先投稿,再补充结果
생성일: 2025년 9월 14일
답변에 사용된 모델 GPT-5 Thinking by Chat01
생성일: 2025년 9월 14일
답변에 사용된 모델 GPT-5 Thinking by Chat01
有的。基于 MIMIC 的 ICU-AF 与 HGI 相关研究,近一年已见至少两篇:
Scientific Reports(2025):用 MIMIC-IV(并有外部验证队列)纳入 ICU 中的房颤患者,联合评估 HbA1c、HGI、SHR、GV 与 ICU 期间及出院后 28 天死亡率;结果显示 HGI 与结局存在关联,尤其低 HGI 与 28 天死亡风险更高(多模型分析仍显著)。(Nature)
Frontiers in Endocrinology(2025,已接收):用 MIMIC-IV 的 3,882 名危重患者,分析 HGI 与应激高血糖比(SHR) 对 ICU 入院 7 天内新发房颤(NOAF) 的预测价值;报道 HGI 与 NOAF 呈倒 “U” 型关系。(Frontiers)
这两个期刊都是影响因子多少分的文章?
按最新 JCR(2025 年发布、对应 2024 JIF):
补充说明:这些“2024 年 JIF”是基于 2024 年的引用 对 2022–2023 年发表文章的统计,并于 2025 年发布。(Frontiers)
(小提醒:影响因子是期刊层面指标,不代表单篇文章的质量或影响力。)
Frontiers in Cardiovascular Medicine这个期刊多少分数?
Frontiers in Cardiovascular Medicine 的最新影响因子(JCR 2024)是 2.9。(frontiersin.org)
“题目:
Abstract
Background
The hemoglobin glycation index (HGI)—the difference between observed and predicted hemoglobin A1c (HbA1c)—has been linked to adverse outcomes, particularly cardiovascular events. Its prognostic value in atrial fibrillation (AF) remains uncertain. We assessed the association between HGI and 30 , 90 , 180 , and 365 day all cause mortality among critically ill patients with AF admitted to the intensive care unit (ICU).
Methods
Using the MIMIC IV database, we identified ICU admissions with a documented diagnosis of AF. The primary outcomes were 30 day and 365 day all cause mortality; 90 and 180 day mortality were secondary outcomes. HGI was categorized into quartiles. We used Kaplan–Meier curves with log rank tests to compare survival across quartiles; Restricted mean survival time analyses were performed with truncation at 30 and 365 days. Cox proportional-hazards models estimated hazard ratios (HRs), and restricted cubic splines (RCS) evaluated nonlinearity. Prespecified subgroup and interaction analyses (age, gender, body mass index, diabetes , hypertension, chronic kidney disease, prior myocardial infarction) controlled the false discovery rate at 5%.
Results
A total of 2,458 ICU patients were included (median age 73 years [IQR 65–80]; 67.7% male). Patients with lower level of HGI had higher risk of 30-day, 90-day, 180-day and 365-day all-cause mortality, according to the Kaplan-Meier curves (log-rank P < 0.01). In Cox proportional-hazards analyses, patients with lower HGI levels had higher all-cause mortality risk.
RCS indicated an inverse J shaped association, with inflection points at HGI −0.54 (30 day) and −0.59 (365 day); relative to values below the inflection, higher HGI was associated with lower mortality (30 day HR 0.53, 95% CI 0.37–0.76; 365 day HR 0.51, 95% CI 0.37–0.71), and beyond the inflection the curve plateaued near the null.
Conclusion
Among ICU patients with AF, the HGI exhibits an inverse-J shaped relationship with mortality; its integration into clinical evaluation and longitudinal monitoring could identify high risk patients, optimize management, and improve survival.
Keywords: Haemoglobin glycation index, Atrial fibrillation, Prognosis, All-cause mortality,MIMIC-IV database
Introduction
Cardiovascular disease (CVD) has remained a leading global cause of death over recent decades. In 2021, an estimated 20.5 million deaths were attributed to CVD, accounting for roughly one third of all deaths worldwide. Within this burden, atrial fibrillation (AF) is the most common sustained arrhythmia and an established contributor to stroke, heart failure, and premature mortality. AF is common among intensive care unit (ICU) patients ; one study reported a prevalence of approximately 14% among ICU patients. Its presence has been associated with higher in-hospital and post-discharge mortality and greater intensity of care.
ICU pathophysiology includes hemodynamic instability, systemic inflammation, autonomic dysregulation, and exposure to vasoactive therapies. These features heighten arrhythmia risk and complicate clinical decision making. One domain with direct relevance to critical illness is disordered glucose metabolism.
Disordered glucose metabolism is a consistent correlate of adverse outcomes in acute care. Stress hyperglycemia, antecedent chronic hyperglycemia, and greater glycemic variability have each been associated with mortality and organ dysfunction across ICU cohorts. Bedside metrics capture distinct facets of glycemia: admission or mean glucose reflect the current metabolic response to illness and treatment, whereas glycated hemoglobin (HbA1c) integrates average glycemia over the prior 8–12 weeks. Importantly, the relationship between bedside random glucose and HbA1c varies across individuals. For the same glucose exposure, some patients manifest higher or lower HbA1c, a phenomenon termed HbA1c–glucose discordance. Potential contributors include variations in erythrocyte lifespan, iron status, renal function, the inflammatory milieu, and intrinsic glycation kinetics.
The hemoglobin glycation index (HGI) was introduced to quantify this discordance. HGI is defined as the difference between observed HbA1c and the HbA1c predicted from plasma glucose using a population derived regression equation. Conceptually, HGI indexes person specific glycation propensity: a positive HGI indicates that measured HbA1c exceeds what would be expected for a given glucose level, whereas a negative HGI indicates the opposite. HGI is a derived construct rather than a direct laboratory measurement. Its computation requires specification of a prediction model, and it is not routinely reported in most ICU clinical information systems. These attributes influence interpretation and external application but do not undermine the conceptual value of HGI in capturing HbA1c–glucose mismatch.
Evidence indicates that HGI can capture risk beyond absolute glycemia. In sepsis cohorts, higher HGI has been associated with increased all-cause mortality. By contrast, studies in critically ill myocardial infarction and in metabolic disease cohorts have reported greater risk at lower HGI values or U-shaped associations, such that both low and high HGI levels portend adverse outcomes. Despite heterogeneity across designs and populations, these patterns collectively support the view that inter-individual glycation propensity relates to susceptibility to tissue injury and adverse outcomes, and that the association may vary in direction and form according to clinical context.
Despite this background, the clinical relevance of HGI in critically ill adults with AF remains undefined. In ICU settings, AF coexists with circulatory and inflammatory stressors that may interact with glycemic biology. Clarifying whether HGI is associated with mortality in ICU patients with AF would extend inquiry beyond average glucose levels and variability metrics to an index that reflects individual glycation characteristics.
Therefore, the purpose of this study was to evaluate whether the baseline hemoglobin glycation index (HGI) at ICU admission is associated with all-cause mortality among critically ill patients with atrial fibrillation (AF). The findings may help explore approaches for earlier risk identification and for improving prognosis in this high-risk population.
Methods
Source of data
This retrospective cohort study used the Medical Information Mart for Intensive Care IV (MIMIC-IV), version 3.1. Maintained by the Massachusetts Institute of Technology (MIT) Laboratory for Computational Physiology, MIMIC-IV is a credentialed-access, de-identified electronic health record database covering emergency department (ED) encounters and ICU admissions at Beth Israel Deaconess Medical Center (BIDMC). Version 3.1 spans 2008–2022 and includes records for more than 65,000 ICU patients and more than 200,000 ED patients. Use of MIMIC-IV was approved by the Institutional Review Boards of BIDMC and MIT. Data access was granted after completion of National Institutes of Health training on the protection of human participants and successful Collaborative Institutional Training Initiative certification. Because the data are de-identified, informed consent was waived.
