Association between hemoglobin glycation index and mortality in critically ill patients with chronic obstructive pulmonary disease: a retrospective cohort study

Abstract

Background

Chronic obstructive pulmonary disease (COPD) is a critical illness with high intensive care unit (ICU) mortality. Traditional glycemic markers like hemoglobin A1c (HbA1c) poorly reflect individual glucose metabolism variability. The hemoglobin glycation index (HGI), derived from the discrepancy between measured HbA1c and values predicted by fasting plasma glucose (FPG), assesses glycometabolic variability but remains unstudied in critically ill COPD patients.

Methods

This retrospective cohort enrolled 1,125 critically ill COPD patients admitted to the ICU, with data derived from the MIMIC-IV database. HGI was calculated as measured HbA1c minus predicted HbA1c (model: HbA1c =  − 0.0095 × FPG + 5.02) and divided into tertiles. Cox regression models, adjusted for demographics, comorbidities, and clinical parameters, evaluated HGI tertile associations with 30-day, 90-day, and 365-day all-cause mortality. Restricted cubic splines (RCS) and threshold analysis explored nonlinear relationships.

Results

Among 1,125 patients, higher HGI tertiles were independently associated with lower mortality at all timepoints. In multivariable-adjusted Model II (age, sex, ethnicity, hematocrit, hemoglobin, SOFA score, SAPS II score, corticosteroid use, sepsis, and mechanical ventilation, diabetes), compared to the lowest tertile (T1), T2 and T3 showed significantly reduced mortality. For the 30-day mortality, the hazard ratios (HRs) were 0.54 (95% confidence interval [CI] 0.36–0.77, P = 0.0012) for T2 and 0.69 (95% CI 0.46–0.98, P = 0.0396) for T3; for the 90-day mortality, the HRs were 0.59 (95% CI 0.42–0.80, P = 0.0015) for T2 and 0.68 (95% CI 0.50–0.96, P = 0.0274) for T3; and for the 365-day mortality, the HRs were 0.67 (95% CI 0.51–0.87, P = 0.0108) for T2 and 0.73 (95% CI 0.57–0.96, P = 0.0277) for T3, all of which showed significant decreasing trends (all P for trend < 0.05). RCS analysis identified a nonlinear relationship between HGI and 30-day mortality (threshold: HGI = 0.865), with HGI increases below this threshold reducing mortality, and no association above it. Subgroup analyses showed no significant interactions.

Conclusions

Lower HGI levels are independently associated with higher short- and long-term mortality in critically ill COPD patients, with a nonlinear threshold effect. HGI may serve as a novel prognostic biomarker, highlighting the need for personalized glycometabolic management.

Introduction

Chronic obstructive pulmonary disease (COPD), characterized by persistent airflow limitation and progressive respiratory compromise, imposes a significant global health burden, contributing to substantial morbidity, mortality, and healthcare expenditure [1, 2]. Acute exacerbations of COPD often necessitate hospital or intensive care unit (ICU) admission, exacerbating the clinical and economic challenges associated with this condition [3, 4]. While advancements in supportive care have improved outcomes, accurate prognostic stratification remains a clinical imperative, as conventional biomarkers such as C-reactive protein (CRP) and neutrophil-to-lymphocyte ratio (NLR) fail to fully capture the complex interplay between metabolic dysregulation and individual glucose homeostasis variability [5, 6].

The hemoglobin glycation index (HGI) offers a unique perspective on glycometabolic heterogeneity, defined as the discrepancy between measured hemoglobin A1c (HbA1c) and values predicted by fasting plasma glucose (FPG) through a validated linear regression model [7]. In contrast, HGI is essentially the residual of the HbA1c–glucose regression. This metric circumvents limitations of traditional glycemic markers by mitigating confounding effects of erythrocyte turnover, ethnic variations, and iron metabolism on HbA1c interpretation [8]. Elevated HGI values have been associated with atherosclerotic plaque instability, endothelial dysfunction, and systemic inflammation, indicating intricate interactions between metabolism and inflammation that impact multiple organ functions [7, 9]. Nevertheless, the role of HGI in respiratory diseases, particularly in critically ill COPD patients, remains largely unexplored.

Glucose metabolism derangements are prevalent in COPD, with emerging evidence highlighting shared molecular pathways and genetic determinants linking COPD pathogenesis with type 2 diabetes mellitus (T2DM) [10]. Clinical observations indicate that COPD patients with T2DM experience heightened risks of severe exacerbations and increased all-cause mortality, with respiratory-related deaths constituting a significant proportion of these outcomes [11, 12]. Importantly, stress-induced hyperglycemia in critical illness differs mechanistically from chronic hyperglycemia, a distinction not fully captured by conventional metrics, such as random blood glucose or HbA1c, which overlook acute pathophysiological nuances [13]. While the Stress Hyperglycemia Ratio shows potential [14], its reliance on acute glucose measurements overlooks chronic damage from advanced glycation end products (AGEs) [15]. Preclinical research has demonstrated that AGEs exacerbate lung fibrosis and airway remodeling through receptor for AGE-mediated pathways [16]. In COPD, chronic hypoxemia often induces compensatory polycythemia, with hemoglobin level fluctuations strongly correlated with adverse outcomes, such as pulmonary vascular remodeling [17, 18]. These findings suggest that integrating long-term glycometabolic markers like HGI could enhance prognostic evaluation systems.

