Association between body temperature and all-cause mortality in patients with sepsis: analysis of the MIMIC-IV database

Abstract

Background

Abnormal body temperature (fever or hypothermia) is a critical symptom in sepsis and is strongly associated with clinical prognosis and disease progression. Given the duality and variability of body temperature fluctuations throughout the disease course, further research is essential to refine clinical strategies for temperature management in sepsis patients.

Methods

We extracted clinical data of sepsis patients from the MIMIC-IV database. A restricted cubic spline (RCS) curve was employed to describe the non-linear relationship between body temperature and clinical outcomes. Based on peak temperature within the first 24 h after admission, patients were categorized into three groups: < 36 °C, 36–38 °C, and > 38 °C. We subsequently matched patients one-to-one into three cohorts using a pairwise propensity score matching (PSM) approach. Alongside clinical data, we conducted log-rank and McNemar tests, and established multiple models, including multiple Cox regression, overlap-weighted (OW) adjusted Cox regression, multiple logistic regression, and OW-adjusted multiple logistic regression, to investigate the impact of temperature on clinical outcomes.

Results

A total of 35,499 sepsis patients were included in my study: 311 with a temperature below 36 °C, 27,538 with a temperature between 36 and 38 °C, and 7650 with a temperature above 38 °C. The RCS analysis revealed a non-linear, U-shaped relationship between body temperature and 28-day, ICU, and in-hospital mortality. Patients with hypothermia had significantly higher 28-day mortality (54.34% vs. 19.28%), ICU mortality (44.37% vs. 12.89%), and in-hospital mortality (49.20% vs. 17.46%) compared to those with hyperthermia. Among patients younger than 65 years, hyperthermia was a protective factor against 28-day mortality relative to normal body temperature, while the opposite was observed in patients aged 65 and older. This trend was consistent in the analysis of ICU and in-hospital mortality.

Conclusions

Among sepsis patients admitted to the ICU, a peak temperature below 36 °C within the first 24 h of admission was associated with higher 28-day mortality. However, no significant difference in clinical prognosis was observed between normothermic and hyperthermic patients.

Background

Sepsis is a clinical syndrome characterized by a high mortality rate, resulting from a dysregulated host response and life-threatening organ dysfunction [1, 2]. Its primary clinical manifestations include chills, palpitations, shortness of breath, and altered mental status [3]. Among these, abnormal body temperature (fever or hypothermia) is not only a key symptom of sepsis but also a critical indicator for assessing the severity and prognosis of the condition [4, 5]. As a vital sign parameter that is easily obtainable and holds significant diagnostic value in clinical practice, it is widely recognized by researchers [6, 7].

Fever typically arises from the body's immune response to infection, which suppresses pathogen growth and enhances immune function, though it may also increase metabolic demands and cardiovascular strain [8]. The need for hypothermia therapy in febrile sepsis remains a clinical controversy. Conversely, hypothermia is often linked to higher mortality rates, especially among older patients or those with weakened immune systems, as it may signify an inadequate immune response or an advanced disease stage [9]. Therefore, temperature management is a crucial aspect of sepsis treatment: antipyretics and physical cooling methods can alleviate hyperthermia symptoms, while warming devices can be employed to address hypothermia, ultimately aiming to reduce complications and improve patient outcomes. Given the dual and heterogeneous nature of temperature abnormalities in sepsis progression, further studies are required to refine clinical strategies for temperature control in septic patients.

Patients with sepsis from the MIMIC-IV database were categorized based on body temperature into three groups: hypothermia group (< 36 °C), normal temperature group (36–38 °C), and hyperthermia group (> 38 °C). Combined with other clinical data, the objective is to explore the relationship between abnormal body temperature and 28-day, ICU, and in-hospital mortality in patients with sepsis.

Methods

Study design

We conducted a retrospective cohort study using a large database, the Medical Information Mart for Intensive Care IV (v3.0). The MIMIC-IV (v3.0) database is a large publicly available dataset containing de-identified health information for patients admitted to the Beth Israel Deaconess Medical Center in Boston for intensive-care unit from 2008 to 2022.

