Association Between Temperature During Intensive Care Unit and Mortality in Patients With Acute Respiratory Distress Syndrome

Abstract

The relationship between body temperature changes and prognosis in patients with acute respiratory distress syndrome (ARDS) remains inconclusive. Our study aimed to investigate the clinical value of body temperature in the management of ARDS. Data from the Medical Information Mart for Intensive Care III database were collected. Adult patients with ARDS were enrolled and further grouped based on their temperature values in the intensive care unit. Both the maximum (temperature_max) and minimum (temperature_min) temperatures were used. The primary outcome was 28-day mortality rate. Polynomial regression, subgroup analysis, and logistic regression analysis were performed in the final analysis. A total of 3922 patients with ARDS were enrolled. There was a U-shaped relationship between 28-day mortality and body temperature. For patients with infection, the elevated temperature_max (≥37.0°C) was associated with decreased mortality, with an odds ratio ranging from 0.39 to 0.49, using temperature_max from 36.5°C to 36.9°C as reference. For patients without infection, a similar tendency was observed, but the protective effect was lost at extremely high temperatures (≥38.0°C, p < 0.05). Elevated temperature_min (≥37.0°C) and decreased temperature_min (<35.0°C) were associated with increased mortality, using the temperature_min from 36.0°C to 36.9°C as a reference. Hypothermia was associated with increased mortality in patients with ARDS, while the effect of hyperthermia (≥37.0°C) on the mortality of patients with ARDS was not fully consistent in the infection and noninfection subgroups. Short-term and transient temperatures above 37.0°C would be beneficial to patients with ARDS, but extreme hyperthermia and persistent temperatures above 37.0°C should be avoided.

Introduction

Acute respiratory distress syndrome (ARDS) is common in the emergency department and intensive care unit (ICU) department and is characterized by noncardiogenic acute hypoxic respiratory insufficiency or failure (ARDS Definition Task Force et al., 2012). In the past several years, ARDS has gained more scientific research attention owing to the outbreak of coronavirus disease 2019 (COVID-19), which could lead to lung injury and ARDS development. ARDS causes a great medical burden owing to its high morbidity and mortality. A multicenter study reported that the morbidity rate of ARDS is ∼10.4%, while the hospital mortality rate is ∼40% (Bellani et al., 2016). Effective treatment and well-done management are essential to achieve positive outcomes in patients with ARDS.

Body temperature is an easy-to-obtain clinical symptom and serves as a sensitive indicator of many diseases, especially infectious diseases. An abnormal body temperature is disadvantageous for patient prognosis. A multicenter randomized clinical trial indicated that the elevated peak temperature in the first 24 hours after ICU admission was related to decreased in-hospital mortality in critically ill patients with infection, but it was a risk factor for those without infection (Young et al., 2012). Hypothermia was reported as an independent risk factor for mortality in the severe sepsis and COVID-19 cohorts (Kushimoto et al., 2013; Liu et al., 2021). In patients with ARDS, several factors may lead to the onset of temperature changes, including infections, noninfectious systemic inflammatory reactions, agitation, and certain medications. However, whether the management of body temperature contributes to the recovery of ARDS patients remains inconclusive. The optimal target temperature for temperature management remains unclear.

A secondary analysis study reported that higher temperature was associated with a better outcome for patients with ARDS, and each 1°C elevation in body temperature may reduce the probability of death by 15% (Schell-Chaple et al., 2015). In the COVID-19 cohort, elevated body temperature was significantly associated with the development of ARDS and severe COVID-19 (Liu et al., 2021; Wang et al., 2020). These limited data indicate a potential relationship between temperature and ARDS, and further research is needed to detect this association.

Based on the above findings, our study aimed to investigate the relationship between body temperature and prognosis of patients with ARDS.

Materials and Methods

Data source

This was a retrospective study using data collected from the Medical Information Mart for Intensive Care III (MIMIC-III database, Vision 1.4) (Johnson et al., 2016). The MIMIC-III database includes over 60,000 hospitalization records of 46,000 critically ill patients between 2001 and 2012 at the Beth Israel Deaconess Medical Center (Boston, MA, USA). Informed consent was waived by the Institutional Review Boards (IRBs) of the Beth Israel Deaconess Medical Center and Massachusetts Institute of Technology because the personal information of patients was hidden in the MIMIC-III database. Author Yi-Peng Fang obtained certification (certification No. 43025968) and gained free access to the MIMIC database after completing the National Institutes of Health web-based course study and passing the examination. PgAdmin4 and PostgreSQL (version 9.6) software were Strengthening the Reporting of Observational Studies in Epidemiology used to extract raw data from MIMIC-III. Our study followed these guidelines (von Elm et al., 2007).