The primary clinical outcomes in COPD, such as respiratory failure, are intrinsically linked to mechanisms, such as skeletal muscle dysfunction (including respiratory muscles) and impaired ventilatory capacity, which are highly sensitive to long-term metabolic disturbances [19, 20]. Current critical care research predominantly focuses on single-timepoint glucose/HbA1c assessments, lacking longitudinal analysis of glycometabolic profiles [21]. This study sought to determine whether HGI is independently associated with mortality in critically ill COPD patients after adjusting for traditional risk factors. By incorporating HGI into prognostic frameworks, the research aims to offer new insights for early risk stratification and provide evidence supporting personalized glucose management strategies.

Methods

Data source

A retrospective cohort study design was employed, leveraging data from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. This repository contains detailed clinical records of over 60,000 ICU admissions at Beth Israel Deaconess Medical Center between 2008 and 2022 [22]. Institutional Review Board (IRB) approval for database utilization was obtained from both Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology Affiliates. Access was granted to an author (Benji Wang) following completion of certification requirements (certification number: 6182750). Given the de-identified nature of the publicly available data set and absence of direct participant interaction, the IRB waived formal ethical approval, and informed consent was not required.

Study population

Inclusion criteria comprised consecutive adult ICU admissions (age > 18 years) with a primary diagnosis of COPD, identified using ICD-9 codes (491.20, 491.21, 491.22, 496) and ICD-10 codes (J44, J44.0, J44.1). Exclusion criteria were as follows: (1) multiple ICU admissions excluding the first encounter; (2) ICU length of stay < 24 h; (3) missing baseline HbA1c or FBG measurements; and (4) > 10% missing data across individual records. Of the 23,818 initially screened COPD patients, 19,730 were excluded due to missing HbA1c or FBG data, resulting in a final study population of 1,125 patients.

Data extraction and clinical outcomes

Data extraction was performed via structured query language (SQL) to retrieve covariates, including age, sex, ethnicity, heart rate, diastolic blood pressure (DBP), mean arterial pressure (MAP), respiratory rate, temperature, weight, FBG, HbA1c, white blood cell (WBC) count, hematocrit, hemoglobin, platelet count, anion gap, potassium, congestive heart failure, atrial fibrillation, coronary artery disease (CAD), renal disease, liver disease, diabetes, stroke, respiratory failure, acute respiratory distress syndrome (ARDS), pneumonia, length of stay in ICU, and sequential organ failure assessment (SOFA) score. The primary outcome was 30-day all-cause mortality, with 90-day and 365-day mortality defined as secondary outcomes.

HGI calculation and grouping

A population-specific linear regression model was developed to characterize the relationship between HbA1c and FBG, yielding the predictive equation: predicted HbA1c = −0.0095 × FPG + 5.02. The HGI was calculated as the difference between measured and predicted HbA1c values. Figure 1 illustrates the correlation between HGI and HbA1c. Participants were stratified into three groups based on HGI tertile distribution: tertile 1 (T1, HGI < −0.574), tertile 2 (T2, −0.574 ≤ HGI < 0.031), and tertile 3 (T3, HGI ≥ 0.031).

Fig. 1.

Correlation analysis of HGI with HbA1c

Statistical analysis

Categorical variables were summarized as percentages and compared using chi-square tests. Continuous variables, following normality assessment, were reported as medians (interquartile ranges) and analyzed via nonparametric rank-sum tests. Cox proportional hazards models evaluated associations between HGI tertiles and mortality outcomes, presenting hazard ratios (HR) with 95% confidence intervals (CI). The linearity of the association between HGI and 30-day all-cause mortality was assessed using restricted cubic splines (RCS), where nonlinearity was detected, a two-piecewise linear regression model with smooth curve plotting identified potential threshold effects, using recursive maximum likelihood methods for optimal breakpoint determination. Subgroup analyses explored heterogeneity across predefined clinical subgroups. All analyses were conducted in R (version 4.4.2), with two-sided P < 0.05 denoting statistical significance.

Results

Baseline characteristics

Baseline demographic and clinical characteristics of the 1,125 study participants, stratified by HGI tertiles, are presented in Table 1. The highest HGI tertile (T3) was characterized by elevated weight, HbA1c levels, potassium concentrations, and higher prevalence of CAD and diabetes, alongside lower pneumonia incidence. In contrast, the lowest tertile (T1) exhibited the highest respiratory rate, WBC counts, SOFA scores, ICU length of stay, and mortality rates (hospital, 30 days, 90 days, and 365 days), coupled with the lowest hematocrit and hemoglobin values. No significant intergroup differences were observed for sex distribution, ethnicity, DBP, MAP, and temperature, suggesting that HGI stratification captures distinct metabolic, hematological, and outcome-related profiles within the cohort.

Table 1.