The author obtained a Collaborative Institutional Training Initiative (CITI program) Certificate (Record ID: 61575775), which enabled access to the database and data extraction. The study complied with the Reporting of Studies Conducted using Observational Routinely Collected Health Data (RECORD) statement.

Selection of participants

Based on the Sepsis-3 criteria, this study included patients from the MIMIC-IV database who were diagnosed with sepsis, defined as those with documented or suspected acute infection and Sequential Organ Failure Assessment (SOFA) score ≥ 2. We initially enrolled 41,296 patients, excluding those under 18 years of age or with an ICU stay of less than 24 h. Patients lacking complete data on maximum temperature within the first 24 h of ICU admission were also excluded, resulting in a final study cohort of 35,499 patients. I divided the patients into three groups according to their maximum body temperature within 24 h after admission: hypothermia group (< 36 °C), normal temperature group (36–38 °C), and hyperthermia group (> 38 °C), and formed three cohorts: cohort 1 (< 36 °C and 36–38 °C), cohort 2 (< 36 °C and > 38 °C), cohort 3 (> 38 °C and 36–38 °C).

Variable extraction and outcome

We extracted the baseline characteristics of the included patients within 24 h of ICU admission using Structured Query Language (SQL) with PostgreSQL (version 16), including age, gender, weight, and illness severity indicators (SOFA score and Charlson score). Data on mechanical ventilation, sedatives, vasopressors, albumin, and comorbidities such as heart failure (HF), atrial fibrillation (AFIB), renal disease, liver disease, chronic obstructive pulmonary disease (COPD), coronary artery disease (CAD), stroke, and malignant tumor were also obtained, with comorbidities identified via recorded ICD-9 codes. Vital signs measured for the first time within 24 h of admission included mean arterial pressure (MAP) and heart rate. Laboratory variables, including white blood cell (WBC) count, hemoglobin, platelet count, sodium, potassium, bicarbonate, chloride, blood urea nitrogen (BUN), lactate, creatinine, pH, partial pressure of oxygen (PO₂), and partial pressure of carbon dioxide (PCO₂), were measured within the first 24 h of ICU admission. The primary outcome was 28-day mortality. Secondary outcomes included in-hospital and ICU mortality.

Statistical analysis

Values are presented as the medians (interquartile ranges) or means (standard deviations) for continuous variables. Categorical variables are presented as total numbers and percentages. Group comparisons were performed using the χ² test or Fisher’s exact test for categorical variables, and Student’s t-test or the Mann–Whitney U test for continuous variables, as appropriate.

We employed Restricted Cubic Splines (RCS) to illustrate the non-linear relationship between patient temperature—specifically, the maximum and minimum body temperatures recorded within 24 h of ICU admission—and the study outcomes. To ensure the robustness of the statistical results, we used pairwise propensity score matching (PSM) and overlap-weighted (OW) based on this to adjust for each included covariate to ensure relative balance [10, 11]. The study identified 3 pairwise 1:1 propensity score–matched cohorts: cohort 1 (< 36 °C group and 36–38 °C group), cohort 2 (< 36 °C group and > 38 °C group), cohort 3 (36–38 °C group and > 38 °C group). Kaplan–Meier (K-M) survival curves were plotted for different cohorts to show 28-day, ICU, and in-hospital survival for patients with sepsis and were compared using log-rank tests.

We used multiple regression to analyze the association between body temperature and outcomes. Baseline variables considered clinically relevant entered the multiple Cox regression and multiple logistic regression as covariates. Using the estimated propensity score as the weight, the OW model is established. Standardized mean difference (SMD) was calculated to evaluate the effectiveness of PSM and OW. Then the matching queue and the weighted queue are tested by multiple Cox regression and multiple logistic regression model. In addition, we constructed a double-robust model that combined a regression model for outcomes and a propensity score model for treatments to further assess the robustness of the model results. Finally, we also performed a subgroup analysis to explore the relationship between abnormal body temperature and clinical outcomes in different subgroups of patients, including age, SOFA score, HF, CAD, albumin use and malignancy. To avoid bias induced by missing data, the analysis of the outcome was duplicated after multiple imputations. Additionally, variables with missing values exceeding 40% were excluded from the analytical models as covariates.

