一更正“指向”。(manuscript.docx) (supplementary files.docx)

10. 图 5、图 6 在正文被提及但主文中无确切图题与内容

    • 现状：正文“机器学习”段落提到 Supplementary Figures 4–6，同页却出现“Figure 5/6”占位且无图题。(manuscript.docx) (supplementary files.docx)
    • 建议：若 ML 结果留在补充材料，请删除主文 Figure 5/6 占位；若保留主文展示，请给出完整图题/图注并确保与补充材料编号不冲突。(manuscript.docx)

11. 明显的文字/排版差错

    • 例如：“Supplem014270578/8entary…” 乱码；表 2 某处出现“…2.536)f”；“加number”提示语出现在表题；CI 数值间有制表符；个别小数点或括号缺失等。建议统一由语言/排版一轮清校修正。(manuscript.docx)

12. 图表信息完整性

    • KM 图注称“显示风险人数（numbers at risk）”，请确保图下方实际展示；若未展示，请增加或在图注删除此语。(manuscript.docx)

---

三、次要问题（Should fix）

• 结果与方法互证：3.5“敏感性/稳健性”中提到“按 5 单位增量、多种修剪”等，建议在补充表给出相应 HR 与 95%CI（简短即可），便于同行复核。(manuscript.docx)
• “P for trend” 统一：在表 2 的四个随访时间点，建议统一为 “P for trend”，不与“Test for trend”混用，并与文中叙述完全一致（30 天为 0.001；60/120/365 天为 ≤0.001）。(manuscript.docx)
• 把“绝对风险”写清（建议增强）：KM 图已经隐含三分位各自 30/60/120/365 天的累计死亡率，建议在 3.3/3.4 中各给出一次绝对事件率（例如 T1/T3 的 30 天死亡率 %），有助临床解读。(manuscript.docx)
• 交互 P 值书写：统一为 P_interaction 或 “P for interaction”。(manuscript.docx)

---

四、图表与引用核对清单（逐一对照）

元素	正文首次出现位置	当前标题/说明	主要问题	建议处理
Figure 1	3.1 与方法“Participants”均有提及	纳排流程图	正常（建议只在结果引用一次）	保留于 3.1，方法处可不必引用图号。(manuscript.docx)
Table 1	3.2 基线特征	BAR 三分位的基线差异	与正文文字不完全一致（体温显著但正文写不显著；IMV 记作 IVW）；单位/精度需统一	纠正文案与缩写；标准化单位与小数。(manuscript.docx)
Figure 2	3.3 主要结局（30 天）及 3.4 延伸随访	题为 Cox，但注释与正文为 KM 曲线（A–D）	题名-内容不匹配；图注称“numbers at risk”需确保呈现	更名为 KM；若有 Cox 森林图，另立新图或转补充。(manuscript.docx)
Table 2	3.3/3.4	多时间点 Cox 模型	个别拼写/格式错误；“Per 1 Unit increase”需加单位；一处“f”残留	统一为 “Per 1 mg·g⁻¹ increase”；全表格式与小数位统一。(manuscript.docx)
Figure 3	3.3/3.4	RCS（A–D：四时间点）	逻辑与叙述一致	建议在图注中注明节段数与结点位置（knots）。(manuscript.docx)
Figure 4	3.3/3.6	30 天亚组森林图	叙述与图号一致	在图注写明“调整同 Model 3；展示 T3 vs T1”并报告 P_interaction。(manuscript.docx)
Supp. Fig. 1–3	3.4/3.6	60/120/365 天亚组	编号与描述匹配	保持；图注与主文术语一致。(supplementary files.docx)
Supp. Fig. 4–6	机器学习段	DCA、PR、SHAP	与正文存在“Figure 5/6”占位冲突	统一为补充图；主文仅概述要点并给出“见 Supp. Figures 4–6”。(manuscript.docx) (supplementary files.docx)
Supp. Tables	多处	S1–S4 已存在；Boruta 为 S6	编号与正文错配（S5/S6）；S1 内容与正文声称不符	统一补充表编号与内容描述，正文相应修订。(supplementary files.docx)