Study design and population
The study population consisted of patients with atrial fibrillation (AF) identified using the International Classification of Diseases, Ninth and Tenth Revisions (ICD-9/10); specific AF codes are provided in Additional file 1: Table S1. Exclusion criteria were: (1) age less than 18 years; (2) not the first ICU admission—when patients had multiple ICU admissions, only data from the first hospitalization and its first ICU admission were retained as the index admission, and all non-index ICU stays were excluded; (3) ICU length of stay less than 24 hours; (4) missing HbA1c or no fasting Plasma glucose (FPG) measurements; and (5) missing outcome information. After exclusions, 2,458 patients were included in the final analysis.
Data extraction
Data were extracted from MIMIC-IV using Navicat Premium (version 17.2.4). The variables extracted for analysis were:(1) Baseline characteristics:age, gender, race, body weight, and height; (2) Vital signs: heart rate, systolic/diastolic blood pressure (SBP/DBP), mean arterial pressure (MAP), respiratory rate (RR) , and oxygen saturation (SpO₂); (3) Comorbidities: hypertension, myocardial infarction(MI), heart failure(HF), cerebrovascular disease (CI), chronic pulmonary disease (COPD), chronic kidney disease(CKD), acute kidney injury (AKI), cirrhosis, malignancy, diabetes, and sepsis; (4) Laboratory parameters: hematocrit, hemoglobin, platelet count, red cell distribution width (RDW), red blood cell count (RBC), white blood cell count (WBC), lymphocyte count, neutrophil count, anion gap, calcium, chloride, blood urea nitrogen (BUN), creatinine, bicarbonate, arterial carbon dioxide tension (PaCO₂), pH, arterial oxygen tension (PaO₂), sodium, potassium, prothrombin time (PT), international normalized ratio (INR), alanine aminotransferase (ALT), aspartate aminotransferase (AST), lactate, FPG, and HbA1c. For blood biochemistry variables, the initial measurement was defined as the first value obtained after hospital admission;
(5) disease severity score: Charlson Comorbidity Index (CCI), Simplified Acute Physiology Score II (SAPS II), Oxford Acute Severity of Illness Score (OASIS), Glasgow Coma Scale (GCS), Systemic Inflammatory Response Syndrome (SIRS), Acute Physiology Score III (APS III), and Sequential Organ Failure Assessment (SOFA) ;(6) Therapies and procedures: mechanical ventilation, continuous renal replacement therapy (CRRT), angiotensin-converting enzyme inhibitors/angiotensin receptor blockers (ACEI/ARB), aspirin, clopidogrel, statins, beta-blockers, warfarin, amiodarone, digoxin, and insulin;(7) Outcomes—dates and times of hospital and ICU admission and discharge, date and time of death, and ICU length of stay. The clinical outcomes of this study were defined as 30-day and 365-day all-cause mortality from the moment of admission to the ICU, with 90-day and 180-day all-cause mortality specified as secondary outcomes; all-cause mortality included both cardiovascular and non-cardiovascular deaths. Variables with more than 20 percent missingness were removed; remaining missing values were imputed using a random forest algorithm.
HGI calculation and grouping
A linear regression equation between HbA1c and FPG was developed and the predicted HbA1c was obtained by the equation: predicted HbA1c = 0.006 * FPG (mg/dL) + 5.325
The HGI was then calculated by subtracting the predicted HbA1c from the actual measured HbA1c. Patients were divided into four groups based on HGI quartiles: quartile 1 (Q1), HGI < -0.623; quartile 2 (Q2), -0.623≤ HGI < −0.241; quartile 3 (Q3), −0.241 ≤ HGI < 0.321 and quartile 4(Q4)HGI ≥ 0.321.The correlation between HGI and HbA1c is shown in Fig. S2.
Statistical analysis
Normality of continuous variables was assessed using the Shapiro–Wilk test.
For non-normally distributed data, variables were summarized as median with interquartile range (IQR) and compared across groups using the Kruskal–Wallis test; for two-group comparisons, the Wilcoxon rank-sum test was applied. Categorical variables were compared using the chi-square test; Fisher’s exact test was prespecified for sparse tables but was not required because Cochran’s assumptions were met.
The Kaplan–Meier method was used to calculate the cumulative risk of mortality, and the log-rank test was used to compare differences across the four quartiles of HGI. At 30 and 365 days, we reported Restricted mean survival time (RMST) differences for Q2–Q4 compared with Q1, with two-sided 95% CIs and Wald tests. Cox proportional hazards models were fitted to evaluate the association between HGI, modeled in quartiles (lowest quartile as the reference), and 30-, 90-, 180-, and 365-day all-cause mortality.
Covariates for adjustment were prespecified based on clinical knowledge and prior literature as potential confounders; variables that changed the HGI hazard ratio by >10% when added to the model were retained. Three hierarchical Cox models were prespecified: Model 1 included HGI only; Model 2 additionally adjusted for age, race, gender, body mass index (BMI), diabetes, hypertension, cerebral infarction, heart failure, chronic kidney disease, and myocardial infarction; Model 3 further adjusted for white blood cell count, hemoglobin, platelet count, creatinine, potassium, and pH; SOFA and SAPS II; heart rate and MAP; and use of statins, insulin, ACEI/ARB and amiodarone.
Restricted cubic splines were used to examine potential linear or nonlinear associations between HGI and 30-, 90-, 180-, and 365-day all-cause mortality. Subgroup analyses were performed to assess heterogeneity across prespecified strata. Interaction P values in subgroup analyses were adjusted for multiplicity using the Benjamini–Hochberg false discovery rate (FDR) procedure. Analyses were conducted in R (version 4.3.3) and DecisionLnnc (version 1.0), and two-sided P < 0.05 was considered statistically significant.
Results0908
Baseline characteristics
In total, 2,458 ICU patients were included—median age 73 years (IQR 65–80), 32.3% female, and 64.8% White. Patients were stratified into quartiles of the HGI with near-equal sizes (Q1–Q4: n=614, 614, 614, and 616). Baseline characteristics of the total population and groups stratified by HGI index groups were presented in Table 1. Across HGI quartiles, group differences were significant for clinical characteristics and laboratory indices. Patients in the lowest HGI quartile (Q1; n=614) had the lowest BMI and HbA1c but the highest fasting plasma glucose across quartiles ; they also showed a slightly higher respiratory rate and tended to have a longer ICU stay among the four groups. Compared with the higher HGI index groups., Q1 patients had significantly lower prevalences of diabetes and hypertension, and less frequent use of antiplatelet therapy , statins, warfarin, and insulin. In contrast, Q1 exhibited higher rates of cirrhosis, sepsis, acute kidney injury, and receipt of CRRT. Other comorbidities and treatments were broadly comparable between groups. Across HGI quartiles (Q1–Q4), mortality differed at 30, 90, 180, and 365 days (P < 0.05), with Q1 consistently showing the highest and Q2 the lowest rates: 30-day (11.73% vs 6.35% vs 8.79% vs 8.77%), 90-day (13.03% vs 7.65% vs 10.26% vs 9.42%), 180-day (13.36% vs 7.65% vs 10.42% vs 9.74%), and 365-day (13.36% vs 7.82% vs 10.42% vs 10.23%).
