Correlation of the triglyceride-glucose index and heart rate with 28-day all-cause mortality in severely ill patients: analysis of the MIMIC-IV database

Abstract

Background

Research has identified a link between the triglyceride-glucose index (TyG-i) and the risk of mortality in severely ill patients. However, it remains uncertain if the TyG-i affects mortality by influencing heart rate (HR).

Methods

This study enrolled 3,509 severely ill participants from the Medical Information Mart for Intensive Care (MIMIC-IV) database who had triglyceride, glucose, and HR data upon entering the ICU. Cox regression models were applied to estimate the effect of the TyG-i and HR on 28-day all-cause mortality (ACM) and 28-day in-hospital mortality (IHM). Additionally, Kaplan-Meier (K-M) survival analysis was employed to explore outcome variations among different patient groups. The association of the TyG-i with HR, Sequential Organ Failure Assessment (SOFA) score, and Simplified Acute Physiology Score (SAPS) II was explored through linear regression analysis. Subgroup analysis explored potential interactions among patient characteristics, while sensitivity analysis gauged the robustness of the findings. Additionally, mediation analysis was conducted to assess whether elevated HR acts as an intermediary factor linking the TyG-i to both 28-day ACM and 28-day IHM.

Results

During the 28-day follow-up, 586 cases (16.7%) died from all causes, and 439 cases (12.5%) died during hospitalisation. Cox results showed that individuals with a heightened TyG-i and elevated HR had the highest 28-day ACM (Hazard Ratio 1.70, P-value below 0.001) and 28-day IHM (Hazard Ratio 1.72, P-value below 0.001) compared to those with a reduced TyG-i and HR. The K-M curves showed that individuals with low TyG-i and low HR had the lowest incidence of 28-day ACM and 28-day IHM. The linear analysis results evidenced that the TyG-i was independently connected to HR (beta = 3.05, P-value below 0.001), and the TyG-i was also independently associated with SOFA score (beta = 0.39, P-value below 0.001) and SAPS II (beta = 1.79, P-value below 0.001). Subgroup analysis revealed a significant association in participants without hypertension, the interaction of an elevated TyG-i and HR strongly correlated with a higher 28-day death risk (interaction P-value below 0.05). Furthermore, HR mediated 29.5% of the connection between the TyG-i and 28-day ACM (P-value = 0.002), as well as 20.4% of the connection between the TyG-i and 28-day IHM (P-value = 0.002).

Conclusion

For severely ill patients, the TyG-i is distinctly correlated with HR, and elevated levels of both are strongly connected to greater 28-day ACM and 28-day IHM risks, especially in patients without hypertension. Furthermore, elevated HR mediates the connection between the TyG-i and 28-day mortality.

Introduction

Millions of people worldwide die from critical illnesses each year, and the mortality in ICU can be as high as 40%, posing a significant burden on society [1]. In intensive care settings, several scoring systems are employed to monitor various aspects of patient health: the SOFA score evaluates organ dysfunction, the Acute Physiology and Chronic Health Evaluation (APACHE) II measures disease severity, and the Nutrition Risk in Critically Ill (NUTRIC) score assesses nutritional risk. However, these tools still have limitations in accurately predicting the prognosis of severely sick subjects [2,3,4,5]. Therefore, the continued search for effective prognostic indicators remains essential to guide the supervision of severely sick patients.

Insulin resistance (IR) manifests when peripheral tissues show reduced responsiveness to insulin. The TyG-i serves as a straightforward and reliable marker for detecting IR [6, 7]. Research indicates that severely sick subjects frequently experience more significant insulin sensitivity impairment in the initial phases of hospitalisation compared to healthy individuals [8,9,10]. In the ICU setting, patients often face a heightened risk of severe metabolic disturbances, in which IR plays a key role. IR leads to diminished levels of counter-regulatory hormones like growth hormone (GH), glucagon, glucocorticoids (GC), and catecholamines and also stimulates cytokine production while intensifying both gluconeogenesis and glycogenolysis [11,12,13,14,15]. Tachycardia frequently manifests in the ICU and is linked to a variety of factors, such as hypovolemia, infection, hypoxemia, and drug effects [16,17,18]. HR as a signal of illness severity can markedly influence the mortality of severely ill patients [19, 20]. Even in the general population, HR, every 10 beats per minute (bpm) increase will also increase the risk for ACM by 9% [21].