Baseline characteristics of the population according to HGI tertiles

Variables	T1 (< −0.574)	T2 (≥ −0.574, < 0.031)	T3 (≥ 0.031)	P value
N	375	375	375
Age, years	70.2 ± 11.0	73.7 ± 10.9	70.1 ± 9.8	< 0.001
Sex, n (%)				0.288
Female	173 (46.1)	154 (41.1)	155 (41.3)
Male	202 (53.9)	221 (58.9)	220 (58.7)
Ethnicity, n (%)				0.124
White	245 (65.3)	275 (73.3)	262 (69.9)
Black	27 (7.2)	16 (4.3)	26 (6.9)
Other	103 (27.5)	84 (22.4)	87 (23.2)
Heart rate, times/minute	84.1 ± 14.1	81.8 ± 13.6	82.9 ± 13.3	0.099
DBP, mmHg	62.9 ± 11.3	62.0 ± 10.9	62.0 ± 11.6	0.478
MAP, mmHg	79.2 ± 10.6	78.4 ± 10.7	78.7 ± 11.1	0.361
Respiratory rate, times/minute	19.9 ± 3.7	18.9 ± 3.2	19.1 ± 3.2	0.003
Temperature, ℃	36.8 ± 0.4	36.8 ± 0.4	36.8 ± 0.4	0.291
Weight, kg	79.9 ± 23.3	81.6 ± 22.9	89.0 ± 25.5	< 0.001
HGI	−1.1 ± 0.5	−0.3 ± 0.2	1.3 ± 1.5	< 0.001
Fasting blood glucose, mg/dL	171.2 ± 94.6	128.4 ± 54.6	158.3 ± 78.7	< 0.001
HbA1c, %	5.6 ± 0.7	6.0 ± 0.5	7.9 ± 1.9	< 0.001
White blood cell, 109/L	11.3 ± 6.1	10.5 ± 8.6	10.6 ± 5.4	0.043
Hematocrit, %	29.8 ± 6.3	31.8 ± 6.9	31.8 ± 6.8	< 0.001
Hemoglobin, g/dL	9.8 ± 2.2	10.5 ± 2.3	10.4 ± 2.2	< 0.001
Platelet, 109/L	179.4 ± 85.9	187.4 ± 86.3	191.0 ± 84.9	0.109
Anion gap, mmol/L	13.0 ± 3.4	12.4 ± 3.2	12.4 ± 3.5	0.042
Potassium, mmol/L	4.0 ± 0.6	4.0 ± 0.6	4.1 ± 0.6	0.012
Congestive heart failure, n (%)				0.116
No	198 (52.8)	216 (57.6)	188 (50.1)
Yes	177 (47.2)	159 (42.4)	187 (49.9)
Atrial fibrillation, n (%)				0.026
No	227 (60.5)	192 (51.2)	219 (58.4)
Yes	148 (39.5)	183 (48.8)	156 (41.6)
CAD				< 0.001
No	202 (53.9)	195 (52.0)	146 (38.9)
Yes	173 (46.1)	180 (48.0)	229 (61.1)
Renal disease, n (%)				0.210
No	276 (73.6)	289 (77.1)	268 (71.5)
Yes	99 (26.4)	86 (22.9)	107 (28.5)
Liver disease, n (%)				0.177
No	348 (92.8)	358 (95.5)	358 (95.5)
Yes	27 (7.2)	17 (4.5)	17 (4.5)
Diabetes, n (%)				< 0.001
No	341 (90.9)	325 (86.7)	235 (62.7)
Yes	34 (9.1)	50 (13.3)	140 (37.3)
Stroke, n (%)				0.008
No	335 (89.3)	305 (81.3)	316 (84.3)
Yes	40 (10.7)	70 (18.7)	59 (15.7)
Respiratory failure, n (%)				0.255
No	273 (72.8)	286 (76.3)	292 (77.9)
Yes	102 (27.2)	89 (23.7)	83 (22.1)
ARDS, n (%)				0.299
No	353 (94.1)	362 (96.5)	357 (95.2)
Yes	22 (5.9)	13 (3.5)	18 (4.8)
Pneumonia, n (%)				< 0.001
No	269 (71.7)	298 (79.5)	311 (82.9)
Yes	106 (28.3)	77 (20.5)	64 (17.1)
Length of stay in ICU, day	4.9 ± 5.9	4.5 ± 5.4	4.0 ± 5.2	0.004
SOFA	5.4 ± 3.5	4.5 ± 3.2	4.6 ± 2.9	< 0.001
Hospital mortality, n (%)	50 (13.3)	26 (6.9)	33 (8.8)	0.010
30-day mortality, n (%)	76 (20.3)	38 (10.1)	46 (12.3)	< 0.001
90-day mortality, n (%)	94 (25.1)	55 (14.7)	61 (16.3)	< 0.001
365-day mortality, n (%)	125 (33.3)	90 (24.0)	95 (25.3)	0.008

DBP diastolic blood pressure, MAP mean arterial pressure, HGI hemoglobin glycation index, HbA1c glycated hemoglobin A1c, CAD coronary atherosclerotic heart disease, ARDS acute respiratory distress syndrome, ICU intensive care unit, SOFA sequential organ failure assessment

