This retrospective observational cohort study was based on the eICU Collaborative Research Database (eICU-CRD), an online international database in the United States (9). The eICU-CRD is a public database created through collaborative efforts between the MIT Laboratory for Computational Physiology and the eICU Institute (9). It contains high-resolution data of over 200,000 ICU admissions monitored across the United States as part of the eICU program from 2014 to 2015 (9). In accordance with the data use agreement of the PhysioNet review committee, the requisite examinations were completed and certification was obtained to enable access to the database (record ID: 40859994). In accordance with the Safe Harbor provision of the U.S. Health Insurance Portability and Accountability Act (HIPAA), the database utilized in this study was de-identified, and all protected health information was removed to ensure the safeguarding of institutional and patient privacy. This retrospective study did not involve direct patient intervention and was exempted from the safety protocol of the PhysioNet review committee. In light of the aforementioned reasons, the requirement for informed consent was not applicable. All methods used in this study were conducted in accordance with the tenets of the Declaration of Helsinki (10).

The study population included all patients diagnosed with congestive heart failure (CHF) upon ICU admission. Specifically, patients were identified using the International Classification of Diseases, 9th Revision (ICD-9) code 428.0, as recorded in the database admission table (11). Study subjects were selected using the following exclusion criteria: (1) patients not on their first ICU admission, (2) patients with an ICU stay of less than 24 h, (3) patients under the age of 18 years, (4) patients lacking data monitoring during ICU treatment, and (5) patients with data having systematic errors or missing HR/T ratio after ICU admission. A flowchart of the inclusion and exclusion criteria is shown in Figure 1.

A Structured Query Language (SQL) was employed to extract data from the eICU-CRD. The extracted data included vital signs, such as heart rate, body temperature, respiratory rate (RR), and mean arterial pressure (MAP), which were recorded as the worst initial measurements within the first 24 h of ICU admission. The definition of the “worst” vital signs refers to the values that are furthest from the midpoint defined by Acute Physiology and Chronic Health Evaluation (APACHE; temperature = 38°C, heart rate = 75 beats per minute, RR = 19 breaths per minute, MAP = 90 mmHg) (9). The baseline characteristics of the patients, including sex, age, race, and body mass index (BMI), were collected from the patient and the patient result tables. Laboratory data, including alanine aminotransferase (ALT), aspartate aminotransferase (AST), Blood urea nitrogen levels (BUN), red blood cell count, hematocrit (Hct), and hemoglobin levels, were collected from the corresponding laboratory tables. The presence of comorbidities, including chronic obstructive pulmonary disease (COPD), acute myocardial infarction (AMI), liver failure, leukemia, and immunosuppression, was determined based on the APACHE IV scores. Treatment given at baseline included intubation, mechanical ventilation, dialysis, and β-blockers. The initial recorded value within the first 24 h of ICU admission was set as the baseline for analysis, using the measurement time as a reference point. The severity of the patients’ conditions upon admission was evaluated using several standardized scoring systems, including the Acute Physiology Score III, APACHE IV score, Glasgow Coma Scale (GCS) score, and Sequential Organ Failure Assessment (SOFA) score. The HR/T ratio was calculated by dividing the heart rate (beats per minute) by body temperature (°C). Patients were divided into quartiles based on their HR/T ratio: Q1 (1.28–1.76), Q2 (2.44–2.72), Q3 (2.88–3.14), and Q4 (3.29–4.13).

The primary outcome measure was 28-day mortality after ICU admission.

Statistical analyses were conducted using the software packages R (version 4.2.0, R Foundation for Statistical Computing, Vienna, Austria) and EmpowerStats (X&Y Solutions Inc., Boston, MA, United States).¹ Descriptive statistics were used to summarize baseline patient characteristics. Continuous variables are presented as mean ± standard deviation (SD) or median with interquartile range (IQR). One-way analysis of variance (ANOVA) was used to ascertain discrepancies among the quartiles of the HR/T ratio in continuous variables. The Kruskal–Wallis test was used to evaluate differences in non-normally distributed continuous variables across groups. Categorical variables are expressed as absolute values and percentages, and differences were compared using the Chi-Square test or Fisher’s exact test, as appropriate.