Survival analysis
Kaplan–Meier curves stratified by HGI quartiles showed separation at all time points. The overall log-rank test was significant for both the prespecified primary end points (30-day and 365-day mortality; P = 0.011 and P = 0.015, respectively) and the secondary end points (90-day and 180-day mortality; P = 0.015 and P = 0.0098, respectively). In post hoc pairwise comparisons, Q1 differed from Q2 at 30, 90, 180, and 365 days (P<0.001, P=0.0016, P<0.001, and P=0.0013, respectively), and Q1 differed from Q4 at 90 and 180 days (P=0.045 and P=0.048, respectively). Other pairwise contrasts were not statistically significant. Pairwise P values are summarized in Fig./Table Sx.
RMST analyses were prespecified with truncation at 30 and 365 days. ΔRMST denotes the between-quartile difference relative to Q1. Relative to Q1, Q2 demonstrated longer survival at both time points (30 days: ΔRMST=1.22 days; 95% CI, 0.54–1.89; P<0.001; 365 days: ΔRMST=19.57 days; 95% CI, 7.65–31.50; P=0.001). Differences for Q3 and Q4 were smaller and not statistically significant at either time point (Table 2).
Table 2. 30‑day and 365‑day RMST by HGI quartiles in AF
HGI quartile RMST,days (95% CI) ΔRMST, days (95% CI) P value
30-day RMST
Q1 (reference) 27.61 (27.06–28.17) — —
Q2 28.83 (28.45–29.21) 1.22 (0.54–1.89) < 0.001
Q3 28.12 (27.63–28.62) 0.51 (−0.23–1.25) 0.177
28.23 (27.75–28.71) 0.62 (−0.11–1.35) 0.097
365-day RMST
Q1 (reference) 318.51(309.12–327.91) — —
Q2 338.09(330.75–345.43) 19.57 (7.65–31.50) 0.001
Q3 328.71(320.28–337.14) 10.20 (−2.42–22.82) 0.113
Q4 330.45(322.26–338.64) 11.94 (−0.52–24.40) 0.060
Thirty-day and 365-day restricted mean survival time (RMST) with 95% CI are reported. ΔRMST denotes the difference from Q1 (reference). P values are from pairwise comparisons with Q1. AF, atrial fibrillation; HGI, hemoglobin glycation index; RMST, restricted mean survival time; CI, confidence interval.
The association between HGI and 30-day and 365-day all-cause mortality
Using Cox proportional hazards models, we evaluated the association between HGI quartiles (Q1–Q4; with Q1 as the reference) and 30- and 365-day all-cause mortality (Table 3).
In model 1, for 30-day mortality, only Q2 showed a significant reduction relative to Q1 (HR = 0.52; 95% CI, 0.35–0.77; P = 0.001), whereas Q3 and Q4 were not statistically significant. For 365-day mortality, only Q2 was significant (HR = 0.56; 95% CI, 0.39–0.80; P = 0.002), with Q3 and Q4 not significant.
In model 2, for 30-day mortality, Q2–Q4 were all significant: Q2 (HR = 0.50; 95% CI, 0.34–0.74; P = 0.001), Q3 (HR = 0.64; 95% CI, 0.45–0.93; P = 0.017), and Q4 (HR = 0.61; 95% CI, 0.40–0.94; P = 0.024). For 365-day mortality, Q2–Q4 were likewise significant: Q2 (HR = 0.54; 95% CI, 0.38–0.77; P = 0.001), Q3 (HR = 0.68; 95% CI, 0.48–0.95; P = 0.022), and Q4 (HR = 0.61; 95% CI, 0.41–0.91; P = 0.015).
In model 3, for 30-day mortality, only Q2 was significant (HR = 0.59; 95% CI, 0.40–0.88; P = 0.009), with Q3 and Q4 not significant; for 365-day mortality, only Q2 remained significant (HR = 0.64; 95% CI, 0.44–0.92; P = 0.015), with Q3 and Q4 not significant. Similar patterns were observed for 90- and 180-day mortality (see Supplementary Table S2).
Table 3. Hazard ratios (95% CI) for 30- and 365-day all-cause mortality by HGI quartiles in patients with AF
Model 1 HR (95% CI) P value Model 2 HR (95% CI) P value Model 3 HR (95% CI) P value
30-day mortality
Q1 Reference Reference Reference
Q2 0.52 (0.35–0.77) 0.001 0.50 (0.34–0.74) 0.001 0.58(0.39–0.86) 0.006
Q3 0.74 (0.52–1.05) 0.092 0.64 (0.45–0.93) 0.017 0.82(0.57–1.20) 0.306
Q4 0.74 (0.52–1.05) 0.087 0.61 (0.40–0.94) 0.024 0.94 (0.61–1.44) 0.766
365-day mortality
Q1 Reference Reference Reference
Q2 0.56 (0.39–0.80) 0.002 0.54 (0.38–0.77) 0.001 0.62 (0.43–0.89) 0.009
Q3 0.77 (0.55–1.06) 0.111 0.68 (0.48–0.95) 0.022 0.86 (0.61–1.21) 0.379
Q4 0.75 (0.54–1.04) 0.086 0.61 (0.41–0.91) 0.015 0.92 (0.62–1.37) 0.675
Model 1: Univariate model
Model 2: Adjust: Age, Race, Gender, BMI, Diabetes, Hypertension , Cerebral Infarction , Heart Failure , Chronic Kidney Disease , Myocardial Infarction .
Model 3: Adjust: all Model 2 covariates plus laboratories: WBC, Hemoglobin, Platelets, Creatinine, Potassium, pH; severity scores: SOFA, SAPS II; medications: Statins, Insulin, ACEI/ARB; vital signs: Heart Rate, Mean Arterial Pressure .
Detection of nonlinear relationships
Restricted cubic spline (RCS) analyses were used to examine the nonlinear association between the HGI index and 30-, 90-, 180-, and 365-day all-cause mortality. The RCS curves showed reverse-J-shaped associations for 30- and 365-day mortality (Fig. 3A–B) and for 90- and 180-day mortality (Supplementary Fig. 2A–B).
To further investigate this nonlinear relationship, both Cox proportional hazard models and two-piecewise Cox proportional hazard models were employed (P for log-likelihood ratio < 0.01 in both models) (see Tables 4 and Supplementary Tables 3–4). When HGI was below −0.541 (30-day endpoint) or −0.593 (365-day endpoint), higher HGI was associated with lower all-cause mortality (30-day: HR = 0.53; 95% CI, 0.367–0.757; P < 0.001; 365-day: HR = 0.51; 95% CI, 0.366–0.710; P < 0.001). Above the respective thresholds, the associations were close to null, with 95% CIs spanning 1.00. Analyses at the 90- and 180-day horizons showed comparable threshold-dependent patterns, consistent with the primary endpoints.
(see Supplementary Tables S5).
Table 4. Threshold effect analysis of HGI index on 30-day and 365-day all-cause mortality in AF patients
HR (95% CI) P value
30-day mortality
Fitting by the standard linear model 0.96 (0.83–1.10) 0.54
Fitting by the two-piecewise linear model
Inflection point −0.54
HGI < −0.54 0.53 (0.37–0.76) < 0.001
HGI ≥ −0.54 1.09 (0.94–1.25) 0.25
P for Log-likelihood ratio 0.002
365-day mortality
Fitting by the standard linear model 0.94 (0.83–1.07) 0.37
Fitting by the two-piecewise linear model
Inflection point −0.59
HGI < −0.59 0.51 (0.37–0.71) <0.001
HGI ≥ −0.59 1.07 (0.94–1.22) 0.31
P for Log-likelihood ratio <0.001
HGI, hemoglobin glycation index; The inflection of threshold effect analysis of HGI index on 30-day all-cause mortality was −0.54; The inflection of threshold effect analysis of HGI index on 365-day all-cause mortality was −0.59.