Correlation of the HGI with outcome events

Cox proportional hazards regression analysis was performed to evaluate the relationship between HGI tertiles and all-cause mortality at 30, 90, and 365 days, using the T1 group as the reference category (Table 2). Unadjusted models revealed significant inverse associations: both the T2 and highest T3 tertiles demonstrated reduced mortality risk across all timepoints compared to T1. For 30-day mortality, HRs were 0.49 (95% CI 0.33–0.72, P = 0.0003) for T2 and 0.59 (95% CI 0.41–0.85, P = 0.0049) for T3; corresponding values for 90-day mortality were 0.56 (95% CI 0.40–0.78, P = 0.0006) and 0.62 (95% CI 0.45–0.86, P = 0.0041); and for 365-day mortality, 0.67 (95% CI 0.51–0.88, P = 0.0036) and 0.71 (95% CI 0.55–0.93, P = 0.0128), with significant trend tests (P < 0.05 for all). These dose-responsive protective associations remained robust in multivariable-adjusted models: Model I (age, ethnicity, sex) and Model II (additional adjustment for hematocrit, hemoglobin, SOFA score, SAPS II score, corticosteroid use, sepsis, mechanical ventilation, and diabetes) consistently showed lower mortality risks for higher HGI tertiles. Specifically, Model II yielded 30-day HRs of 0.54 (95% CI 0.36–0.77, P = 0.0012) for T2 and 0.69 (95% CI 0.46–0.98, P = 0.0396) for T3; 90-day HRs of 0.59 (95% CI 0.42–0.80, P = 0.0015) and 0.68 (95% CI 0.50–0.96, P = 0.0274); and 365-day HRs of 0.67 (95% CI 0.51–0.87, P = 0.0108) and 0.73 (95% CI 0.57–0.96, P = 0.0277), all with significant trend significance (P < 0.05). These findings indicate that higher HGI values are independently associated with a graded reduction in short- and long-term mortality risk, resilient to confounding by clinical covariates.

Table 2.

HRs (95% CIs) for all-cause mortality across groups of HGI level

HGI level	Non-adjusted		Model I		Model II
	HR (95% CIs)	P value	HR (95% CIs)	P value	HR (95% CIs)	P value
30-day all-cause mortality
Tertiles
< −0.574	1.0 (ref)		1.0 (ref)		1.0 (ref)
≥ −0.574, < 0.031	0.49 (0.33, 0.72)	0.0003	0.45 (0.31, 0.67)	< 0.0001	0.54 (0.36, 0.77)	0.0012
≥ 0.031	0.59 (0.41, 0.85)	0.0049	0.62 (0.43, 0.89)	0.0103	0.69 (0.46, 0.98)	0.0396
P trend	0.0084		0.0171		0.0466
90-day all-cause mortality
Tertiles
< −0.574	1.0 (ref)		1.0 (ref)		1.0 (ref)
≥ −0.574, < 0.031	0.56 (0.40, 0.78)	0.0006	0.51 (0.36, 0.71)	< 0.0001	0.59 (0.42, 0.80)	0.0015
≥ 0.031	0.62 (0.45, 0.86)	0.0041	0.65 (0.47, 0.90)	0.0092	0.68 (0.50, 0.96)	0.0274
P trend	0.0075		0.0174		0.0344
365-day all-cause mortality
Tertiles
< −0.574	1.0 (ref)		1.0 (ref)		1.0 (ref)
≥ −0.574, < 0.031	0.67 (0.51, 0.88)	0.0036	0.59 (0.45, 0.78)	0.0002	0.67 (0.51, 0.87)	0.0108
≥ 0.031	0.71 (0.55, 0.93)	0.0128	0.72 (0.55, 0.95)	0.0182	0.73 (0.57, 0.96)	0.0277
P trend	0.0233		0.0410		0.0294

HR hazard ratio, CI confidence interval

Models were derived from Cox proportional hazards regression models

Non-adjusted model adjust for: none

Adjust I model adjust for: age, ethnicity and sex

Adjust II model adjust for: age, ethnicity, sex, hematocrit, hemoglobin, SOFA, SAPS II, corticosteroid use, sepsis, and mechanical ventilation, diabetes

RCS analysis uncovered a nonlinear relationship between HGI and 30-day mortality risk (P for non-linear = 0.0001; Fig. 2). Threshold effect modeling identified an inflection point at an HGI of 0.865: above this threshold (HGI ≥ 0.865), no significant association with 30-day mortality was observed (HR: 1.12, 95% CI 0.94–1.34, P = 0.2142), whereas below the threshold (HGI < 0.865), each one-unit increase in HGI was associated with a 28% reduction in mortality risk (HR: 0.72, 95% CI 0.60–0.87, P = 0.0007; Table 3). The group with lower HGI (HGI < 0.865) comprised 913 patients with a mortality rate of 14.3%, while the group with higher HGI (HGI ≥ 0.865) included 212 patients with a mortality rate of 13.2% (Fig. 3). The 28% reduction represents the average relative change in hazard per one-unit increase in HGI within the lower segment (HGI < 0.865), and the nonlinearity is modeled via a two-piecewise Cox model, in which the effect of HGI differs before and after the inflection point.