IR has been extensively shown to be strongly linked with dysfunction of endothelial dysfunction (ED), increased oxidative stress, coagulation disorders, and an inflammatory reaction [22,23,24]. Elevated HR can potentially lead to increased blood pressure and ED, as well as arrhythmias and myocardial ischemia (MI) [25, 26]. Both IR and elevated HR are significant predictors of mortality risk in severe patients. Nonetheless, there exists restricted research on how the TyG-i and HR may interact to affect mortality risk in seriously sick patients. The objective of this research is to analyse how the TyG-i and HR correlate with the 28-day ACM in patients suffering from severe illnesses. Additionally, the research will evaluate whether a heightened HR serves as an intermediary factor linking the TyG-i to both 28-day ACM and 28-day IHM.

Methods

Study population

Data used in this study were sourced from version 2.2 of the MIMIC-IV database, which is maintained by the Laboratory for Computational Physiology at the Massachusetts Institute of Technology (MIT). Data concerning health were derived from the ICUs at Beth Israel Deaconess Medical Center (BIDMC) in Boston [27]. Data access and extraction were performed by Yuekai Shao (certificate number 59828695). the data retrieval process. The Institutional Review Boards (IRBs) at BIDMC granted permission for this investigation and waived the requirement for authorised consent from patients.

The investigation encompassed 40,060 individuals, each 18 years or older, experiencing their initial admission to the ICU. Absence of triglyceride, glucose, and HR data on the first day of ICU entry will be excluded. Ultimately, the final cohort comprised 3,509 subjects (Fig. 1).

Fig. 1

Flow chart of study participants

Data acquisition

Patient baseline data were extracted using Navicat Premium (version 16). Basic demographic details collected included age, gender, ethnicity, and body mass index (BMI). Reported accompanying health conditions comprised diabetes, heart failure(HF), shock, stroke, acute myocardial infarction (AMI), atrial fibrillation (AF), hypertension, and chronic kidney disease (CKD). Key physiological measurements taken included HR, systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse oximetry oxygen saturation (SpO₂). Medication use encompassed hypoglycemic therapy (insulin and other oral agents), calcium channel blockers (CCB), beta-blockers, furosemide, vasopressors, and angiotensin-converting enzyme inhibitors (ACEI/) /angiotensin receptor blockers (ARB). The analysis covered triglycerides (TG), white blood cell (WBC) count, and fasting plasma glucose (FPG) metrics from laboratory tests. Length of stay (LOS) encompassed hospital and ICU stays. Outcomes were assessed through 28-day ACM and 28-day IHM metrics.

TG and FPG were measured as initial values on the first day of ICU intake, while HR, other vital signs, and additional laboratory indicators were the average values measured on the first day of ICU intake. Hypoglycemic therapy was recorded throughout the entire hospital stay, while other medication use was documented on the first day of ICU entry. Height and BMI statistics were excluded due to missing values exceeding 20%. The final dataset had less than 5% missing values, which were addressed through imputation using a random forest algorithm.

Outcomes

The main endpoint was 28-day ACM, with the secondary endpoint being 28-day IHM.

Calculation of TyG-i

The TyG-i was computed by applying the equation: Ln [FPG (mg/dl)×TG (mg/dl)/2] [28].