Subgroup analysis
To assess whether the association between HGI and all cause mortality was consistent across clinical contexts, we assessed whether it differed across strata defined by age, gender, BMI, diabetes, hypertension, chronic kidney disease, and myocardial infarction in each time window. Across subgroups, the second HGI quartile (Q2) consistently corresponded to the lowest risk, most clearly in older adults (>65 years) (at 30, 90, 180, and 365 days: P=0.010, P=0.024, P=0.018, and P=0.017, respectively) and in participants with BMI ≥25 kg/m² (at 30, 90, 180, and 365 days: P=0.010, P=0.030, P=0.030, and P=0.047, respectively), and also in those without diabetes (at 30, 180, and 365 days: P=0.021, P=0.031, and P=0.043; borderline at 90 days).(Figure 4; Supplementary Figures S3 and S4).
Taken together, these subgroup patterns persisted after false-discovery-rate control at 5%. Interactions were retained for chronic kidney disease (q=0.0105, 0.0297, 0.0261, 0.0400 at 30, 90, 180, and 365 days, respectively) and myocardial infarction (q=0.0299, 0.0376, 0.0261, 0.0448 at 30, 90, 180, and 365 days, respectively), and for hypertension at 365 days (q=0.0485). Interactions for age, gender, BMI, and diabetes were not retained. Directionally, Q4 was associated with lower risk in participants with prior myocardial infarction and in those with chronic kidney disease, but appeared higher in those without these conditions; for hypertension, the divergence was driven by Q3, indicating lower risk in those without hypertension and higher risk in those with hypertension (Figure 4; Supplementary Figures S3 and S4; Supplementary table).
Discussion0909
Using the MIMIC-IV database, we evaluated associations between ICU-measured HGI and 30-, 90-, 180-, and 365-day all-cause mortality in patients with AF.
HGI showed a reverse J shaped association with all-cause mortality, and this pattern persisted after adjustment for demographics, vital signs, laboratory values, comorbidities, clinical scores, and medications.
Patients with low HGI had higher in-hospital mortality and longer ICU stays.
Below the inflection point, risk increased sharply as HGI decreased;above it, the curve plateaued, and further increases in HGI were associated with minimal additional risk.
These findings underscore the potential clinical relevance of HGI for risk evaluation and stratification and for informing management in ICU patients with atrial fibrillation.
Evidence from large cohorts and meta-analyses indicates that, even after accounting for conventional risk factors, diabetes confers an approximately 30–40% higher risk of incident atrial fibrillation.
Critically ill patients with atrial fibrillation often have complex clinical profiles and face elevated risks of adverse outcomes due to multiple interacting insults .
Admission blood glucose alone cannot capture chronic glycemic exposure because it does not reflect long-term fluctuations.
HbA1c is strongly correlated with mean glucose; however, mean glucose accounts for approximately 60–80% of the variance in HbA1c, leaving roughly 20–40% unexplained by glycemia—a residual that may have prognostic relevance for adverse outcomes .
HGI quantifies this residual component and is calculated as observed HbA1c minus glucose-predicted HbA1c.HGI provides a phenotypic index of glycation propensity and glycemic variability, especially relevant among ICU patients with AF.
The variation captured by HGI likely reflects factors beyond ambient glucose—including erythrocyte redox status, intrinsic glycation activity, genetic susceptibility of red cells, and hemoglobin oxygenation—which may underlie interindividual differences in the HbA1c–fasting glucose relationship.
Accumulating evidence indicates that higher HGI is associated with greater prevalence and incidence of microvascular complications—such as diabetic retinopathy and chronic kidney disease—independent of conventional glycemic measures.
In cardiovascular cohorts, elevated HGI is associated with increased risk of atherosclerotic events and may delineate subgroups that experience differential benefit or harm from intensive glucose-lowering strategies.
Beyond diabetes, higher HGI is associated with insulin resistance, carotid atherosclerosis, and nonalcoholic fatty liver disease in general-population and nondiabetic cohorts, suggesting that HGI captures broader metabolic risk.
Across acute-care settings—including stroke, heart failure, and critical illness—both high and low HGI have been associated with mortality or poor functional recovery, supporting a nonlinear relationship between glycation phenotype and prognosis.
This phenotype may be particularly pertinent among ICU patients with AF.
A coherent mechanistic framework implicates the advanced glycation end products (AGE)–receptor for AGE (RAGE) axis, together with RAAS activation and autonomic–metabolic stress, in driving oxidative injury and profibrotic TGF-β/Smad signaling.Epigenetically mediated metabolic memory sustains hyperglycemia induced signaling beyond episodes of transient hyperglycemia.
AGE-mediated collagen cross-linking stiffens the extracellular matrix and confers resistance to proteolysis, thereby limiting reverse remodeling.
In parallel, endothelial dysfunction and microvascular injury—exacerbated by recurrent stress hyperglycemia—promote nitric oxide depletion, a prothrombotic shift, and conduction heterogeneity.
Insulin resistance and inflamed epicardial adipose tissue add lipotoxic and mitochondrial stress, modulating CaMKII activity, ion-channel expression, and connexins.
Together, structural fibrosis (conduction slowing and spatial dispersion) and ionic remodeling (effective refractory period shortening and afterdepolarizations) supply both substrate and triggers for AF.
These processes converge into positive feedback loops linking glycemic dynamics, oxidative–fibrotic remodeling, and electrophysiologic instability, with the strength of evidence varying across steps.
Against this backdrop, between-study inconsistencies in the HGI likely arise from three domains: (1) non-glycemic determinants of HbA1c; (2) how predicted HbA1c and HGI are derived and modeled; and (3) population characteristics and treatment strategies .
Computational choices also matter: predicted HbA1c has been generated from heterogeneous regressions (e.g., fasting plasma glucose vs. mean glucose monitoring; cohort-specific vs. externally calibrated equations), and HGI has been analyzed as a continuous variable, by quantiles, or by dichotomization—decisions that can alter effect sizes, statistical significance, and even the direction of associations .
Notably, HGI derived from a single fasting glucose captures a different facet of glycemia than CGM-based estimates, a distinction has been highlighted in the DEVOTE secondary analysis.
In addition, differences in populations and glycemic strategies also shape HGI–outcome associations: in the ACCORD trial (intensive target HbA1c < 6.0%), intensive pharmacotherapy was associated with higher all-cause mortality and the trial was terminated early; by contrast, the ADVANCE trial (gliclazide modified-release regimen; target ≤ 6.5%) showed mainly microvascular benefits without excess mortality. Together with an ACCORD secondary analysis suggesting that harms clustered in high-HGI subgroups under intensive therapy, these contrasts illustrate how population and treatment strategy can shape observed risk patterns.
Taken together, heterogeneity in biology, analytics, and population/treatment strategies plausibly explains the divergent findings.
Consistent with this, multiple observational cohorts in critical-care and cardiovascular settings show a reverse-J association between HGI and outcomes, with higher mortality or complications at the low-HGI end (acute myocardial infarction, ischemic stroke, mixed ICU cohorts). Low HGI may index non-glycemic depression of HbA1c or overtreatment-related discordance, both of which can track heightened vulnerability despite apparently favorable HbA1c.
Across 30, 90, 180, and 365 days, adjusted spline curves consistently indicated a reverse J shaped association, with a nadir near HGI ≈ 0 and an inflection around −0.6().
Beyond the inflection, slopes were shallow and confidence bands wide, providing limited evidence for a monotonic increase.