Fig. 2.

RCS analysis of the association between HGI and 30-day all-cause mortality. A horizontal black dotted line serves as the reference hazard ratio (HR = 1), while the orange curve plots the calculated HR values. The orange shaded area corresponds to the 95% CI. The analysis uncovers a non-linear association: mortality risk plateaus above the inflection point (HGI = 0.865), whereas below this threshold, it exhibits a sharp decline

Table 3.

Threshold effect analysis of HGI on 30-day mortality

30-day mortality	HR	95% CIs	P value
Standard linear regression model	0.89	0.78, 1.01	0.0765
Inflection point
< 0.865	0.72	0.60, 0.87	0.0007
≥ 0.865	1.12	0.94, 1.34	0.2142
Log-likelihood ratio test			0.0100

HR hazard ratio, CI confidence interval

Fig. 3.

30-Day mortality rates stratified by the inflection point. Patients were stratified based on an HGI cutoff of 0.865. The group with lower HGI (HGI < 0.865) comprised 913 patients with a mortality rate of 14.3%, while the group with higher HGI (HGI ≥ 0.865) included 212 patients with a mortality rate of 13.2%

Subgroup analysis

Subgroup analyses were conducted to assess potential effect modification of the association between HGI levels and 30-day all-cause mortality across predefined clinical subgroups (Table 4). Stratification by age, sex, ethnicity, vital signs (heart rate, respiratory rate, DBP, MBP, temperature), weight, laboratory parameters (anion gap, platelet count, potassium, WBC count), and comorbid conditions (including congestive heart failure, atrial fibrillation, CAD, stroke, respiratory failure, ARDS, pneumonia, renal disease, liver disease, diabetes) revealed no statistically significant interactions (P for interaction range: 0.0586–0.8511), indicating consistency of the observed association across diverse patient subgroups.

Table 4.