All statistical analyses were performed using RStudio and p < 0.05 was considered statistically significant. We employed the survival, car, MatchIt, and forestplot packages to develop models and conduct data analysis.

Results

Baseline characteristics

Using the Sepsis 3.0 criteria, 41,296 patients with sepsis were identified in the database. After excluding patients younger than 18 years, those with an ICU stay of less than 24 h, and individuals with missing maximum body temperature data, the final study cohort consisted of 35,499 patients (Fig. 1). Patients were classified into three groups based on body temperature, and their baseline characteristics are summarized in Table S1. A pairwise propensity score matching (PSM) approach was then employed, matching patients one-to-one into three cohorts, with their characteristics detailed in Tables 1, 2, 3. Additionally, we have included in the supplementary material the patient’s hadm ID, stay ID, subject ID, and the time at which they met the Sepsis-3 diagnostic criteria for reference.

Fig. 1

The detailed process of data extraction

Table 1 Baseline characteristics before and after propensity score matching of cohort 1

Table 2 Baseline characteristics before and after propensity score matching of cohort 2

Table 3 Baseline characteristics before and after propensity score matching of cohort 3

Patients with a body temperature of less than 36 °C had a higher SOFA score (< 36 °C vs 36 °C–38 °C, 8.58 ± 3.88 vs 5.93 ± 3.45) (< 36 °C vs > 38 °C, 8.58 ± 3.88 vs 6.27 ± 3.54) compared with the other two groups, and a higher proportion of patients received mechanical ventilation (< 36 °C vs 36 °C–38 °C, 74.60% vs 47.17%)(< 36 °C vs > 38 °C, 74.60% vs 63.88%), vasopressors (< 36 °C vs 36 °C–38 °C, 64.31% vs 41.01%)(< 36 °C vs > 38° C, 64.31% vs 45.19%), sedatives (< 36 °C vs 36 °C–38 °C, 66.56% vs 46.61%)(< 36 °C vs > 38 °C, 6.56% vs 60.65%), and albumin(< 36 °C vs 36 °C–38 °C, 16.08% vs 15.05%)(< 36 °C vs > 38 °C, 16.08% vs 10.93%) within 24 h of ICU admission. Furthermore, laboratory tests indicated that patients with hypothermia exhibited higher levels of WBC (< 36 °C vs 36 °C–38 °C, 14.83 K/μL vs 13.20 K/μL) (< 36 °C vs > 38 °C, 14.83 K/μL vs 13.58 K/μL), BUN (< 36 °C vs 36 °C–38 °C, 38.40 mg/dL vs 32.24 mg/dL) (< 36 °C vs > 38 °C, 38.40 mg/dL vs 26.14 mg/dL), lactate (< 36 °C vs 36 °C–38 °C, 4.08 mmol/L vs 2.26 mmol/L) (< 36 °C vs > 38 °C, 4.08 mmol/L vs 2.27 mmol/L), and creatinine(< 36 °C vs 36 °C–38 °C, 2.20 mg/dL vs 1.70 mg/dL) (< 36 °C vs > 38 °C, 2.20 mg/dL vs 1.49 mg/dL) in their initial measurements within 24 h of ICU admission. Kaplan–Meier curves indicated a significantly worse prognosis for hypothermic patients (p < 0.001) (Fig. 2) (Table S2-4).

Fig. 2

Kaplan–Meier survival curves showing differences in clinical outcomes among patients with sepsis. Kaplan–Meier survival curves for the 28-day (A), ICU (B), and in-hospital (C) mortality among patients with sepsis

Patients with a temperature exceeding 38 °C were more likely to require mechanical ventilation (63.88% vs 47.17%), vasopressors (45.19% vs 41.01%), and sedatives (63.88% vs 47.17%), and had a higher likelihood of experiencing a stroke (13.84% vs 10.36%) compared to those with normal body temperature. However, the incidence of HF (25.48% vs 33.97%), AFIB (13.84% vs 16.59%) and CAD (23.56% vs 32.61%) was lower in this group.