Selection of participants

All adult patients (age ≥18 years) with ARDS were identified based on the Berlin definition, as previously reported (ARDS Definition Task Force et al., 2012; Huang et al., 2021). For multiple-admission patients, only the first hospitalization records were used in the present study. Temperature records (item-id = “676,” “678,” “223761” or “223762”) from chart events table were screened out. The conversion between Celsius and Fahrenheit was based on the following formula: Celsius = (Fahrenheit-32)/1.8. Because the different sites for measuring body temperature would influence the value of body temperature, we only included oral temperature records. The maximum (temperature_max) and minimum (temperature_min) values of body temperature during the ICU stay were calculated. Hyperthermia and hypothermia were defined as temperatures ≥37.3°C and <36.0°C as standard.

Definition of variables and outcomes

The study variables included demographics, type of admission, comorbidities, disease severity scoring system, vital signs, laboratory parameters, and intervention methods. According to a previous study and the International Classification of Diseases-9 (ICD-9 code), we selected patients with infection (Angus et al., 2001), brain injury (Cai et al., 2022), and cardiac arrest (ICD-9 code = 4275). The details of the ICD-9 codes used to screen for complications are shown in Supplementary Table S1. The disease severity scores on ICU admission were also assessed. The maximum, mean, and minimum values of vital signs and laboratory parameters within the first 24 hours after ICU admission were used. All records of dexamethasone, prednisolone, prednisone, and hydrocortisone from the prescription table were considered as evidence of glucocorticoid exposure. The medical records of atracurium from the prescription table were used as evidence of muscle relaxant exposure.

The primary outcome measure was 28-day mortality rate. The secondary outcome indicators included 7-day, 14-day, 90-day, 1-year, in-hospital and ICU mortality, length of hospital stay (hospital-LOS), and ICU-LOS.

The winsor2 command in the Stata software, with a threshold range from 1 to 99, was used to treat abnormal values, except for age. The abnormal age value (>300 years) was replaced with 91. Variables with missing values exceeding 30% were excluded from the final analysis. Missing data were imputed with the mean or medium according to their distribution when less than 15% was missing. Multiple imputation methods were used to process the missing data when the missing proportion ranged from 10% to 20%. Details of the missing values are listed in Supplementary Table S2.

Statistical analysis

Normally distributed continuous variables are described as means and standard deviations, and non-normally distributed continuous variables are represented as medians with interquartile ranges. The Student's t test and Wilcoxon rank-sum test were used for two-group comparisons of normally and non-normally distributed continuous variables. For multiple group comparisons, one-way analysis of variance and Kruskal–Wallis tests were used for normally and non-normally distributed continuous variables. Categorical variables are described as numbers and percentages (%), and comparisons were performed using chi-square or Fisher's exact tests.

To explore the potential confounders systematically, directed acyclic graph (DAG) analysis was performed (Lederer et al., 2019), and the causal diagram was constructed using DAGitty software (Textor et al., 2011). In the present study, our minimal sufficient adjustment set included the following factors: age, sex, anemia, brain injury, cardiac arrest, infection, hypothyroidism, metastatic cancer, rheumatoid arthritis, glucocorticoid exposure, muscle relaxant exposure, continuous renal replacement therapy (CRRT), and ventilation (Supplementary Fig. S1). Subgroup analysis was performed for the infection and noninfection subgroups. A logistic regression model adjusted for all potential confounders was used to determine the relationship between different levels of temperature and 28-day mortality. The effects were represented by odds ratios (ORs) and 95% confidence intervals (CIs). The variance inflation factor was applied to describe the multicollinearity and how well our model fitted the observation.

Data cleaning, statistical analyses and illustrations were performed using Stata (version 15.0). In all statistical analyses, p < 0.05 was considered statistically significant.