Subgroup analysis of the associations between HGI and 30-day all-cause mortality

Variables	No. of patients	HGI level			P for interaction
		< −0.574	≥ −0.574, < 0.031	≥ 0.031
Age, years					0.3934
< 67.0	375	1.0 (ref)	0.54 (0.24, 1.22)	0.55 (0.27, 1.13)
≥ 67.0, < 76.2	375	1.0 (ref)	0.23 (0.09, 0.62)	0.72 (0.39, 1.34)
≥ 76.2	375	1.0 (ref)	0.50 (0.29, 0.84)	0.53 (0.29, 0.97)
Sex					0.3681
Female	482	1.0 (ref)	0.51 (0.30, 0.89)	0.46 (0.26, 0.80)
Male	643	1.0 (ref)	0.47 (0.27, 0.82)	0.73 (0.45, 1.19)
Ethnicity					0.5159
White	782	1.0 (ref)	0.45 (0.27, 0.75)	0.58 (0.36, 0.93)
Black	69	1.0 (ref)	0.21 (0.05, 2.03)	0.25 (0.03, 2.27)
Other	274	1.0 (ref)	0.65 (0.34, 1.21)	0.71 (0.39, 1.30)
Heart rate, beats/minute					0.0771
< 77.0	374	1.0 (ref)	0.35 (0.16, 0.78)	0.64 (0.32, 1.27)
≥ 77.0, < 87.0	373	1.0 (ref)	0.60 (0.31, 1.17)	0.25 (0.10, 0.61)
≥ 87.0	374	1.0 (ref)	0.53 (0.29, 0.98)	0.82 (0.49, 1.38)
Respiratory rate, beats/minute					0.2227
< 17.0	374	1.0 (ref)	0.31 (0.12, 0.80)	0.41 (0.17, 1.00)
≥ 17.0, < 20.0	373	1.0 (ref)	0.34 (0.15, 0.77)	0.63 (0.33, 1.22)
≥ 20.0	374	1.0 (ref)	0.81 (0.49, 1.34)	0.75 (0.45, 1.25)
DBP, mmHg					0.4314
< 57.0	371	1.0 (ref)	0.40 (0.21, 0.76)	0.37 (0.19, 0.71)
≥ 57.0, < 66.0	371	1.0 (ref)	0.48 (0.24, 0.96)	0.63 (0.33, 1.19)
≥ 66.0	372	1.0 (ref)	0.61 (0.30, 1.24)	0.91 (0.48, 1.71)
MBP, mmHg					0.3301
< 74.0	374	1.0 (ref)	0.33 (0.18, 0.61)	0.40 (0.22, 0.73)
≥ 74.0, < 81.0	373	1.0 (ref)	0.53 (0.26, 1.07)	0.55 (0.28, 1.11)
≥ 81.0	374	1.0 (ref)	0.70 (0.34, 1.41)	0.95 (0.50, 1.81)
Temperature, °C					0.7207
< 36.6	342	1.0 (ref)	0.40 (0.20, 0.82)	0.65 (0.35, 1.18)
≥ 36.6, < 36.9	346	1.0 (ref)	0.41 (0.19, 0.90)	0.42 (0.20, 0.89)
≥ 36.9	344	1.0 (ref)	0.57 (0.30, 1.10)	0.81 (0.44, 1.49)
Weight, kg					0.1877
< 71.0	363	1.0 (ref)	0.37 (0.21, 0.68)	0.46 (0.25, 0.86)
≥ 71.0, < 90.0	355	1.0 (ref)	0.82 (0.39, 1.74)	1.17 (0.58, 2.37)
≥ 90.0	374	1.0 (ref)	0.49 (0.23, 1.04)	0.42 (0.21, 0.86)
Aniongap, mmol/L					0.1151
< 11.0	297	1.0 (ref)	0.41 (0.17, 1.02)	0.24 (0.08, 0.74)
≥ 11.0, < 14.0	426	1.0 (ref)	0.60 (0.31, 1.13)	0.51 (0.26, 0.98)
≥ 14.0	393	1.0 (ref)	0.44 (0.24, 0.79)	0.90 (0.55, 1.46)
Platelet, 109/L					0.3449
< 141	369	1.0 (ref)	0.56 (0.29, 1.08)	0.47 (0.23, 0.95)
≥ 141, < 209	378	1.0 (ref)	0.34 (0.16, 0.69)	0.43 (0.22, 0.83)
≥ 209	374	1.0 (ref)	0.58 (0.30, 1.13)	0.91 (0.51, 1.65)
Potassium, mmol/L					0.2578
< 3.8	354	1.0 (ref)	0.36 (0.18, 0.72)	0.49 (0.25, 0.96)
≥ 3.8, < 4.3	385	1.0 (ref)	0.87 (0.43, 1.77)	0.99 (0.51, 1.92)
≥ 4.3	380	1.0 (ref)	0.36 (0.18, 0.71)	0.46 (0.25, 0.86)
WBC, 109/L					0.6628
< 8.2	370	1.0 (ref)	0.48 (0.23, 1.01)	0.65 (0.32, 1.32)
≥ 8.2, < 11.6	378	1.0 (ref)	0.47 (0.22, 1.00)	0.38 (0.17, 0.85)
≥ 11.6	375	1.0 (ref)	0.58 (0.33, 1.04)	0.78 (0.47, 1.31)
Congestive heart failure					0.7842
No	602	1.0 (ref)	0.55 (0.32, 0.93)	0.57 (0.33, 0.99)
Yes	523	1.0 (ref)	0.43 (0.24, 0.77)	0.60 (0.37, 0.98)
Atrial fibrillation					0.3191
No	638	1.0 (ref)	0.36 (0.20, 0.64)	0.52 (0.32, 0.85)
Yes	487	1.0 (ref)	0.65 (0.38, 1.13)	0.70 (0.40, 1.24)
CAD					0.8511
No	543	1.0 (ref)	0.52 (0.32, 0.86)	0.68 (0.41, 1.12)
Yes	582	1.0 (ref)	0.44 (0.23, 0.82)	0.57 (0.33, 0.98)
Stroke					0.3326
No	956	1.0 (ref)	0.52 (0.33, 0.81)	0.64 (0.43, 0.97)
Yes	169	1.0 (ref)	0.29 (0.13, 0.66)	0.35 (0.15, 0.79)
Respiratory failure					0.3670
No	851	1.0 (ref)	0.49 (0.30, 0.81)	0.50 (0.30, 0.82)
Yes	274	1.0 (ref)	0.51 (0.28, 0.94)	0.83 (0.48, 1.43)
ARDS					0.2121
No	1072	1.0 (ref)	0.50 (0.34, 0.75)	0.54 (0.36, 0.80)
Yes	53	1.0 (ref)	0.40 (0.08, 1.87)	1.28 (0.48, 3.42)
Pneumonia					0.0586
No	878	1.0 (ref)	0.36 (0.21, 0.62)	0.48 (0.30, 0.77)
Yes	247	1.0 (ref)	0.86 (0.49, 1.53)	1.11 (0.63, 1.95)
Renal disease					0.5924
No	833	1.0 (ref)	0.54 (0.35, 0.85)	0.61 (0.39, 0.94)
Yes	292	1.0 (ref)	0.34 (0.14, 0.79)	0.55 (0.28, 1.07)
Liver disease					0.2256
No	1064	1.0 (ref)	0.52 (0.34, 0.77)	0.65 (0.45, 0.95)
Yes	61	1.0 (ref)	0.30 (0.07, 1.39)	0.15 (0.02, 1.18)
Diabetes					0.6299
No	901	1.0 (ref)	0.48 (0.31, 0.73)	0.65 (0.43, 0.99)
Yes	224	1.0 (ref)	0.49 (0.17, 1.42)	0.44 (0.19, 1.03)

DBP diastolic blood pressure, MBP mean blood pressure, WBC white blood cell, CAD coronary atherosclerotic heart disease, ARDS acute respiratory distress syndrome