Tests for the overall association and for nonlinearity were significant at the 30 and 365 day horizons;results at other time points were directionally consistent. (Report exact P values if available.)
Threshold models were concordant: below the inflection, each 1 unit increase in HGI (toward 0) was associated with a lower hazard, whereas above the inflection marginal effects were near null (95% CIs crossed 1.0).
Taken together, excess mortality clustered at the low HGI tail, with minimal incremental risk beyond the identified inflection.
Exploratory subgroup analyses—interpreted with multiplicity in mind—suggested effect modification consistent with biological heterogeneity.
Patients without chronic kidney disease or prior myocardial infarction displayed steeper risk gradients at higher HGI, whereas those with CKD or previous MI showed attenuated or reversed, likely because competing pathways dominate their prognosis.
The steeper high-HGI gradients in the strata without CKD or MI suggest that modifiable glycemic–vascular mechanisms are unmasked when irreversible structural disease is absent.
In established CKD, assay interference and uremic, inflammatory, and arrhythmic risks may attenuate the mortality association of HGI, rendering its prognostic signal less apparent.
After prior MI, scar burden, neurohormonal activation, and guideline-directed therapies often dominate outcomes, further attenuating HGI’s relative signal.
By contrast, patients without these fixed injuries may display greater endothelial plasticity and thrombo-inflammatory responsiveness to glycation stress. Other prespecified subgroup analyses (age, gender, BMI, and diabetes status) did not reveal statistically significant interactions with HGI and were not emphasized further. At very low HGI, depression of HbA1c that is independent of ambient glycemia can accompany high physiologic stress. Shortened erythrocyte lifespan during systemic inflammation, iron-restricted erythropoiesis with hepcidin upregulation and reticulocytosis, and uremic carbamylation can create a negative glycation gap, yielding deceptively low HbA1c despite substantial metabolic strain.
This hematologic/immunometabolic milieu—often co occurring with malnutrition–inflammation–catabolism—promotes mitochondrial dysfunction, dysregulated coagulation and vasoplegia, and susceptibility to secondary infection, phenomena poorly captured by glucose indices alone.
The net effect is a spuriously low HbA1c amid high stress, consistent with the disproportionate short-term mortality observed at the low-HGI tail.
When renal and myocardial reserve are preserved, chronically higher HGI plausibly reflects sustained glycation burden and glycemic volatility[45].
Activation of the AGE–RAGE axis augments NADPH oxidase–dependent oxidative stress and depletes nitric oxide bioavailability; glycocalyx injury and endothelial dysfunction tilt toward a prothrombotic, hypofibrinolytic state (elevated PAI-1, denser fibrin networks).
Superimposed sympathetic activation and oscillating glucose generate recurrent oxidative bursts that foster thrombo-inflammation and microvascular injury, while ionic remodeling (e.g., CaMKII signaling, ion-channel and connexin changes) may increase conduction heterogeneity relevant to AF.
This framework explains the modest upward slopes at higher HGI overall and the steeper gradients in subgroups where organ reserve is intact.
Collectively, the data suggest a reserve dependent model: low HGI often reflects illness and assay related depression of HbA1c and acute vulnerability, whereas high HGI indexes glycation mediated vascular risk chiefly when renal and myocardial reserves are intact.
We present this reserve-dependent framework as hypothesis-generating rather than definitive, inviting validation through prospective replications, quasi-experimental designs, and mechanistic interrogation.
Clinically, these data support individualized interpretation of HGI: identify and treat reversible causes of spuriously low HGI in the overall population, while considering dysglycemic and vascular pathways in subgroups where higher HGI retains prognostic salience.
Such heterogeneity motivates targeted evaluation rather than uniform intensification of glucose-lowering therapy.
Mechanistically anchored phenotyping—combining HGI with inflammation indices, renal function, and measures of glycemic variability—may help identify patients most likely to benefit from vascular-protective strategies.
Such approaches should be accompanied by vigilance for spuriously low HbA1c, ensuring that low HGI triggers investigation for anemia, hemolysis, and assay interference.Accordingly, our findings should be interpreted as hypothesis generating and require confirmation in independent, multicenter cohorts with standardized assessment of HbA1c modifiers.
This study has several limitations. First, it is a retrospective, single-center analysis using MIMIC-IV data from Beth Israel Deaconess Medical Center and limited to critically ill patients with AF; therefore, the findings may not be generalizable to the broader U.S. population and may be subject to selection bias. Second, because of the observational design, causal inference regarding the association between HGI and mortality is not possible. Third, lipid parameters were missing in more than 30% of patients and were not analyzed, which may have reduced statistical power and model stability. Fourth, despite extensive adjustment, residual and unmeasured confounding cannot be excluded. Fifth, non-glycemic determinants of HbA1c (e.g., altered red-cell lifespan, recent transfusion, chronic kidney disease) can influence HGI and complicate its interpretation in critical illness. Prospective, multicenter validation with standardized assessment of HbA1c modifiers is warranted.
Conclusion:
In critically ill patients with AF, the HGI reflects glycation propensity beyond mean glucose and is linked to adverse outcomes in a reverse-J pattern. Integrating HGI into bedside assessment may refine risk stratification and support individualized management. Comprehensive HGI monitoring can help clinicians promptly identify high-risk AF patients, select optimal therapeutic strategies, and ultimately reduce mortality while improving prognosis. As a practical indicator for risk stratification and prognostic evaluation, the association between HGI and outcomes in critically ill AF warrants further study.