---

五、可直接粘贴的“修订版”示例文本（关键段落）

3.1 Study population（修订建议稿）
We identified 3,192 adult ICU admissions with cardiogenic shock in MIMIC‑IV. After excluding ICU stay <24 h (n=190), missing height or weight (n=837), and missing BUN or albumin within 24 h (n=1,154), 1,011 unique patients remained for analysis (Figure 1). Patients were stratified into tertiles of BAR using cohort‑specific cut‑points (T1: ≤7 mg·g⁻¹; T2: >7–12 mg·g⁻¹; T3: >12 mg·g⁻¹; n=337 in each). During 30‑day follow‑up from ICU admission, 377/1,011 (37.3%) deaths occurred (primary outcome). （请用实际分位点替换示例切点） (manuscript.docx)

3.2 Baseline characteristics（修订建议稿）
Across ascending BAR tertiles, patients were older and more often male; severity scores increased (SOFA median 7→9; SAPS II 39→51; both P<0.001). AKI/CKD/sepsis/DM/AF were more prevalent in higher tertiles, whereas hypertension and malignant arrhythmia were less frequent in T1. Laboratory trends included higher BUN/creatinine/potassium/INR/lactate and lower albumin/hemoglobin/hematocrit/platelets/sodium/PaO₂ (all P<0.05). CRRT use rose 14%→36% (P<0.001), whereas ACEI/ARB/statins decreased; IMV and β‑blocker usage did not differ significantly (P>0.05). Table 1 provides full distributions. （删除中文与疑问语气，确保与表 1 P 值一致） (manuscript.docx)

3.3 Primary outcome: 30‑day mortality（修订建议稿）
Kaplan–Meier curves by BAR tertiles showed clear separation (log‑rank P<0.001; Figure 2A). In Cox models, BAR remained independently associated with 30‑day mortality: per 1 mg·g⁻¹, HR 1.026 (95% CI 1.011–1.041; P=0.001); T3 vs T1, HR 1.803 (1.262–2.576; P=0.001), with a significant P for trend=0.001 (Table 2). Restricted cubic splines indicated an approximately linear dose–response without an evident threshold (Figure 3A). Subgroup findings were directionally consistent with no significant interactions (Figure 4). (manuscript.docx)

3.4 Secondary outcomes: 60/120/365‑day mortality（修订建议稿）
Associations were stable at 60/120/365 days, both per‑unit and tertile contrasts (all P≤0.001), with sustained KM curve separation (Figure 2B–D), near‑linear RCS patterns (Figure 3B–D), and no effect modification (Supplementary Figures S1–S3). Table 2 summarizes adjusted estimates. (manuscript.docx) (supplementary files.docx)

3.5 Prespecified subgroups and interactions（修订建议稿）
The T3 vs T1 association exceeded unity across demographics, comorbidities, and clinical strata with P_interaction >0.05 throughout (Figure 4; Supplementary Figures S1–S3). (manuscript.docx) (supplementary files.docx)

3.6 Sensitivity analyses and diagnostics（修订建议稿）
Findings were consistent across complete‑case analyses, alternative BAR codings, and trimming strategies; proportional‑hazards and functional‑form checks were satisfactory. Detailed distributions, missing‑data patterns, and univariable associations are shown in Supplementary Tables S1–S4. （统一 MICE 次数；如有需要在补充列出 per‑5 单位 HR 与 CI） (manuscript.docx) (supplementary files.docx)

3.7 Exploratory machine‑learning analysis（修订建议稿）
Eight algorithms yielded favorable PR curves and DCA, and SHAP ranked BAR among top contributors; these internally validated results support, but do not replace, regression findings (Supplementary Figures S4–S6; Supplementary Table S6 for Boruta). （若留主文图，请改号并给出完整图题/图注；否则仅在此概述并指向补充） (manuscript.docx) (supplementary files.docx)