Discussion

This study explored the association between HGI levels and all-cause mortality at 30-day, 90-day, and 365-day follow-ups in critically ill patients with COPD. The findings demonstrated that higher HGI values were significantly linked to reduced mortality risk compared with the lowest HGI tertile across all timepoints, exhibiting a consistent decreasing trend. Our main findings are counter to our original hypothesis and to the conventional pathophysiological expectation that higher chronic glycemic burden (higher HGI) would be associated with worse outcomes. However, this also suggests an independent correlation, whereby lower HGI levels are associated with heightened short-term and long-term all-cause mortality risk in critically ill COPD patients. In addition, RCS analysis revealed a nonlinear relationship between HGI and 30-day all-cause mortality, with an inflection point at an HGI of 0.865. Below this threshold, each one-unit increase in HGI was associated with a 28% reduction in 30-day mortality risk, whereas no significant association was observed above this value. Subgroup analyses further showed no statistically significant interactions between HGI and predefined subgroups, reinforcing the robustness of these associations. By focusing on a COPD-specific cohort and employing a comprehensive analytical approach, our findings indicate that HGI provides incremental prognostic information for the risk stratification of all-cause mortality. This indicator is distinct from traditional risk factors and exhibits independent potential for clinical assessment, warranting further in-depth investigation in future studies.

HbA1c is formed through the nonenzymatic glycation of intracellular hemoglobin A1 by glucose, yet discrepancies exist between actual and predicted HbA1c levels, with underlying mechanisms remaining unclear [23]. The association between HbA1c and FPG exhibits substantial interindividual variation, potentially influenced by factors affecting glucose metabolism. HGI quantifies this variation in HbA1c [8], serving as an indicator of interindividual glycemic variability that may contribute to microvascular complication risk across populations [24, 25]. Previous research has shown a significant correlation between HGI and all-cause mortality in critically ill coronary artery disease patients, particularly those with low HGI [26]. Another study in ICU-admitted heart failure patients found that HGI was independently associated with increased 30-day and 365-day mortality risks, with high HGI (> 0.709) serving as a valuable marker for identifying high risk individuals [27]. Sarcopenic or cachectic COPD patients may have low HbA1c/HGI despite being at high risk of death [28]. Our finding that lower HGI is associated with higher mortality may partly reflect this phenotype of advanced catabolic state and poor nutritional reserve, and this interpretation aligns with the notion that HGI in COPD might be more of a marker of frailty and chronic systemic disease than of “healthy” glycemic control [29, 30]. Moreover, the correlation between HbA1c and HGI does not preclude HGI from capturing incremental risk information, and our findings suggest HGI may be a more robust marker of long-term dysglycemic burden and “glycation phenotype” than HbA1c alone in critically ill patients.

Our findings in COPD patients align with emerging evidence suggesting that glycation markers reflect complex interactions among glucose metabolism, inflammation, and oxidative stress in chronic lung diseases [31]. For instance, HGI, as a surrogate for hemoglobin glycation, may indirectly reflect systemic carbonyl stress or redox status [32, 33]. In chronic inflammatory states, increased reactive oxygen species (ROS) may enhance hemoglobin glycation while promoting tissue damage and mortality [34, 35]. HGI may also indicate insulin resistance or energy metabolism disorders. A study has shown that HGI mediates death risk via SOFA/Simplified Acute Physiology Score II (SAPS-II), highlighting its link to multi-organ dysfunction [36]. The enrolled patients in this study demonstrated relatively low SOFA scores. It is important to note that the SOFA score was primarily developed to assess organ dysfunction in sepsis [37] and may not fully capture the severity of isolated respiratory failure, which is often the predominant issue in COPD exacerbations [38]. Consequently, the SOFA score might underestimate the true severity of respiratory compromise in this patient population, meaning that many COPD patients could exhibit relatively low SOFA scores despite being at substantial clinical risk. Extremely low HGI has been associated with poor short- and long-term survival in critically ill heart failure patients [39], possibly due to inadequate energy supply to the heart and brain, or exacerbated systemic inflammation. Notably, elevated systemic immune-inflammation indices independently increase death risk in critically ill COPD patients [40], and higher NLR correlates with 28-day mortality [41].

In patients with COPD, corticosteroids (e.g., prednisolone) are frequently used for the management of acute exacerbations [42]. However, their administration may exacerbate the risk of hyperglycemia. By enhancing insulin resistance and promoting hepatic gluconeogenesis, corticosteroids can further elevate blood glucose levels, thereby compounding the underlying chronic glycation status [43]. For instance, poor glycemic control in COPD patients is associated with increased mortality risk [28]. While corticosteroid therapy aims to reduce systemic inflammation, it may inadvertently amplify hyperglycemic effects, particularly in individuals with pre-existing metabolic dysregulation. The HGI serves as a marker reflecting the background chronic glycation state [26]. It may help identify COPD individuals whose chronic metabolic milieu renders them more susceptible to acute insults, such as stressful events or corticosteroid use. Specifically, HGI could potentially function as a predictive tool to pinpoint a subgroup of COPD patients with heightened metabolic vulnerability. These individuals are more likely to experience compounded hyperglycemic complications when exposed to corticosteroid treatment or physiological stressors [44].