Results

Sample size and baseline information

There were 3922 patients with ARDS enrolled in our final study, including 3097 (78.96%) survivors and 825 (21.04%) nonsurvivors, according to their 28-day situation. Figure 1 showed the flow chart of patient selection. As shown in Table 1, patients in the survival group were younger (p < 0.001). Survivors had a lower percentage of emergency admissions, infections, cardiac arrests, brain injuries, and metastatic cancers than nonsurvivors (all p < 0.001). The severity scores and maximum value of serum creatinine were also lower, but serum platelet count was higher in the survival group than in the nonsurvival group (all p < 0.001). Survivors also used less CRRT, muscle relaxants, and glucocorticoids than nonsurvivors (all p < 0.05). However, the percentage of ventilation usage was higher among survivors (p < 0.001). Patients in survival group had the higher temperature₀ (37.06 ± 0.83 vs. 36.91 ± 1.03, p < 0.001), temperature_min (35.63 ± 0.83 vs. 35.33 ± 1.04, p < 0.001), and temperature_mean (37.10 ± 0.48 vs. 36.94 ± 0.70, p < 0.001) values than those in nonsurvival group. No significant difference was found in temperature_max value (38.34 ± 0.80 vs. 38.35 ± 1.02, p = 0.765).

FIG. 1.

The flow chart of patient selection. ARDS, acute respiratory distress syndrome; ICU, intensive care unit; MIMIC-III, Medical Information Mart for Intensive Care III.

Table 1.

The Baseline Information and Body Temperature in Acute Respiratory Distress Syndrome Patients

	Overall (n = 3922)	Survivors (n = 3097)	Nonsurvivors (n = 825)	p
Age, years	63.80 ± 16.77	62.65 ± 16.57	68.12 ± 16.82	<0.001
Male, %	2314 (59.00)	1846 (59.61)	468 (56.73)	0.135
Emergency admission, %	3085 (78.66)	2312 (74.65)	773 (93.70)	<0.001
Comorbidities, %
Infection	2227 (56.78)	1661 (53.63)	566 (68.61)	<0.001
Cardiac arrest	200 (5.10)	92 (2.97)	108 (13.09)	<0.001
Brain injury	467 (11.91)	320 (10.33)	147 (17.82)	<0.001
Metastatic cancer	191 (4.87)	103 (3.33)	88 (10.67)	<0.001
Rheumatoid arthritis	73 (1.86)	58 (1.87)	15 (1.82)	0.918
Hypothyroidism	303 (7.73)	242 (7.81)	61 (7.39)	0.688
Anemia	678 (17.29)	544 (17.57)	134 (16.24)	0.372
Severity score
SOFA on admission	5 (3, 7)	5 (3, 7)	7 (4, 10)	<0.001
APSIII on admission	43 (32, 64)	40 (31, 54)	58 (44, 78)	<0.001
Vital signs
Heart rate, bpm	89.40 ± 15.70	88.72 ± 15.18	91.94 ± 17.29	<0.001
Respiratory rate, cpm	19.35 ± 4.46	18.91 ± 4.22	21.02 ± 4.93	<0.001
Mean blood pressure, mmHg	77.70 ± 10.13	78.09 ± 9.86	76.24 ± 10.97	<0.001
Minimum SpO2, %	93 (89, 95)	93 (90, 95)	91 (86, 94)	<0.001
Laboratory examination
Maximum white blood cell, 109/L	14.3 (10.6, 19.1)	14.3 (10.8, 18.8)	14.5 (9.8, 20.2)	0.502
Minimum hemoglobin, 1012/L	9.60 ± 2.12	9.57 ± 2.11	9.70 ± 2.15	0.112
Minimum platelet, 109/L	167 (112, 231)	168 (117, 229)	158 (90, 238)	<0.001
Maximum creatinine, mg/dL	1.0 (0.8, 1.5)	1.0 (0.8, 1.3)	1.3 (0.9, 2.1)	<0.001
Culture positive, %	1195 (34.45)	897 (33.55)	298 (37.48)	0.040
Ventilation on the first day, %	3110 (79.30)	2502 (80.79)	608 (73.0)	<0.001
Ventilation during ICU stay, %	3452 (88.02)	2744 (88.60)	708 (85.82)	0.029
CRRT during ICU stay, %	193 (4.92)	81 (2.62)	112 (13.58)	<0.001
Muscle relaxants exposure, %	288 (7.34)	182 (5.88)	106 (12.85)	<0.001
Glucocorticoid, %	683 (17.41)	472 (15.24)	211 (25.58)	<0.001
Body temperature (°C)
Temperature0	37.03 ± 0.88	37.06 ± 0.83	36.91 ± 1.03	<0.001
Temperaturemax	38.34 ± 0.85	38.34 ± 0.80	38.35 ± 1.02	0.765
Temperaturemin	35.56 ± 0.89	35.63 ± 0.83	35.33 ± 1.04	<0.001
Temperaturemean	37.07 ± 0.54	37.10 ± 0.48	36.94 ± 0.70	<0.001

Continuous variables are displayed as mean (standard deviation) or median (first quartile–third quartile); categorical variables are displayed as count (percentage).