---

六、图题/表题的“定稿式”改写（示例）

• Figure 2（更名）：Kaplan–Meier survival curves by prespecified BAR tertiles at (A) 30, (B) 60, (C) 120, and (D) 365 days after ICU admission; numbers at risk shown below each panel; log‑rank P‑values compare tertiles. （如无风险人数显示，请删除该句）(manuscript.docx)
• Table 2（规范化）：Cox proportional‑hazards models for 30/60/120/365‑day all‑cause mortality by BAR (per 1 mg·g⁻¹ and tertiles). Model 1: unadjusted; Model 2: +age/sex/race; Model 3: +SOFA, SAPS II, AKI, CKD, sepsis, DM, malignant arrhythmia, creatinine, lactate, INR, IMV, CRRT.（统一小数位与 P 值书写）(manuscript.docx)
• Supplementary Table/Figure 编号：按“五–六项”建议统一到 S1–S6，正文相应全部替换。(supplementary files.docx)

---

七、清单式修改任务（便于逐条完成）

1. 统一 BAR 三分位切点 与全稿单位、图表标注（主文与补充一致）。(manuscript.docx)
2. 将 Figure 2 更名为 KM；如另有 Cox 森林图，新设图号或转补充。(manuscript.docx)
3. 删除 3.3 内重复段落，压缩到一版标准叙述。(manuscript.docx)
4. 修正 小节编号：3.5→3.7 顺序重排。(manuscript.docx)
5. MICE 次数统一（m=20 或 m=5 二选一），正文与补充同步。(manuscript.docx)
6. IMV 全文替代 IVW，表 1/表 2/图注/补充一致。(manuscript.docx)
7. 3.2 去中文与疑问句，并与表 1 的显著性结果一致（体温为显著差异）。(manuscript.docx)
8. 表 2 行题改为 Per 1 mg·g⁻¹ increase；清理“f”“加number”等残留；统一 CI、P 值格式。(manuscript.docx)
9. 补充材料编号重排，正文全部更正指向（S1–S6）。(supplementary files.docx)
10. 若 KM 图注保留“numbers at risk”，务必在图下呈现；否则删除该语。(manuscript.docx)
11. 统一变量单位与精度：pH（两位小数）、RBC（×10⁶/µL）、PaO₂/PaCO₂、温度单位（℃或℉二选一且全文一致）。(manuscript.docx)
12. 语言与排版全稿清校：括号、短横线、LaTeX 下标、间距、制表符等。(manuscript.docx)

---

八、简短肯定与投递前建议

• 你的结果段核心信息充分且与表格/图形互相印证：例如 30 天 per‑unit HR 1.026 与 T3 vs T1 HR 1.803 与表 2完全一致，RCS/亚组与延长随访保持方向与显著性一致，体现了良好的稳健性。(manuscript.docx)
• 投递前，建议按上述任务清单完成统一与纠错；此外在结果中补 1–2 句绝对风险（各 tertile 的 KM 累计死亡率）以增强临床可读性；补充材料与正文一一对应尤为关键。(manuscript.docx) (supplementary files.docx)

---

如需，我可以把你“结果”整节按期刊书写规范直接重排成定稿版（含统一编号、图题/表题、标准化统计书写与一次性去重），并给出一套“作者回应审稿人”模板段落，便于后续对审修改。

下面我将以顶级临床期刊（如 NEJM/JAMA/Circulation）常见写法，先给出一版可直接替换到稿件中的英文 Discussion（已按高分期刊逻辑重排并补齐常缺要素）；随后附上逐条修改建议与检查清单，指出还需要在正文或补充材料中补强/统一的细节，便于你一键对照完善。

说明：以下论述严格基于你稿件的结果、图表与补充材料，不引入外部研究结论；涉及具体效应量、亚组与方法细则处均以你文稿与补充表/图为据。(manuscript.docx) (supplementary files.docx)