Table 1.Baseline characteristics of patients grouped according to HGI index quartiles
Variable Total(n=2458) Q1(n=614) Q2(n=614) Q3(n=614) Q4(n=616) P value
Demographics
Age, years 73 (65-80) 71 (64-79) 73 (65-79) 75 (68-81) 72 (65-79) <0.001
Female, n(%) 794 (32.30) 192 (31.27) 208 (33.88) 213 (34.69) 181 (29.38) 0.172
White race, n (%) 1593 (64.81) 401 (65.31) 415 (67.59) 401 (65.31) 376 (61.04) 0.003
Height, cm 173 (165-178) 173 (165-180) 170 (163-178) 173 (163-178) 173 (165-178) 0.019
RR, breaths/min 28.855 (25.312-33.103) 27.561 (24.459-31.531) 28.261 (25.044-32.311) 28.813 (25.294-32.847) 30.67 (26.778-35.123) <0.001
Vital signs
SBP,mmhg 114 (101-129) 113 (99.25-127) 113 (101-128) 115 (101-129) 115 (103-132) 0.063
DBP,mmhg 62 (54-73) 63 (55-75) 63 (54-74) 62 (54-72) 62 (53-73) 0.061
MBP,mmhg 76 (67-87) 76 (68-87) 76 (67-88) 75 (67-85) 76 (67.75-86.25) 0.333
Heart rate, beats/min 80 (74-91) 80 (73-91) 80 (73-90) 80 (73.25-91) 80 (74-91) 0.566
RR, breaths/min 16 (14-20) 17 (14-21) 16 (14-19) 16 (13-19) 16 (14-20) 0.007
SpO₂, % 100 (97-100) 100 (96-100) 100 (97-100) 100 (96.25-100) 99 (97-100) 0.645
Comorbidities
Hypertension, n (%) 1141 (46.42) 262 (42.67) 300 (48.86) 316 (51.47) 263 (42.69) 0.002
Diabetes , n (%) 884 (35.96) 88 (14.33) 85 (13.84) 180 (29.32) 531 (86.20) <0.001
Congestive heart failure, n (%) 1055 (42.92) 263 (42.83) 234 (38.11) 257 (41.86) 301 (48.86) 0.002
Cerebrovascular disease, n (%) 501 (20.38) 119 (19.38) 117 (19.06) 135 (21.99) 130 (21.10) 0.529
Chronic pulmonary disease (COPD), n (%) 336 (13.67) 82 (13.36) 67 (10.91) 108 (17.59) 79 (12.82) 0.006
Myocardial infarction, n (%) 518 (21.07) 140 (22.80) 111 (18.08) 99 (16.12) 168 (27.27) <0.001
Chronic kidney disease, n (%) 552 (22.46) 127 (20.68) 112 (18.24) 117 (19.06) 196 (31.82) <0.001
Sepsis, n (%) 232 (9.44) 73 (11.89) 45 (7.33) 44 (7.17) 70 (11.36) 0.003
Acute kidney injury, n (%) 833 (33.89) 231 (37.62) 182 (29.64) 182 (29.64) 238 (38.64) <0.001
Cirrhosis, n (%) 68 (2.77) 30 (4.89) 11 (1.79) 18 (2.93) 9 (1.46) 0.001
Malignancy, n (%) 381 (15.50) 90 (14.66) 102 (16.61) 103 (16.78) 86 (13.96) 0.425
Clinical scores
SOFA score 5 (3-7) 5 (3-8) 5 (3-7) 5 (3-7) 5 (3-8) <0.001
GCS score 15 (14-15) 15 (14-15) 15 (14-15) 15 (14-15) 15 (14-15) 0.120
OASIS score 32 (27-38) 32 (26-38) 30(25-6.75) 32 (27-38) 33 (28-38) <0.001
SAPS II score 37 (30-46) 37 (30-46) 36 (29-44) 37 (31-44) 39 (31.75-48) <0.001
SIRS score 3 (2-3) 3 (2-3) 3 (2-3) 3 (2-3) 3 (2-3) 0.042
APS III score 39 (30-53) 41 (30-55) 37 (28-49) 38 (30.25-51) 41 (32-57.25) <0.001
Charlson comorbidity index 5 (4-7) 5 (3-7) 5 (3-6) 5 (4-7) 6 (5-8) <0.001
HGI -0.241(-0.623-0.327) -0.899 (-1.124--0.735) -0.423 (-0.527--0.331) -0.004 (-0.141-0.134) 1.212 (0.669-2.094) <0.001
Medications
ACEI/ARB, n (%) 919 (37.39) 214 (34.85) 216 (35.18) 222 (36.16) 267 (43.34) 0.005
Antiplatelet therapy, n (%) 1720 (69.98) 410 (66.78) 451 (73.45) 415 (67.59) 444 (72.08) 0.023
Digoxin, n (%) 130 (5.29) 31 (5.05) 34 (5.54) 35 (5.70) 30 (4.87) 0.903
Amiodarone, n (%) 1046 (42.55) 273 (44.46) 267 (43.49) 260 (42.35) 246 (39.94) 0.412
β-blocker, n (%) 2173 (88.41) 530 (86.32) 557 (90.72) 551 (89.74) 535 (86.85) 0.040
Insulin, n (%) 1432 (58.26) 336 (54.72) 307 (50.00) 293 (47.72) 496 (80.52) <0.001
Warfarin, n (%) 948 (38.57) 210 (34.20) 260 (42.35) 263 (42.83) 215 (34.90) 0.001
Statin, n (%) 501 (20.38) 102 (16.61) 126 (20.52) 139 (22.64) 134 (21.75) 0.046
ICU interventions
Ventilation, n (%) 2311 (94.02) 569 (92.67) 580 (94.46) 580 (94.46) 582 (94.48) 0.449
CRRT, n (%) 159 (6.47) 62 (10.10) 29 (4.72) 22 (3.58) 46 (7.47) <0.001
Laboratory data
Hematocrit, % 30.5 (26.3-35.7) 30.75 (25.9-35.875) 30.8 (26.6-36.2) 30.8 (26.7-35.8) 29.85 (25.975-35.225) 0.163
Hemoglobin, g/dL 10.1 (8.5-11.8) 10.1 (8.4-11.8) 10.2 (8.725-11.9) 10.2 (8.7-11.8) 9.8 (8.5-11.425) 0.060
Platelets, ×10⁹/L 157 (119-210.75) 155 (114-206.75) 154.5 (119-209.75) 157 (119-210.75) 160 (125.75-216.25) 0.732
RDW, % 13.9 (13.1-15.1) 13.8 (13-15.2) 13.7 (13.1-14.7) 14 (13.3-15.1) 14.1 (13.2-15.3) 0.031
RBC, ×10¹²/L 3.36 (2.86-3.91) 3.315 (2.74-3.87) 3.37 (2.89-3.918) 3.4 (2.922-3.928) 3.35 (2.87-3.99) 0.060
WBC, ×10⁹/L 11.8 (8.8-15.6) 12.25 (9-16.275) 11.5 (8.8-15.275) 11.4 (8.8-15.075) 12.2 (8.7-15.6) 0.133
HbA1c, % 5.9 (5.5-6.5) 5.3 (5.1-5.5) 5.6 (5.5-5.8) 6 (5.9-6.2) 7.5 (6.8-8.4) <0.001
Anion gap, mmol/L 12 (10-15) 12 (10-16) 12 (10-14.75) 12 (10-14) 12 (10-15) 0.011
Calcium, mg/dL 8.3 (8-8.7) 8.3 (7.9-8.675) 8.3 (8-8.7) 8.3 (8-8.7) 8.4 (8.1-8.8) 0.101
Chloride, mEq/L 107 (103-110) 106 (102-110) 107 (104-110) 106 (103-110) 106 (102-109) 0.005
Potassium, mmol/L 4.3 (3.9-4.6) 4.3 (3.9-4.7) 4.2 (3.9-4.6) 4.2 (3.9-4.6) 4.4 (4-4.7) 0.008
Sodium, mmol/L 138 (136-140) 138 (135-140) 138 (136-141) 139 (136-141) 138 (136-140) 0.003
Bicarbonate, mmol/L 25 (23-27) 25 (22-27) 25 (24-27) 26 (24-28) 25 (23-27) <0.001
Lactate, mmol/L 1.8 (1.3-2.6) 1.9 (1.3-2.7) 1.7 (1.2-2.4) 1.9 (1.3-2.6) 1.95 (1.4-2.7) <0.001
PaCO₂, mmHg 41 (36-46) 41 (37-46) 40 (36-45) 40 (36-45) 41 (37-46) 0.534
pH 7.39 (7.35-7.43) 7.38 (7.33-7.42) 7.4 (7.35-7.44) 7.41 (7.36-7.44) 7.39 (7.34-7.43) <0.001
PaO₂, mmHg 270 (108-351) 250.5 (93.25-338) 290 (134.25-360) 269 (107-348) 278.5 (111-353) 0.002
Creatinine, mg/dL 1 (0.8-1.3) 1 (0.8-1.4) 0.9 (0.7-1.2) 0.9 (0.8-1.2) 1 (0.8-1.4) <0.001
BUN, mg/dL 18 (14-27) 18 (13-28) 17 (13-22.75) 18 (14-25) 20 (15-31) 0.002
INR 1.4 (1.2-1.6) 1.4 (1.2-1.7) 1.4 (1.2-1.6) 1.4 (1.2-1.6) 1.4 (1.3-1.6) 0.417
PT, s 15.5 (13.6-17.7) 15.6 (13.4-18.1) 15.4 (13.4-17.4) 15.7 (13.8-17.8) 15.5 (13.875-17.4) 0.399
FPG, mg/dL 124 (106-151) 136.5 (115.25-171.75) 120 (106-138.75) 116 (101-137) 128 (107.75-169) <0.001
Length of stay (LOS)
ICU length of stay, days 3.19 (1.54-6.108) 3.635 (1.652-7.148) 3.125 (1.43-5.46) 3.095 (1.672-5.662) 3.185 (1.588-5.787) 0.005
Hospital length of stay, days 10.01 (7.03-15.703) 10.09 (6.795-16.8) 9.745 (7.07-14.8) 9.665 (6.93-14.965) 10.89 (7.83-15.955) 0.001
Outcomes
30-day mortality, n (%) 219 (8.91) 72 (11.73) 39 (6.35) 54 (8.79) 54 (8.77) 0.012
90-day mortality, n (%) 248 (10.09) 80 (13.03) 47 (7.65) 63 (10.26) 58 (9.42) 0.017
180-day mortality, n (%) 253 (10.29) 82 (13.36) 47 (7.65) 64 (10.42) 60 (9.74) 0.011
365-day mortality, n (%) 257 (10.46) 82 (13.36) 48 (7.82) 64 (10.42) 63 (10.23) 0.018
Table note. Continuous variables are reported as median (interquartile range); categorical variables as number of patients (%). Abbreviations: ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; APS III, acute physiology score III; CCI, Charlson comorbidity index; CRRT, continuous renal replacement therapy; FPG, fasting plasma glucose; GCS, Glasgow coma scale; HbA1c, glycated hemoglobin; HGI, hemoglobin glycation index; ICU, intensive care unit; INR, international normalized ratio; MBP, mean blood pressure; OASIS, Oxford acute severity of illness score; PaCO₂, arterial partial pressure of carbon dioxide; PaO₂, arterial partial pressure of oxygen; PLT, platelet count; PT, prothrombin time; RBC, red blood cell count; RDW, red cell distribution width; RR, respiratory rate; SAPS II, simplified acute physiology score; SBP, systolic blood pressure; DBP, diastolic blood pressure; SOFA, sequential organ failure assessment; SpO₂, oxygen saturation; WBC, white blood cell count; BMI, body mass index.