Additionally, the RCS analysis revealed a non-linear, U-shaped relationship between the maximum body temperatures recorded within 24 h of ICU admission and clinical outcomes, including 28-day, ICU, and in-hospital mortality. Patients with sepsis exhibit the best prognosis when their body temperature reaches 37.33 ℃. The odds ratio (OR) increases when the temperature deviates either below or above this value (Fig. 3). Furthermore, the data analysis revealed a comparable non-linear relationship between minimum body temperature and clinical outcomes in sepsis patients within the first 24 h of ICU admission (Figure S2).

Fig. 3

RCS curve for the maximum body temperature and clinical outcome. Solid red lines are odds ratios, with light red regions showing 95% confidence. RCS curve for the body temperature and the 28-day (A), ICU (B), in-hospital (C) mortality

Primary outcome

Following multiple imputation of missing data, we conducted a multivariable logistic regression analysis, revealing that patients with hypothermia exhibited a lower 28-day mortality rate compared to those with normal (OR 0.25, 95%CI 0.19–0.32, p < 0.001) or elevated body temperatures (OR 0.26, 95%CI 0.19–0.34, p < 0.001). However, no significant difference was observed between individuals with normal body temperature and those with hyperthermia (OR 0.98, 95%CI 0.91–1.05, p < 0.714). Utilizing the estimated propensity score, overlap weighting (OW) was applied to standardize differences between cohorts. Post-normalization, the imbalance of covariates among the three groups was notably reduced (Fig. 4, Table S5–7), yet the findings remained consistent (Table 4).

Fig. 4

Change in standardized mean difference (SMD) before and after matching. Change in SMD before and after matching of cohort 1 (A), cohort 2 (B), cohort 3 (C)

Table 4 Primary and secondary outcome analyses with different models for 3 cohorts

Secondary outcome and sensitivity analyses

Hypothermia was also linked to reduced in-hospital and ICU mortality. Sensitivity analyses were conducted across all assessment models, including multiple Cox regression, OW-adjusted Cox regression, multiple logistic regression, OW-adjusted multiple logistic regression and doubly robust estimation. McNemar test was performed on the matched data. All analyses reached the same conclusion: hypothermic patients had lower 28-day mortality compared to those with normothermia or hyperthermia. However, across all models, differences in clinical outcomes between normothermic and hyperthermic patients were observed only in the log-rank analysis, while no significant differences were found in other models (Table 4).

Subgroup analyses

Subgroup analysis showed that hypothermia patients had the highest risk of 28-day mortality, ICU mortality, and in-hospital mortality in almost all subgroups. In patients younger than 65 years of age, hyperthermia emerged as a protective factor against 28-day mortality relative to normal body temperature (OR 0.79, 95%CI 0.71–0.88, p < 0.001), whereas the opposite was true in older patients older than or equal to 65 years of age (OR 1.11, 95%CI 1.02–1.2, p = 0.014). This phenomenon is also generally consistent in the analysis of ICU mortality and in-hospital mortality. In coronary artery disease patients, hyperthermia was associated with higher ICU (OR 1.42, 95%CI 1.23–1.64, p < 0.001) and in-hospital mortality (OR 1.29, 95%CI 1.13–1.46, p < 0.001), but had no significant effect on 28-day mortality (OR 1.06, 95%CI 0.93–1.2, p = 0.384). Regardless of albumin use, lower body temperature was a risk factor for clinical outcomes in sepsis patients when compared to those with normal or elevated body temperatures. In patients with non-malignant tumors, hyperthermia emerged as a protective factor against 28-day mortality (OR 0.89, 95%CI 0.83–0.96, p = 0.002) compared with patients with normal body temperature, but this phenomenon was not evident in patients with malignant tumors (OR 0.89, 95%CI 0.78–1.02, p = 0.098) (Fig. 5) (Figure S1).