Paradoxically, tiers T2 and T3, which exhibited a higher prevalence of obesity and diabetes, were associated with a lower adjusted mortality risk. This finding is consistent with the "obesity paradox" reported in other conditions [45]. Potential explanations for this inverse relationship include the hypothesis of greater metabolic–nutritional reserves providing a survival benefit during periods of physiological stress, the possibility of lead-time bias or more aggressive management in these patients [46]. The nonlinear relationship identified by RCS analysis adds complexity, with the threshold effect at HGI = 0.865 suggesting a biphasic pattern: below this threshold, each HGI unit increase reduces 30-day mortality by 28%, whereas no effect is observed above it. The 28% reduction is an average effect for patients below the threshold (HGI < 0.865). This reduction was consistently observed for each one-unit increase in HGI within this range. This association is not linear and the inflection point represents a shift in the relationship between HGI and mortality, which could suggest different mechanisms at play above and below this threshold. Potential mechanisms include hemoglobin glycation site saturation, altered glycated hemoglobin–inflammation mediator binding, or differential effects on oxygen transport. For example, moderate glycation may enhance hemoglobin oxygen affinity in hypoxic tissues, while excessive glycation could impair erythrocyte deformability or activate pro-inflammatory pathways via AGEs and their receptors (RAGEs) [31, 47, 48].

In diabetes, high HGI reflects a greater glycation burden relative to glucose levels, contributing to AGE accumulation, endothelial dysfunction, and microvascular damage [49, 50]. In COPD, mortality is driven by respiratory failure [51], cardiovascular events [52], sepsis [53], and comorbid cachexia and chronic inflammation [54, 55]. Few studies have directly investigated HGI in COPD; most research on glycemic markers focuses on hyperglycemia or HbA1c, yielding mixed results, particularly during acute exacerbations [56]. Our findings underscore that HGI provides unique information beyond standard glycemic measures. It captures inherent biological variation in hemoglobin glycation propensity, which appears to be highly relevant to COPD pathophysiology and outcomes. The lack of subgroup interactions reinforces the generalizability of these findings across the studied COPD cohort, indicating that HGI’s predictive value is not confined to specific clinical phenotypes. However, HGI should currently be viewed as a risk marker that may reflect chronic metabolic and inflammatory vulnerability, rather than a direct causal mediator of ICU mortality.

This study has several limitations. First, as a retrospective analysis, residual confounding from unmeasured variables (e.g., muscle mass, nutritional markers, and medication use) cannot be fully excluded. Second, the observed association between low HGI and mortality may reflect confounding by underlying conditions, such as frailty or malnutrition, rather than a direct causal relationship. In addition, due to the lack of dynamic blood glucose data, it is impossible to evaluate how changes in HGI over time affect the results. Third, the single-center nature of MIMIC-IV, potential selection and coding bias, the cohort primarily included hospitalized COPD patients, and the lack of follow-up beyond hospital or recorded encounters, limiting generalizability to ambulatory or milder cases. Fourth, the predominantly White composition of our cohort may limit the generalizability of our findings to more diverse populations. Fifth, the biological mechanisms underlying the HGI–mortality association remain speculative and require validation in mechanistic studies (e.g., in vitro erythrocyte glycation models). Finally, these limitations may impact the estimation of patient mortality rates; therefore, future multicenter, prospective studies are warranted to validate our findings. In addition, the identified threshold (HGI = 0.865) requires external validation in larger, independent data sets.

Conclusion

Lower HGI levels are independently associated with higher short- and long-term mortality in critically ill COPD patients, demonstrating a nonlinear threshold effect. We concluded that HGI serves as a valuable prognostic indicator for poor outcomes in critically ill COPD patients, and could facilitate risk stratification for mortality in this population. Given the stress hyperglycemia common in critically ill patients, those with low HGI warrant close monitoring during ICU admission.

Author contributions

L.W.P. formulated the research design, oversaw data collection, conducted analytical procedures, and drafted the results section. F.F.L. and W.W.Z. contributed to study conceptualization, assisted in data acquisition and analysis, and participated in result interpretation during manuscript development. B.H.C. and J.W. provided critical insights into result interpretation and performed detailed manuscript revisions. B.J.W. offered methodological guidance on framework implementation for this study and assumed primary responsibility for final content validation. All co-authors engaged in iterative manuscript review and approved the final version.

Funding

This study received financial support from the Wenzhou Municipal Basic Scientific Research Project (Grant No. Y20220495) and the Zhejiang Provincial Natural Science Foundation of China (Grant No. LMS25H150004).

Data availability

Clinical datasets supporting the study’s findings were derived from the MIMIC-IV. While this database is publicly accessible at no cost, prospective users must complete the National Institutes of Health’s online training program, "Protecting Human Research Participants," as a prerequisite for accessing research permissions.

Declarations

Ethics approval and consent to participate

Ethical approval for the MIMIC-IV database was granted by the institutional review boards (IRBs) at Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology. Since the database contains no protected health information, the IRBs authorized a waiver of the informed consent requirement as part of the approval process.

Competing interests

The authors declare no competing interests.