---

Discussion（journal‑ready, polished）

Principal findings

In this large retrospective cohort of 1,011 adults with cardiogenic shock (CS) drawn from MIMIC‑IV, we found that the blood urea nitrogen‑to‑albumin ratio (BAR) measured within the first 24 hours of ICU admission was independently and positively associated with all‑cause mortality at 30, 60, 120, and 365 days. The association was evident when BAR was modeled continuously and by tertiles, persisted across hierarchical adjustment, and appeared approximately linear on restricted cubic splines without an obvious threshold. At 30 days, each +1 mg·g⁻¹ higher BAR corresponded to a 2.6% higher hazard of death (adjusted HR 1.026, 95% CI 1.011–1.041), and the highest tertile carried an ~80% higher adjusted risk versus the lowest tertile; patterns were directionally consistent at extended horizons. Prespecified subgroups showed no evidence of effect modification. (manuscript.docx)

Interpretation and context

BAR integrates information from two routinely available analytes that capture complementary aspects of CS pathophysiology. BUN reflects prerenal azotemia, neurohormonal activation, and catabolic drive, whereas albumin—a negative acute‑phase reactant—tracks systemic inflammation, capillary leak, and nutritional reserve. A higher BAR therefore plausibly indexes the combined burden of hypoperfusion‑renal handling abnormalities and systemic inflammatory‑metabolic stress that characterizes CS. Notably, BAR retained prognostic value after comprehensive adjustment for conventional risk markers (e.g., SOFA, SAPS II, lactate, creatinine, INR) and remained nearly linear across the observed range, suggesting that risk accrues progressively with rising BAR rather than being confined to an extreme tail. The non‑significant difference between the middle and lowest tertiles, in the context of a significant trend, is compatible with a graded risk pattern in which clinically meaningful excess risk concentrates at higher BAR levels. (manuscript.docx)

Clinical implications

Because BUN and albumin are inexpensive and universally measured, BAR can be incorporated at the bedside on day 1 to complement established scores for early risk stratification and triage. Two practical implications follow. First, reporting absolute risks by tertile (e.g., Kaplan–Meier cumulative mortality at 30/60/120/365 days) would facilitate shared decision‑making and resource allocation (e.g., escalation thresholds for monitoring, hemodynamic reassessment, or renal support). Second, rather than adopting the current cohort‑specific tertile cut‑points as clinical thresholds, future work should define transportable decision thresholds linked to absolute risk and net benefit, preferably through external validation and decision‑analytic modeling. (manuscript.docx)

Methodological strengths

This study leveraged a large, granular ICU dataset, prespecified multivariable models, time‑to‑event methods across multiple horizons, evaluation of functional form with splines, and subgroup/interaction analyses. We also handled missingness using multiple imputation and performed complete‑case and alternative‑coding checks, with concordant estimates across specifications. An exploratory machine‑learning pipeline (with PR curves, decision‑curve analysis, and SHAP explanations) provided convergent evidence that BAR contributes prognostic information beyond traditional covariates, although these analyses were internally validated only. (manuscript.docx) (supplementary files.docx)

Limitations

Several limitations merit emphasis. First, the retrospective, single‑center nature of MIMIC‑IV entails risks of residual confounding and misclassification of CS based on administrative codes. Second, we focused on baseline BAR within 24 hours; dynamic changes and time‑updated effects were not evaluated. Third, some clinically relevant variables (e.g., selected inflammatory markers or lipids) had >20% missingness and were excluded from primary models, potentially reducing model precision; although multiple imputation was applied for variables with ≤20% missingness, imputation cannot fully recover unobserved information. Fourth, treatment timing beyond the first 24 hours was not modeled, so we cannot exclude time‑dependent confounding or immortal‑time bias despite restricting covariates to baseline. Fifth, we did not quantify the incremental value of adding BAR to existing scores using reclassification or discrimination metrics tailored to survival data (e.g., ΔC‑index, IDI, NRI) nor present formal calibration assessments over time. Finally, the machine‑learning results relied on internal cross‑validation without external testing and should be considered supportive, not definitive. (manuscript.docx) (supplementary files.docx)