Figure legends
Figure 1. Study flowchart of patient selection
Legend. Patients are identified in the de-identified MIMIC-IV database (2008–2019). Adult ICU patients (≥18 years) with atrial fibrillation and ICU length of stay >24 h are screened. Patients without HbA1c or fasting Plasma glucose (FPG) measured within 24 h of ICU admission are excluded. For individuals with multiple ICU admissions, only the first (index) ICU stay is analyzed. After applying inclusion and exclusion criteria, 2,458 patients remain and are assigned to hemoglobin glycation index (HGI) quartiles for analysis: Q1 n = 614, Q2 n = 614, Q3 n = 614, Q4 n = 616. Abbreviations: MIMIC-IV, Medical Information Mart for Intensive Care IV; AF, atrial fibrillation; ICU, intensive care unit; HbA1c, glycated hemoglobin; FPG, fasting Plasma glucose; HGI, hemoglobin glycation index; K–M, Kaplan–Meier; RCS, restricted cubic spline; HR, hazard ratio; CI, confidence interval; Q, quartile; BMI, body mass index; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; ACEI/ARB, angiotensin-converting enzyme inhibitor/angiotensin receptor blocker; WBC, white blood cell count.
Figure 2. Kaplan–Meier survival curves for all-cause mortality
Legend. Kaplan–Meier (K–M) curves for 30-day (A) and 365-day (B) all-cause mortality stratified by baseline HGI quartiles: Q1 ≤ −0.625; −0.625 < HGI ≤ −0.241 (Q2); −0.241 < HGI ≤ 0.321 (Q3); HGI > 0.321 (Q4); sample sizes Q1 n = 614, Q2 n = 614, Q3 n = 614, Q4 n = 616. Time zero is ICU admission; patients are censored at the end of follow-up for each horizon. Group differences are assessed with the log-rank test (p values shown in panels). Numbers at risk are displayed beneath each plot at prespecified time points.
Figure 3. Restricted cubic spline analysis of HGI and all-cause mortality
Legend. Restricted cubic spline (RCS) curves from Cox proportional-hazards models depict the association between continuous HGI and 30-day (A) and 365-day (B) all-cause mortality. Curves show adjusted hazard ratios (solid line) with 95% CIs (shaded band) across the HGI range, using the minimum-risk value HGI = −0.2877 as the referent (HR = 1.0). Models are adjusted as in multivariable Model-3: age, race, gender, BMI; diabetes, hypertension, cerebral infarction, heart failure, chronic kidney disease, myocardial infarction; laboratories (WBC, hemoglobin, platelets, creatinine, potassium, pH); severity scores (SOFA, SAPS II); medications (statins, insulin, ACEI/ARB); vital signs (heart rate, mean arterial pressure). Four knots are placed at the 5th, 35th, 65th, and 95th percentiles of the HGI distribution. p values for the overall association and for nonlinearity are reported in the panels.
Figure 4. Subgroup analyses (forest plots)
Legend. Forest plots show subgroup-specific HRs (95% CIs) for the association between HGI quartiles and all-cause mortality at 30 days (A) and 365 days (B) from Cox models adjusted as in Model-3. Within each subgroup, Q1 is the reference and HRs for Q2, Q3, and Q4 versus Q1 are displayed; the vertical line denotes HR = 1. p for interaction is obtained from a multiplicative HGI-quartile × subgroup term (values shown in the figure). Prespecified subgroups and cut points: age (≥65 vs <65 years); gender (male vs female); BMI (≥25 vs <25kg/m²); diabetes (yes vs no); hypertension (yes vs no); chronic kidney disease (yes vs no); prior myocardial infarction (yes vs no). Squares indicate point estimates (size proportional to subgroup weight); horizontal bars indicate 95% CIs.”这是我的初稿,你看看我的研究是否与他们两个的研究有重复,有不同和创新吗?