Statistical analysis

Initial characteristics were detailed utilizing the median along with the interquartile range (IQR) for continuous variables, while categorical variables were described via frequency (proportion). The evaluation of categorical variables used the Pearson chi-square test, while the Wilcoxon rank-sum test was applied to compare continuous variables. Three progressively complex multivariable-adjusted models were developed to analyse the data. The initial model, Model 1, did not include any adjustments. Model 2 incorporated adjustments for demographic factors, including age, gender, ethnicity, and weight. The most comprehensive, Model 3, extended beyond Model 2 to include additional variables such as SpO₂, SBP, DBP, hypoglycemic therapy, beta-blockers, ACEI/ARB, CCB, furosemide, vasopressors, hypertension, diabetes, shock, AMI, AF, HF, stroke, and CKD. Collinearity was examined using variance inflation factors (VIF), all of which were less than 5.

The TyG-i (median 8.913) and HR (median 81.24 bpm) were each grouped into two categories based on their respective medians. Patients were then classified into four distinct groups according to a combined evaluation of TyG-i and HR levels: those with low TyG & low HR, low TyG & high HR, high TyG & low HR, and high TyG & high HR. Cox proportional hazards models evaluated the TyG-i, HR, and their joint assessment in relation to 28-day ACM and 28-day IHM. K-M analysis Curves and Log-Rank tests evaluated event rates and differences between groups. Linear regression models were used to investigate its correlations with HR, SOFA score, and SAPS II. In the analysis, TyG-i served as both a continuous variable and was divided into quartiles.

Sensitivity analyses included adjustments for WBC levels (sensitivity analysis 1) and classification of patients into nine categories based on TyG-i (tertiles) and HR (tertiles) (sensitivity analysis 2). Subgroup analyses examined the impact of TyG and HR levels on 28-day ACM and 28-day IHM across different strata (age, gender, ethnicity, weight, diabetes, hypertension, AF, hypoglycemic therapy, vasopressors) using likelihood ratio tests to assess modification and interaction. Finally, mediation analysis investigated the mediating role of elevated HR in connections between the TyG-i with 28-day ACM and 28-day IHM.

Statistical analyses were conducted using R software (version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria). Results with a P-value less than 0.05 (two-sided) were considered statistically significant.

Results

Baseline characteristics

Based on 28-day survival outcomes, patients were categorised into two groups: survivors and non-survivors (Table 1). Non-survivors, in comparison to survivors, typically exhibited characteristics such as advanced age, elevated HR, lower blood pressure, and a higher incidence of conditions including hypertension, shock, AF, and HF. They had shorter hospital stays, but longer ICU stays. Non-survivors showed higher usage of beta-blockers, CCB, furosemide, and vasopressors, whereas ACEI/ARB and hypoglycemic therapy use were comparatively lower. Furthermore, non-survivors demonstrated higher SOFA scores and SAPS II.

Table 1 Patients’ baseline characteristics grouped by their survival status

TyG-i, HR, and 28-day mortality

Table 2 outlines the comprehensive Cox regression analyses, indicating that for every incremental unit rise in TyG-i, there is an associated 15% increase in the risk of 28-day ACM (P-value = 0.011) and a 17% increase in the risk of 28-day IHM (P-value = 0.007). Similarly, each 10 bpm rise in HR corresponded to a 13% escalation in 28-day ACM (P-value below 0.001) and a 12% elevation in 28-day IHM (P-value below 0.001). In the joint assessment of the TyG-i and HR, subjects with greater TyG-i and elevated HR, relative to those with reduced TyG-i (below the median) and reduced HR (below the median), exhibited the highest rates of 28-day ACM (Hazard Ratio 1.70, P-value below 0.001) and 28-day IHM (Hazard Ratio 1.72, P-value below 0.001).

Table 2 Association between the TyG-i and HR with the risk of deaths

K-M analysis curves

The K-M curves showed that individuals with low TyG-i and low HR had the lowest incidence of 28-day ACM and 28-day IHM (Fig. 2).