Future directions

Prospective, multicenter validation is needed to test the transportability of BAR‑based risk stratification in contemporary CS care and to identify clinically actionable thresholds linked to absolute risk and net benefit. Studies should evaluate dynamic BAR trajectories, interactions with renal replacement and albumin administration, and the incremental value of BAR when added to SOFA/SAPS II or CS‑specific models using discrimination, calibration, and net reclassification approaches. Where feasible, robust‑bias analyses (e.g., E‑values) could quantify the strength of unmeasured confounding needed to explain away the observed associations (for example, an adjusted HR of ~1.80 for T3 vs T1 corresponds to an E‑value of ~3.0, with a lower‑bound E‑value ~1.8). (manuscript.docx)

Conclusions

BAR, an inexpensive ratio available within 24 hours of ICU admission, provides durable prognostic information for short‑ and longer‑term mortality in CS, with a near‑linear risk gradient and consistency across subgroups. BAR should be viewed as a complement to—not a replacement for—established scores and laboratory markers; external validation and decision‑analytic work are warranted before adopting specific thresholds in clinical pathways. (manuscript.docx)

---

针对你的原始 Discussion 的逐条修改意见（按高分期刊偏好）

结构与行文（强烈建议）

1. 重排段落顺序为：** Principal findings → Interpretation & context → Clinical implications → Strengths → Limitations → Future directions → Conclusions**（上文已按此顺序重写）。你当前版本把“稳健性/机器学习”穿插在结果复述中，建议移至“Strengths/Future directions”，避免与 Results 重复。(manuscript.docx)
2. 减少结果重复：Discussion 只保留 1–2 处关键效应量（如 30 天 per‑unit HR 与 T3 vs T1 HR），其余效应量用“方向一致、幅度相近”概括，避免与 Results“逐项复述”。(manuscript.docx)
3. 加入绝对风险表述：建议在首段或“Clinical implications”中给出 各 tertile 的 KM 累计死亡率（30/60/120/365 天各 1 行），便于临床判断效益大小（稿件结果已具备数据基础）。(manuscript.docx)

方法与稳健性（应补）
4) 增量价值：新增一段报告 BAR 加入 SOFA/SAPS II 后的 ΔC‑index、IDI、NRI，并绘制随时间的 calibration curve；如版面受限，可置于补充材料并在 Discussion 点出新增益与可信区间。(manuscript.docx)
5) 比例风险与函数形式：正文请简述 Schoenfeld 检验结果（通过/未见违背），并在图注或补充注明 RCS 结点位置与自由度；这两点是高分期刊常被审稿人追问的“可复核性”要素。(manuscript.docx)
6) 缺失与选择偏倚：在 Limitations 中补充“比较纳入 vs 排除（因缺失 BUN/Alb 等）患者的基线差异”的一句，并把对应表格放补充，回应“选择性纳入”的可能性。(supplementary files.docx)
7) 稳健性扩展：可在补充材料再给一组敏感性分析（例如 按 5 mg·g⁻¹ 递增、排除基线 AKI/CKD 或接受 CRRT 的人群），并在 Discussion 一句带过。(manuscript.docx)

机器学习（表述级别调整）
8) 将 ML 的定位从“验证 BAR 的中心地位”调整为“探索性、与回归结果一致的互证”，并明确“仅内部交叉验证，缺少外部测试”。删除夸大性措辞，保留 PR、DCA、SHAP 的“方向一致”结论即可。(manuscript.docx)

术语、单位与一致性（与全文联动）
9) 单位统一：将 “Per 1 unit increase” 统一写为 “Per 1 mg·g⁻¹ increase”；首次出现 BAR 时写全称与单位，并在图表题注同步。(manuscript.docx)
10) 缩写统一：全文与图注统一使用 IMV（invasive mechanical ventilation），删除“IVW”。在 Discussion 避免再次引入新缩写。(manuscript.docx)
11) 切点/分组：如保留 tertiles，请在方法或图注写明 三分位切点为队列内数据驱动，非临床阈值；Discussion 坚持此措辞，避免被审稿人认为“过度临床化”。(manuscript.docx)