Fig. 5

Subgroup analysis of the association between the body temperature and primary outcome. Subgroup analysis of the association between body temperature and 28-day mortality in cohort 1 (A), cohort 2 (B), cohort 3 (C)

Discussion

This study suggests a non-linear relationship between the maximum body temperature within the first 24 h of admission and clinical outcomes in patients with sepsis. Patients with hypothermia had a worse prognosis compared to those with normal or elevated temperatures. In non-elderly patients (< 65 years old) with sepsis, hyperthermia appeared to be a protective factor, while in elderly patients (≥ 65 years old), the opposite trend was observed, though without significant differences. These results may be linked to altered immune responses in sepsis patients.

Hypothermia is strongly associated with poor outcomes in sepsis, consistent with previous findings. Numerous clinical studies have shown that sepsis patients with hypothermia experience higher mortality rates and longer hospital stays than those with normothermia, likely due to weakened immune responses, organ dysfunction, and diminished infection control [12, 13]. A multicenter prospective study by Shigeki Kushimoto et al. demonstrated that differences in 28-day mortality were evident between hypothermic and non-hypothermic patients [14]. Additionally, a retrospective analysis of 525 sepsis patients in Dutch ICUs identified hypothermia within the first 24 h as a risk factor for outcomes including 30-day, 60-day and 1-year mortality, as well as acute kidney injury (AKI) incidence. The poor prognosis associated with hypothermia may be linked to persistent lymphocytopenia [15], elevated markers of endothelial activation [16], and hypertension. In our study, we also observed that the hypothermia group had a higher proportion of white blood cells, as well as an increased risk of stroke and coronary artery disease.

Furthermore, prior basic studies have suggested that the onset of hypothermia-type sepsis may be associated with vasodilatory hypothermia and a lethally hypodynamic state induced by nitric oxide synthase (iNOS)-derived nitric oxide (NO) in late sepsis [17]. Additionally, research has examined the inhibitory effects on the immune system, particularly the decline in neutrophil function and phagocytosis, which may restrict the body's capacity to effectively eliminate sources of infection. During infection, polymorphonuclear cells (PMNs), such as neutrophils, form neutrophil extracellular traps (NETs) to contain and eliminate pathogens. Studies have demonstrated that NET formation is impaired at 35 °C and 42 °C, while optimal at 40 °C, highlighting how temperature affects immune response [18]. This evidence underscores the critical importance of managing hypothermia in sepsis treatment. Additionally, hypothermia is uncommon at the time of death, even when multi-organ failure is at its most severe [19, 20]. Therefore, the association between hypothermia and higher mortality rates may not necessarily indicate that hypothermia exacerbates the dysfunction caused by sepsis. This hypothesis requires further investigation and discussion for confirmation.

Contrary to the established view of hypothermia as a poor prognostic factor, the role of hyperthermia in sepsis remains debated. On one hand, moderate increases in body temperature can inhibit bacterial growth, enhance immune function by promoting antibody and cytokine synthesis, and help eliminate pathogens, providing protective effects. A clinical study that combined data from two large, independent multinational ICU cohorts (636051 patients) found that elevated body temperature was associated with reduced in-hospital mortality, with the lowest risk of death observed in patients whose peak temperature ranged from 39.0 °C to 39.4 °C [21]. On the other hand, persistent high fever may exacerbate the inflammatory response, precipitating a cytokine storm, disrupting the body's immune regulation, and elevating metabolic load, ultimately resulting in organ damage and multiple organ dysfunction. Laupland et al. found that patients with high temperatures (> 39.5 °C) were more likely to experience arrhythmias and tachycardia, resulting in worse clinical outcomes [4]. Similarly, a large epidemiological study linked high fever (≥ 39.5 °C) with significantly increased mortality in sepsis patients [22]. Moderate temperature regulation, such as mild hypothermia, can enhance blood coagulation in patients with sepsis [23] and diminish the extent of systemic organ failure [24].