结论先说在前:
与你给出的两篇最新研究相比,你的工作既有重叠(同为 MIMIC‑IV 的 ICU‑AF 人群,关注 HGI 与结局),也有清晰的差异与增量创新。与 Scientific Reports(2025)那篇“AF-ICU 人群多糖代指标与短期死亡”的研究相比,你样本更大、随访更长、方法更精细(RMST、分段 Cox、FDR 控制的交互分析、明确阈值),方向上也聚焦在 HGI 的非线性与阈值;与 Frontiers in Endocrinology(2025,已接收)的“重症人群 7 天 NOAF 风险”研究则结局完全不同(发生 vs. 死亡)、侧重点互补。整体来看,你的稿件属于同主题上的实质性延伸,不是简单重复。
维度 | 你的初稿(本研究) | Scientific Reports 2025(已发表) | Frontiers in Endocrinology 2025(已接收) |
---|---|---|---|
数据源 / 人群 | MIMIC‑IV v3.1;ICU 成人人群,确诊 AF;n=2,458 | MIMIC‑IV + 单中心外部验证(中国滨海县医院);ICU AF 患者;MIMIC n=952,外部 n=286 (Nature) | MIMIC‑IV;重症成人总体,结局是 7 天内新发 AF(NOAF);n=3,882(非既往 AF 队列)(Frontiers) |
暴露/指标 | HGI 为主(基于 FPG 预测 HbA1c:HbA1c_pred = 0.006×FPG+5.325),分位组 + 连续;未与 SHR/GV 联合入模 | 同时评估 HbA1c、HGI、SHR、GV;给出分组、平滑曲线与阈值;GV 为最强预测因子之一 (Nature) | 关注 HGI 与 SHR 对 NOAF 的预测;不涉及 GV 与死亡 (Frontiers) |
主要结局 | 30/90/180/365 天全因死亡;次要:RMST、分段阈值 | ICU 期间死亡 + 入 ICU 后 28 天死亡;并做外部验证及 WQS 加权贡献分析 (Nature) | ICU 入院 7 天内 NOAF 风险(发生率 19.3%)(Frontiers) |
统计方法 | K–M/Log-rank、RMST(30 & 365 天)、多层级 Cox;RCS 非线性;两段式(阈值)Cox;亚组交互并 FDR 校正 | K–M、Cox、多模型、GAM 平滑与阈值分析、WQS、时变 ROC;外部验证 | 多变量 Cox、RCS;关注 U 形/倒 U 形关系(HGI 倒 “U”,SHR 线性反向)(Frontiers) |
关键结果模式 | 低 HGI 风险更高;反 J 型;阈值约 −0.54(30 天)/ −0.59(365 天);Q2 风险最低;Q2 效应在全调模型仍显著 | 低 HGI 与更高 ICU/28 天死亡相关(K–M:ICU 死亡 log‑rank p=0.006);报告 HGI 阈值 ≈ −0.4(ICU 死亡) 与 ≈ 2.2(28 天) 的分段分析;GV 预测力最强;Q2 vs Q1 HR≈0.5(ICU)(Nature) | HGI 与 NOAF 呈倒 “U” 型;SHR 与 NOAF 呈线性负相关;提示两者独立预测 NOAF,可用于分层管理 (Frontiers) |
外部验证 | 无 | 有(单中心队列) (Nature) | — |
与 He 等(Scientific Reports, 2025)基于 MIMIC‑IV 的 ICU‑AF 研究相比,我们的研究在三方面做出了实质性扩展:其一,我们在更大的 AF 队列(n=2,458)中,将结局随访从 ICU/28 天拓展至 90/180/365 天,系统描绘了中远期死亡的时间依赖风险;其二,我们采用 RMST 与两段式 Cox 明确了 HGI ≈ −0.6 的一致阈值,并以 RCS 验证其反 J 型非线性;其三,我们通过FDR 控制的交互分析揭示了肾/心储备对 HGI‑结局关联的情境依赖性。此外,与同期 Frontiers in Endocrinology(2025)关于 NOAF 风险的工作相比,本研究聚焦于既有 AF 的预后,两者在人群与结局层面互补,共同支持 HGI 的非线性/阈值特征及其在重症管理中的潜在应用。(Nature)
小结:你的稿件不是“重复发表”。如果补上文中建议的增值分析与清晰定位,新意将更突出:** 更长随访 + 可操作阈值 + 方法学强化 + 情境异质性**,对 ICU‑AF 的临床分层具有直接意义。
“SHR/GV 作为并列暴露:鉴于 He 等文中 GV 预测力最强、SHR 亦相关,审稿人可能会问“单看 HGI 是否稳健/是否添益”。两条可选策略:”这一点关于SHR的研究已经有了,我还需要进行补充对比吗?在上述已发表的论文中,SHR和HGI谁的效果更好?如果HGI的效果低于GV、SHR,那么我应该如何解释这种现象(这是否反对了我自己的研究结果)
简要回答先上:
一句话定位:把你的论文定位为“在 ICU-AF 人群中,系统刻画 HGI 的非线性-阈值-时程(至 1 年),并在加入 SHR/GV 后仍具有独立且互补的风险信息”。这既不与文献相悖**,也能满足“HGI 是否添益”这一常见审稿问题。
SHR如何计算呢?我现在的问题在于我没有计算过SHR这一指标,立刻计算这个结果短时间内无法解决,我想立刻投稿试试,然后在等待期间进行计算,如果审稿人进行提问的话我再进行补充可以吗?
可以的——你现在就投稿,聚焦 HGI,把 SHR/GV 留作“预先规划的补充分析/未来工作”,完全说得通。等审稿人问到时,再在同一队列上补算 SHR/GV 做敏感性与增量价值分析即可。下面给你两件事:① SHR 怎么算(一步到位);② 稿件里如何“合理地先不做 SHR/GV”,以及被问到时怎么回应。
定义:SHR = 入院(或入 ICU 24h 内)的“应激”血糖 ÷ HbA1c 推算的长期平均血糖(eAG)。(Oxford Academic)
在你现有数据里怎么做
你已经有“入院 24h 内的 FPG 和 HbA1c”。最简便的先行口径:
注意点(写在方法或补充材料里)
可以。把你的研究定位在 HGI 的非线性与阈值(≈−0.6)+ 时程至 1 年这一“深挖”上;同时在文稿中预先交代范围与局限,并说明将于修稿阶段补充 SHR/GV 以评估增量价值。这样既诚实又专业。
可直接用的写法(放引言/方法 & 讨论/局限)
如果审稿人提问时的标准回应(预设答复要点)
Stress hyperglycemia ratio (SHR). We calculated SHR as admission glucose divided by the HbA1c-derived estimated average glucose (eAG). eAG was computed by Nathan’s formula: eAG (mg/dL) = 28.7×HbA1c(%) − 46.7. The primary specification used the first plasma glucose within 24 hours of ICU admission; sensitivity analyses used the mean of all glucose measurements within 24 hours. Units were harmonized to mg/dL. We excluded extreme/unreliable HbA1c values (e.g., <4% or >14%) and repeated analyses after excluding patients with recent transfusion, severe anemia, or end-stage kidney disease. (Oxford Academic)
如果你愿意,我可以把你现有表 1 的变量清单转成一段可直接贴进“将来补做 SHR/GV 的统计计划”的附录大纲。
我需要补充SHR的哪些信息呢?ROC曲线可以吗?作为补充材料不进行那么全面的分析可以吗?至少需要的数据结果是什么?
可以。你把 SHR 放到补充材料做一个“精简版”即可,先保证可复现、看得出独立性/增量性,不需要面面俱到。下面给你一份最小可行清单(MVP)+可选增强项,直接照做就行。
A. 定义与计算(方法里 3–5 行)
B. 一张表 + 一张图 + 两个统计量(补充材料就够)
上述 1 张表 + 1 张图 + 2 个检验,足以让审稿人判断“SHR 是否独立”“是否添益”。AUC(t) 就是你问的 ROC,做时间依赖 ROC更规范(生存数据有删失)。
方法(补充)
We computed the stress hyperglycemia ratio (SHR) as admission glucose divided by the HbA1c-derived estimated average glucose (eAG = 28.7×HbA1c − 46.7, mg/dL). The primary specification used the first plasma glucose within 24 h after ICU admission; sensitivity analyses used the mean of all glucose values within 24 h. Units were harmonized to mg/dL. Patients with potentially unreliable HbA1c (recent transfusion, severe anemia, end-stage kidney disease) were excluded in sensitivity analyses.
结果(补充)
SHR showed [简要方向:正相关/无显著等] with 30- and 365-day mortality in fully adjusted models (per-SD HR …; Q4 vs Q1 HR …). Adding SHR to the HGI model modestly improved discrimination at 30/365 days (ΔC …, p-LRT …) while HGI remained independently associated with mortality (Q2 vs Q1 HR …). Time-dependent AUCs at 30/365 days were: covariates only …; +HGI …; +SHR …; +HGI+SHR …. Calibration was similar across models (calibration slope …; Brier …).
需要,我可以把你现在的协变量列表直接改写成 R 代码骨架(survival + riskRegression/timeROC) 的可复制片段,包含 SHR 计算、四分位切点、Cox、AUC(t)、LRT 与 ΔC 的输出模板。