To address potential confounding factors, three distinct adjustment models were constructed to assess the relationship between HR/T and 28-day mortality. The adjustment for covariates was based on the clinical significance of confounders (12) and a change of at least 10% in the matched odds ratio (OR) (13). The results are presented as OR with 95% confidence intervals (95% CI). The HR/T ratio was analyzed as a categorical variable based on quartiles (Q1-Q4), with the lowest quartile serving as the reference category. The p values for the trends across the HR/T ratio quartiles were calculated for all three models. Additionally, generalized additive models (GAM) were employed to illustrate the correlation between the HR/T ratio and mortality, with the findings presented through smoothed curve-fitting plots. We calculated the E-value to determine the potential influence of unmeasured confounders on the association between HR/T and 28-day ICU mortality (14). The E-value quantifies the extent to which unmeasured confounders potentially nullify the observed association between HR/T and 28-day ICU mortality. For all analyses, statistical significance was set at p < 0.05.

A total of 3,790 patients were categorized by HR/T ratio quartiles and comparison of demographic characteristics, vital signs, laboratory test results, disease severity, and comorbidities across different quartiles is presented in Table 1. Compared with patients in the lowest HR/T ratio quartile, those in Q4 were generally younger and exhibited a lower BMI, while sex and racial distributions were relatively balanced. Significant intergroup differences were observed in laboratory parameters, including AST, ALT, hemoglobin, and Hct, across the different HR/T ratio groups (p < 0.05).

Overall, the 28-day ICU mortality rate was 5.01% (190/3790). Notably, the 28-day mortality rate increased significantly with a higher HR/T ratio, from 3.69% in Q1 to 7.49% in Q4 (p < 0.001).

A univariate logistic model was used to demonstrate the unadjusted associations between the baseline variables and 28-day mortality in patients in ICU (Table 2). There was a statistically significant positive correlation between HR/T ratio and mortality risk in patients with HF (p < 0.001). With regard to the source of admission, patients in the operating room exhibited a higher mortality risk than those in the emergency department (p = 0.002). Regarding the assessment of disease severity, the Acute Physiology Score III, APACHE IV, and SOFA scores demonstrated an elevated risk of mortality.

A variety of covariate adjustment strategies were employed, and the results of the multivariate regression model showed an association between the HR/T ratio and 28-day ICU mortality (Table 3). When the HR/T ratio was treated as a continuous variable, the unadjusted model demonstrated a positive correlation between an increasing HR/T ratio and 28-day mortality (OR = 1.45, 95% CI: 1.21–1.73). After adjustment for gender, age, ethnicity, and mechanical ventilation use in Model 2, the results were similar to those of the unadjusted model (OR = 1.49, 95% CI: 1.25–1.79, p < 0.001). Building on adjusted Model 2, after accounting for additional confounding variables, including diabetes, CHF, COPD, MAP, AMI, pneumonia, rhythm, β-blocker use, SOFA score, BUN, Albumin, Hct, ALT, there was an enhanced association between the HR/T ratio and 28-day mortality in patients with HF, which remained statistically significant (OR = 1.55, 95% CI: 1.17–2.04, p = 0.002). Upon analysis of the HR/T ratio by quartile in adjusted Model 3, the highest quartile (Q4, 3.29–4.13) exhibited a markedly elevated mortality risk compared with the lowest quartile (Q1, 1.28–1.76; OR = 2.66, 95% CI: 1.34–5.30).

Using GAMs and curve fitting, an association between the HR/T ratio and 28-day ICU mortality in patients with HF remained apparent (Figure 2). After adjusting for covariates such as gender, age, ethnicity, diabetes, mechanical ventilation use, CHF, COPD, MAP, AMI, pneumonia, rhythm, β-blocker use, SOFA score, BUN, albumin, Hct, and ALT, a trend of increasing 28-day ICU mortality with rising HR/T ratios was observed. Additionally, the calculated E-value of 2.42 suggests stability and indicates that unmeasured confounders are unlikely to substantially affect the results.