The paradoxical outcome may be driven by various factors, including the patient's age, the presence of specific complications (such as malignancy, stroke) [25], outside temperature [26], and the patient's gut microbiota composition [27]. A multicenter retrospective study indicated that hyperthermia (BT ≥ 38.3 °C) was a protective factor for 28-day mortality in sepsis patients under 75 years old, while in older patients, fever had no significant impact on clinical outcomes, possibly due to reduced thermogenic cytokine production and frequent comorbidities that impair thermoregulation [28]. Comparable results were observed in my study. Moreover, fever is a common symptom in patients with hematological malignancies and some solid tumors, such as primary liver cancer, renal cell carcinoma, and sarcoma, potentially related to cancer-induced alterations in cytokine activity [29]. Sepsis patients with malignancies typically have a poorer prognosis. Subgroup analyses in this study revealed that hyperthermia was a protective factor for 28-day mortality in patients without malignancy, but not significantly so in those with malignancy, consistent with findings from previous research.

Given the heterogeneity of body temperature fluctuations during sepsis, managing temperature effectively poses a significant challenge for clinicians. For patients with elevated temperatures, it is generally accepted that appropriate cooling measures are essential to protect vital organs, such as the brain and heart, which are particularly sensitive to temperature increases. Studies have demonstrated that external cooling can reduce vasopressor requirements in septic shock patients and lower 14-day mortality [30]. However, the necessity of antipyretic treatment in febrile sepsis patients remains contentious. A multicenter randomized clinical trial involving 700 patients with infections and temperatures exceeding 38.3 °C found that early intravenous acetaminophen had no significant impact on 90-day mortality or ICU-free days [31]. Furthermore, while rewarming therapy is widely considered beneficial, there remains a lack of robust clinical and experimental evidence to elucidate its role and mechanisms in sepsis treatment [32]. Induced hypothermia therapy has also drawn considerable interest in recent research. Despite promising outcomes in animal models [33,34,35], randomized clinical trials have shown that lowering core body temperature to 32–34 °C did not improve mortality [36]. As such, further research is necessary to inform clinical decisions regarding temperature management in sepsis patients, and this study offers a certain degree of clinical foundation for such considerations.

However, this study has certain limitations. First, the RCS analysis showed a non-linear relationship between peak body temperature within the first 24 h of admission and clinical outcome, with optimal prognosis observed at 37.33 °C. However, a single temperature measurement may not fully capture the patient’s condition. For example, trauma patients exhibit complex circadian rhythms in temperature, and early increases in rhythm frequency and amplitude are associated with sepsis development [37]. Additionally, Bhavani and colleagues employed longitudinal temperature trajectories to categorize hospitalized patients with infections into four distinct subtypes: “hyperthermic, slow resolvers”, “hyperthermic, fast resolvers”, “normothermic”, and “hypothermic”. Their analysis revealed significant differences in inflammatory markers (such as C-reactive protein and erythrocyte sedimentation rate) and clinical outcomes across these subtypes [9]. Thus, real-time temperature monitoring may be necessary to better understand the immunophenotype of infected patients [9, 38]. Second, previous studies have indicated that sepsis complications can affect temperature fluctuations. Subgroup analyses of coronary heart disease, atrial fibrillation, and malignancy were conducted. The findings indicated no significant differences in mortality. Further investigation is required to examine the effects of body temperature variations on clinical outcomes in more specific subgroups, such as those with stroke, solid tumors, and hematologic malignancies.

Conclusions

In summary, the results demonstrated that hypothermia was linked to a higher risk-adjusted 28-day, ICU, and in-hospital mortality in sepsis patients. Among non-elderly patients, elevated body temperature served as a protective factor against sepsis compared to normothermia, whereas in elderly patients, the opposite trend was observed, though without statistical significance. This study elucidated the relationship between body temperature fluctuations and clinical outcomes, offering valuable insights and recommendations for temperature management in sepsis.