限度与因果（把握语气）
12) 全文避免“predictor/causal”混用；在 Discussion 强调“prognostic marker/risk stratification”，明确“风险标志物不等于可干预靶点”，尤其是在提及 albumin 给药或 CRRT 时。(manuscript.docx)

可加的量化语句（选做但加分）
13) 在“Future directions”加入 E‑value 句（示例已写：T3 vs T1 HR≈1.80 → E‑value≈3.0；下限 1.26 → ≈1.8），显示对未测混杂的定量把握，常见于审稿意见的加分点。(manuscript.docx)

---

可直接替换到正文的中文小标题与英文句式模板

• What this study adds（可在正文前加“Key Points”或“Clinical Perspective”框）

  • We show that BAR measured within 24 h of ICU admission independently stratifies short‑ and long‑term mortality risk in CS, with an approximately linear dose–response and consistency across subgroups. (manuscript.docx)
  • Because BAR relies on two ubiquitous tests, it may support early bedside triage when combined with established scores. (manuscript.docx)

• Limitations（两句版，便于压缩字数）

  • This single‑center retrospective analysis is susceptible to residual confounding and coding misclassification; dynamic BAR trajectories and treatment timing were not modeled. (manuscript.docx)
  • Incremental value over SOFA/SAPS II by reclassification and calibration metrics was not quantified and warrants external validation. (manuscript.docx)

---

快速检查清单（给作者与合作者分工用）

• Discussion 段落顺序按上文模板调整，删除与 Results 的逐句重复。(manuscript.docx)
• **加入 4 个时间点的绝对风险（KM 累计事件率）**一句，并提供具体数值（可置补充表/图，正文引用）。(manuscript.docx)
• 补充 ΔC‑index/IDI/NRI 与校准曲线（补充材料展示；正文 1–2 句概述）。(manuscript.docx)
• 报告 Schoenfeld 检验结论并在图注标注 RCS knots。(manuscript.docx)
• 统一单位与缩写（BAR mg·g⁻¹；IMV）；确保 Discussion 与图注一致。(manuscript.docx)
• 纳入 vs 排除（因缺失）患者对比表放补充，正文一语带过。(supplementary files.docx)
• 机器学习定位为探索性；删除过度结论，保留“与回归一致”。(manuscript.docx)
• （可选）E‑value 一句，展示稳健性思维。(manuscript.docx)

---

如果你愿意，我也可以把上述polished Discussion直接替换并在文档中全稿统一单位/缩写、同步图题表题；同时生成一页式**“作者对审稿人回复模板”**（涵盖稳健性、缺失、增量价值、比例风险与外部验证等常见问题），方便投稿后快速响应。 (manuscript.docx) (supplementary files.docx)

下面给出一版按 NEJM/JAMA/Circulation 写作习惯重排并润色的 Methods（可直接替换到稿件中）；随后附上关键修改要点与待补信息清单，帮你把方法学更加严谨、与 NHANES 规范完全对齐。

---

Methods (journal‑ready, polished)

Study design and data source

We performed a cross‑sectional and prospective follow‑up analysis using publicly available data from the National Health and Nutrition Examination Survey (NHANES), a continuous, nationally representative survey conducted by the National Center for Health Statistics (NCHS), Centers for Disease Control and Prevention. NHANES employs a complex, multistage probability design with oversampling of specific subpopulations to improve the precision of estimates. All NHANES protocols were approved by the NCHS Research Ethics Review Board; written informed consent was obtained from all participants. This secondary analysis used de‑identified data and was considered exempt from additional institutional review. Detailed field operations, quality control procedures, and laboratory methods are described in the NHANES documentation.

Study population

We combined eight 2‑year cycles (2003–2018). Of 80,312 participants examined during these cycles, we excluded individuals <20 years of age (n=35,522), those missing all four inflammatory indices (PLR, NLR, MLR, SII) (n=8,600), and those lacking mortality linkage or coronary artery disease (CAD) information (n=3,507), yielding a final analytic sample of 32,683 adults. A flow diagram and detailed counts by exclusion criterion are provided in the Supplement. For participants with partial missingness in covariates, multiple imputation was used as described below.