Fig. 2

K-M chart of mortality by the TyG-i and HR levels. (A, 28-day ACM; B, 28-day IHM)

Correlation between TyG-i and HR

In linear regression models incorporating the TyG-i and adjusting for confounding variables, HR maintained a significant separate connection with greater TyG-i (Table 3). Specifically, each additional 1-unit rise in the TyG-i is related to a 3.05 bpm raise in HR (P-value below 0.001). When sorting the TyG-i, individuals in the top quartile exhibited an HR 4.81 bpm superior to those in the least quartile (P-value below 0.001).

Table 3 The relationship between the TyG-i and HR

Correlation of TyG-i with SOFA score and SAPS II

Upon adjusting for various confounding factors, both the SOFA score and SAPS II were independently associated with increased TyG-i values (Additional file 1: Table S1). Specifically, Each 1-unit raise in the TyG-i was linked with a 0.39-point enhancement in the SOFA score (P-value below 0.001) and a 1.79-point increase in the SAPS II (P-value below 0.001). When the TyG-i was employed as categorical variables, the SOFA score in the top quartile of TyG-i was 0.92 points greater than the least quartile (P-value below 0.001), and the SAPS II in the top quartile was 4.20 points greater than the least quartile (P-value below 0.001).

Subgroup and sensitivity analyses

In subgroup analyses: In patients without hypertension, elevated TyG-i and HR significantly interacted, heightening the risk of both 28-day ACM and 28-day IHM, with the interaction showing statistical significance (interaction P-value below 0.05). No notable interactions emerged related to age, gender, ethnicity, weight, diabetes, AF, shock, hypoglycemic therapy, and the use of vasopressors (Additional file 1: Fig. S1).

In sensitivity analyses, even after further adjustment for WBC, the top hazard ratios remained largely unchanged (Additional file 1: Table S2). When subjects were segmented into nine categories based on TyG-i tertiles and HR tertiles, those in the top tertiles for both TyG-i and HR continued to exhibit the top hazard ratios (Additional file 1: Table S3).

Mediation analysis

The mediation analysis revealed that an increased HR accounted for 29.5% (P-value = 0.002) of the connection between a higher TyG-i level and 28-day ACM (Fig. 3A). Likewise, elevated HR mediated 20.4% (P-value = 0.002) of the connection between greater TyG-i and 28-day IHM (Fig. 3B).

Fig. 3

The mediating effect of HR on the TyG-i related mortality (A, 28-day ACM; B, 28-day IHM)

Adjusted for age, gender, ethnicity, weight, SBP, DBP, SpO₂, hypoglycemic therapy, beta-blockers, ACEI/ARB, CCB, furosemide, vasopressors, hypertension, diabetes, shock, AMI, AF, HF, stroke, and CKD

Discussion

This investigation is the first to demonstrate an independent connection between the TyG-i and HR in severely ill patients. It also reveals that elevated levels of both factors are significantly linked to increased risks of 28-day ACM and 28-day IHM, particularly in severely ill subjects without concomitant hypertension. Moreover, elevated HR was recognised as a mediator in the connection between TyG-i and both types of mortality within 28 days.

Previous research has indicated that IR correlates with increased mortality risk among severely ill patients. They often demonstrate pronounced IR early in their hospitalisation, with a substantial reduction in insulin sensitivity ranging from 50 to 70% compared to normal subjects [8,9,10]. Studies using the MIMIC database have consistently demonstrated that greater TyG-i values are related to more severe outcomes in critical conditions such as stroke and AF [29, 30]. This connection holds true across the broader severely ill population, where both baseline TyG-i values and dynamic changes in TyG-i significantly predict death risk [11, 31]. IR may result in reduced secretion of counter-regulatory hormones such as growth hormone (GH), glucagon, glucocorticoids (GC), and catecholamines. It can also increase cytokine production and amplify metabolic activities, including gluconeogenesis and glycogenolysis [11,12,13,14,15]. These mechanisms are closely associated with ED, oxidative stress, coagulation imbalance, and inflammatory responses [22,23,24].