The superscripts of lowercase letters present the result of statistical analysis. When two groups have the same superscript of lowercase letters, there is no significant difference between them. The different superscripts of lowercase letters present a significant different, with p < 0.05.

APS, Acute Physiology Score; bpm, beats per minute; cpm, cycle per minute; CRRT, continuous renal replacement therapy; ICU, intensive care unit; mmHg, millimeter of mercury; SOFA, Sequential Organ Failure Assessment; temperature₀, initial temperature after intensive care unit admission; temperature_max, maximum temperature during ICU stay; temperature_min, minimum temperature; temperature_mean, mean temperature.

Maximum value of temperature and clinical outcomes

Table 2 showed the unadjusted clinical outcomes according to the temperature_max categories in patients with ARDS. Hyperthermia_max (>37.0°C) was associated with significantly lower 28-day, 90-day, and hospital mortality rates (all p < 0.05) than patients in the normal range of temperature_max (36.0–37.2°C). Patients with hyperthermia_max had significantly longer hospital LOS and ICU-LOS than those in the normal range of temperature_max (all p < 0.05). No significant differences were observed in ICU mortality, acute kidney injury (AKI), and Stage 3 AKI development among the three categories (all p > 0.05). The relationship between the temperature_max value and 28-day mortality was further described by polynomial regression and bar graph (Fig. 2A). Their relationship exhibited a U-shape in overall ARDS cohort. With the increase in temperature_max value, the trend of 28-day mortality first decreased and then increased, and inflection points were observed between 37.5°C and 38.4°C.

Table 2.

Unadjusted Outcomes By Temperature_max Value in the Entire Cohort of Patients with Acute Respiratory Distress Syndrome

Outcomes	Temperaturemax (°C) during ICU department
	<36.0 (n = 5), n (%)	36.0–37.2 (n = 405), n (%)	≥37.3 (n = 3512), n (%)
28-Day mortality	4 (80.0)a	121 (29.9)b	700 (19.9)c
90-Day mortality	4 (80.0)a	148 (36.5)a	917 (26.1)b
Hospital mortality	4 (80.0)a	113 (27.9)b	708 (20.2)c
ICU mortality	4 (80.0)a	67 (16.5)a	585 (16.7)a
Hospital-LOS, days	6.4 (2.0, 8.8)a,b	8.1 (4.8, 15.3)a	12.9 (7.6, 21.8)b
ICU-LOS, days	1.3 (1.1, 1.6)a	2.2 (1.3, 3.8)a	6.1 (2.9, 12.8)b
AKI	3 (60.0)a	204 (50.4)a	1819 (51.8)a
Vasopressin	3 (60.0)a,b	182 (44.9)b	2170 (61.8)a
AKI stage 3	1 (20.0)a	62 (15.3)a	692 (19.7)a

Continuous variables are displayed as mean (standard deviation) or median (first quartile–third quartile); categorical variables are displayed as count (percentage).

The superscripts of lowercase letters present the result of statistical analysis. When two groups have the same superscript of lowercase letters, there is no significant difference between them. The different superscripts of lowercase letters present a significant different, with p < 0.05.

AKI, acute kidney injury; ICU, intensive care unit; LOS, length of stay; temperature_max, maximum temperature during ICU stay.

FIG. 2.

The relationship between temperature_max value and 28-day mortality in patients with ARDS by Polynomial regression and bar graphs. (A) Overall cohort, (B) Infection subgroup, (C) Noninfection subgroup. CI, confidence interval; ICU, intensive care unit.

Previous studies have found that hyperthermia played not exactly the same effect in infectious and noninfectious patients. Therefore, we further investigated the relationship between temperature_max and clinical outcomes in infectious and noninfectious cohorts. As shown in Table 3, hyperthermia_max was associated with decreased 28-day, 90-day, and hospital mortality rates in the infectious cohort (all p < 0.05), but no significant difference was found in the noninfectious cohort (all p > 0.05). Patients with hyperthermia_max had significantly longer hospital-LOS and ICU-LOS than those in the normal range of maximum temperature_max value in both infectious and noninfectious cohorts (all p > 0.05). Notably, U-shaped relationships were observed in both cohorts presented by polynomial regression (Fig. 2B, C). As shown in the bar graphs, the lowest 28-day mortality rate occurred in the 39.0–39.4°C subgroup in the infectious cohort. The inflection points appeared in the range from 37.5°C to 37.9°C in the noninfectious cohort. The 28-day mortality rate was markedly increased when the temperature_max value reached 38.5°C in the noninfectious cohort. However, no such tendency was observed in the infectious disease cohort.