Data collection and measurements

Trained staff obtained demographic and questionnaire data during household interviews. Mobile Examination Centers (MECs) subsequently conducted standardized physical examinations (e.g., height, weight, waist circumference, and three blood pressure measurements) and venipuncture. For assays requiring fasting, participants followed NHANES protocols (≥8 hours). Laboratory testing included routine chemistry, complete blood count with differential, glucose and lipid profiles.
Inflammation‑related blood cell indices were computed from the same blood draw using standard definitions:

• PLR = platelet count / lymphocyte count
• NLR = neutrophil count / lymphocyte count
• MLR = monocyte count / lymphocyte count
• SII = platelet count × (neutrophil count / lymphocyte count)
  Counts were kept in consistent units (e.g., ×10⁹/L or 10³/µL) across all formulas.

Lifestyle variables were defined a priori. Smoking status was categorized as never (<100 lifetime cigarettes), former (≥100 lifetime cigarettes and quit ≥1 year), or current (≥100 lifetime cigarettes and currently smoking). Alcohol consumption was categorized as none, 1–5 times/month, 5–10 times/month, or >10 times/month. Body‑mass index (BMI) was computed as weight (kg)/height (m)².

Outcomes

The primary cross‑sectional outcome was prevalent CAD, defined as a positive response to any of the following physician‑diagnosed conditions: coronary heart disease, angina pectoris, or myocardial infarction.
Secondary longitudinal outcomes were all‑cause mortality and cardiovascular mortality obtained through NCHS linkage to the National Death Index (NDI) using unique participant identifiers, with follow‑up through December 31, 2019. Time‑to‑event analyses used the interview/examination date as time zero; participants were censored at the end of follow‑up or date of death.

Covariates

Potential confounders were selected a priori based on clinical relevance and prior literature. Demographic and socioeconomic factors included age, sex, race/ethnicity, education level, poverty‑income ratio (PIR), and BMI. Laboratory markers included fasting plasma glucose (FBG), HbA1c, total cholesterol (TC), triglycerides (TG), HDL‑C, and LDL‑C. Lifestyle variables included smoking and alcohol categories. Variable definitions and units, as well as coding details, are provided in the Supplement.

Handling of the complex survey design

All analyses accounted for NHANES’ stratification (SDMVSTRA), clustering (SDMVPSU), and sampling weights using design‑based methods. For multi‑cycle analyses, 2‑year weights were divided by 8 to create 16‑year weights. We used the most restrictive weight applicable to the variables in a given model (e.g., fasting subsample weights for models including fasting labs; MEC examination weights when fasting labs were not included). Variance estimation used Taylor series linearization.

Missing data

For variables with <10% missingness (e.g., PIR, BMI, TG, FBG, HbA1c, HDL‑C, LDL‑C, smoking, alcohol), we performed multiple imputation by chained equations (R mice, version 4.3.0), generating m=10 completed datasets. We used the random forest method for continuous and categorical variables, included the four indices (PLR/NLR/MLR/SII), outcomes, design variables (weights, strata, PSUs) as auxiliary inputs, and restricted predictors to pre‑exposure covariates to avoid look‑ahead bias. Estimates were combined using Rubin’s rules after fitting survey‑weighted models within each imputed dataset. Complete‑case analyses served as sensitivity checks.

Statistical analysis

All analyses were conducted in R 4.3.0 with the survey and survival packages. Because PLR, NLR, MLR, and SII were right‑skewed, we applied natural‑log transformation for continuous modeling; a 1‑unit increase on the log scale corresponds to an e‑fold (~2.72‑fold) increase in the original measure. Each biomarker was analyzed separately to avoid collinearity among indices and was also categorized into quartiles to assess non‑linearity and threshold patterns.