HR plays a pivotal role in influencing myocardial oxygen consumption. A meta-analysis involving the general population revealed that a 10 bpm increase in resting HR correlated with a 9% rise in ACM risk [21]. Tachycardia observed in ICU settings can stem from various factors, including arrhythmias, medication effects, infections, or inadequate control of sedation and analgesia, thereby heightening myocardial oxygen consumption [16,17,18]. Elevated HR typically coincides with heightened sympathetic nervous system activity, which can elevate blood pressure, induce vascular oxidative stress, and predispose individuals to arrhythmias and MI [25, 26]. As a critical vital sign, HR is prominently featured in many mortality-predicting physiology-based scoring systems used in ICUs, such as APACHE II and SAPS II [32]. Research by Sander, Park, and others has underscored that elevated HR is linked to increased ICU mortality and prolonged ICU stays among severely ill cases [20, 33]. In contrast, a meta-analysis involving 2,103 severely ill subjects demonstrated that beta-blocker therapy significantly reduced mortality [34].

In the ICU setting, it is frequently noticed that patients exhibit elevated TyG-i and HR. Nevertheless, research on the connection between TyG-i and HR is notably sparse and has predominantly focused on the general population. Poon et al. demonstrated that even in non-diabetic individuals, IR can impair cardiac autonomic function among the elderly [35]. Li et al. conducted a study with 6,078 individuals aged 60 and above, discovering that elevated resting HR coupled with an increase in TyG-i substantially contributed to the onset of cardiovascular disease [36]. Given the critical roles of IR and elevated HR in predicting adverse outcomes in severely ill patients, the hypothesis of this study was validated: elevated HR mediates the link between TyG-i and 28-day ACM as well as 28-day IHM. Several mechanisms underline the intricate relationship between IR, elevated HR, and mortality risk. Studies indicate that IR may elevate sympathetic nervous activity, oxidative stress, and inflammation levels while impairing nitric oxide (NO) synthesis and contributing to ED factors frequently associated with elevated HR [37, 38]. Normally, NO inhibits sympathetic nervous activity, but IR disrupts NO production via the phosphoinositide 3-kinase pathway, leading to heightened HR, shortened diastole, and inadequate myocardial perfusion and filling [39].

Furthermore, it was observed that severe patients without comorbid hypertension were more sensitive to the increased mortality risk associated with elevated TyG-i and HR. Currently, there is a lack of targeted research on this difference. One possible reason for this disparity is that severely sick subjects with comorbid hypertension often have a certain degree of IR and elevated HR prior to admission. Compared to non-hypertensive patients, hypertensive patients may have a greater tolerance for the increased TyG-i and HR in the early acute response upon ICU admission [33, 34]. Notably, seriously sick subjects without a history of hypertension may require stricter blood glucose and HR management.

This investigation has several limitations that should be acknowledged. Firstly, being an observational study, this research cannot definitively establish the causal connection between TyG-i, HR, and death risk in seriously ill cases. Nonetheless, rigorous statistical methods were employed to ensure robust findings. Secondly, due to the nature of data retrieval from a database, it was not possible to verify whether all blood glucose and lipid measurements were obtained under fasting conditions. Additionally, HR was assessed as the average value on the initial day of ICU admission, with varying numbers of measurements per patient. Lastly, the study did not account for the long-term level and variability of HR, potentially causing bias in HR assessment.

Conclusion

The results of this investigation indicate that in severely ill patients, there exists a distinct connection between TyG-i and HR, and elevated levels of both are notably linked to greater risks of 28-day ACM and 28-day IHM, especially among subjects without concurrent hypertension. Moreover, elevated HR was discovered as a mediator in the connection between TyG-i and 28-day mortality. Simultaneously assessing the TyG-i and HR may help identify and differentiate severely ill patients with higher risk levels; this will guide clinicians to implement more accurate glucose, HR regulation strategies, and other therapeutic measures for high-risk individuals to improve their prognosis.