• CAD (primary, cross‑sectional): We used survey‑weighted multivariable logistic regression, reporting odds ratios (ORs) with 95% confidence intervals (CIs).
• Mortality (secondary, longitudinal): We used survey‑weighted Kaplan–Meier estimators and survey‑weighted Cox proportional‑hazards models, reporting hazard ratios (HRs) with 95% CIs for all‑cause and cardiovascular mortality. The proportional hazards assumption was examined using Schoenfeld residuals.
• Functional form and thresholds: We used restricted cubic splines (pre‑specified knots at common quantiles) to examine dose–response; if non‑linearity was detected by likelihood‑ratio testing, we fit two‑segment (piecewise) models to explore potential inflection points.
• Model building and diagnostics: Model 1 was unadjusted; Model 2 adjusted for demographic/socioeconomic variables (age, sex, race/ethnicity, education, PIR, BMI); Model 3 additionally included laboratory and lifestyle covariates (HbA1c, FBG, TG, HDL‑C, LDL‑C, smoking, alcohol). We assessed multicollinearity with variance inflation factors (VIF) and iteratively removed covariates with VIF ≥5.
• Discrimination for CAD: We constructed survey‑weighted ROC curves and calculated the area under the curve (AUC) with 95% CIs via bootstrap resampling.
  A two‑sided α of 0.05 defined statistical significance.

Sensitivity analyses

To assess robustness, we (1) re‑estimated associations in complete‑case samples; (2) repeated CAD analyses after excluding any participants with missingness in covariates; (3) compared continuous versus quartile‑based exposure coding; and (4) restricted mortality analyses to participants with fasting laboratory data using fasting subsample weights.

---

关键修改要点与待补信息清单（供你逐条完善）

1. 权重选择要与最严格子样本对齐：

   • 含空腹血糖/血脂的模型请使用 fasting subsample 权重（WTSAF2YR/8），不含空腹变量的模型使用 MEC 权重（WTMEC2YR/8）；仅问卷变量可用访谈权重（WTINT2YR/8）。同一模型中变量来自不同子样本时，应以最受限的权重为准。

2. 四个炎症指数请分开建模为主分析：

   • PLR/NLR/MLR/SII彼此高度相关，顶级期刊通常分别建模并单独给出主要表格；合并进入同一模型仅可作为敏感性/探索性展示，并报告 VIF。

3. 清楚声明对数变换：

   • 指明为自然对数（ln）；解释“ln 指标每增加 1 单位 = 原始值约增加 2.72 倍”。

4. 复杂抽样的实现要点：

   • 在 mice 插补时把权重、分层、PSU作为辅助变量；在每个插补数据集中构建 survey design 对象后再拟合回归，最后用 Rubin 规则合并（不少稿件容易忽略这一步）。

5. PH 假设与 RCS 细节：

   • 在正文或图注注明 RCS 结点数与位置；在补充材料给出 Schoenfeld 检验或时间交互检验的结果。

6. CAD 定义透明化：

   • 在补充材料列出使用的 NHANES 问卷条目（变量代号），并声明“任一阳性即判定 CAD”。

7. ROC/AUC 需说明权重与区间估计方法：

   • 建议写明“survey‑weighted ROC，CI 由自助法获得（例如 2,000 次重采样）”。

8. 报告样本量与事件数：

   • 在 Methods 或 Results 的首段/图示中给出最终样本量 32,683，以及 CAD 案例数、全因与心血管死亡事件数。

9. 敏感性分析：

   • 建议新增一项：将四个指数按 四分位建模，报告P for trend；或对极端值进行winsorization（如 0.5th/99.5th 百分位）后复算，以显示稳健性。

10. 用语与格式统一：

    • “two‑sided α=0.05”；“OR/HR (95% CI)”；计量单位在正文与表格一致；变量首现处写全称与缩写。

---

如果你愿意，我可以把上述Methods 定稿版直接替换到你的文档中，并按清单补齐权重说明、变量代码表、RCS 结点与 PH 检验结果的图注与补充材料对应关系，确保与图表和统计代码一一对应。