Table 3.

Unadjusted Outcomes By Temperature_max Value in Acute Respiratory Distress Syndrome Patients With or Without Infection

Outcomes	Patients with infection, n (%)			Patients without infection, n (%)
	<36.0 (n = 1)	36.0–37.2 (n = 217)	≥37.3 (n = 2009)	<36.0 (n = 4)	36.0–37.2 (n = 188)	≥37.3 (n = 1503)
28-Day mortality	1 (100.0)a,b	83 (38.2)b	482 (24.0)a	3 (75.0)a	38 (20.2)b	218 (14.5)b
90-Day mortality	1 (100.0)a,b	102 (47.0)b	656 (32.7)a	3 (75.0)a	46 (24.5)a,b	261 (17.4)b
Hospital mortality	1 (100.0)a,b	77 (35.5)b	508 (25.3)a	3 (75.0)a	36 (19.1)b	200 (13.3)b
ICU mortality	0 (0.0)a	45 (20.7)a	408 (20.3)a	2 (50.0)a	22 (11.7)a	177 (11.8)a
Hospital-LOS	8.8 (NA)a,b	11.9 (5.9, 19.8)a	17.7 (10.7, 27.0)b	4.2 (1.5, 12.2) a,b	6.4 (4.1, 10.9)a	8.8 (6.0, 13.8)b
ICU-LOS	1.6 (NA)a,b	2.7 (1.7, 4.2)a	9.3 (4.5, 17.3)b	1.2 (1.1, 1.7)a	1.8 (1.2, 3.1)a	3.6 (2.1, 7.0)b
AKI	1 (100.0)a	114 (52.5)a	1110 (55.3)a	2 (50.0)a	90 (47.9)a	709 (47.2)a
Vasopressin	1 (100.0)a,b	94 (43.3)b	1227 (61.1)a	2 (50.0)a,b	88 (46.8)b	943 (62.7)a
AKI stage 3	0 (0.0)a	41 (18.9)a	492 (24.5)a	1 (25.0)a	21 (11.2)a	200 (13.3)a

Continuous variables are displayed as mean (standard deviation) or median (first quartile–third quartile); categorical variables are displayed as count (percentage).

AKI, acute kidney injury; ICU, intensive care unit; LOS, length of stay; NA, not applicable; temperature_max, maximum temperature during ICU stay.

The superscripts of lowercase letters present the result of statistical analysis. When two groups have the same superscript of lowercase letters, there is no significant difference between them. The different superscripts of lowercase letters present a significant different, with p < 0.05.

To further explore the effect of temperature_max value on 28-day mortality in patients with ARDS, logistical regression analysis was performed, with temperature_max values from 36.5°C to 36.9°C serving as a reference. As shown in Table 4, in the unadjusted and adjusted models, increasing maximum temperature_max values were associated with decreased 28-day mortality in patients with ARDS (all OR <1.0, p < 0.05). When temperature_max value reached 37.0–37.9°C, the increase of temperature_max value reduced the risk of 28-day death by 59% (OR: 0.41, p < 0.001) compared with temperature_max value ranging from 36.5°C to 36.9°C. The decrease of temperature_max value did not influence the 28-day risk in patients with ARDS (p > 0.05). A similar analysis was performed for the infected and noninfected cohorts. Table 5 showed that the increase in temperature_max value associated with over 50% reduction (OR: 0.39–0.49, all p < 0.05) in the risk of 28-day death in the infectious cohort compared with the reference value. In the noninfectious cohort, the increase in temperature_max was also considered as an independent protective factor within a reasonable range from 37.0°C to 38.4°C (OR: 0.29–0.43, all p < 0.05).

Table 4.

Crude and Adjusted Odds Ratios Using Temperature_max As the Design Variable in Patients With Acute Respiratory Distress Syndrome

Unadjusted model			Adjusted model
Variable	OR (95% CI)	p	Variable	OR (95% CI)	p
Temperaturemax (<36.0)	7.47 (0.81–68.82)	0.076	Temperaturemax (<36.0)	8.16 (0.81–82.59)	0.075
Temperaturemax (36.0–36.4)	3.27 (1.27–8.37)	0.014	Temperaturemax (36.0–36.4)	2.05 (0.75–5.64)	0.163
Temperaturemax (36.5–36.9)	Ref		Temperaturemax (36.5–36.9)	Ref
Temperaturemax (37.0–37.4)	0.50 (0.33–0.77)	0.002	Temperaturemax (37.0–37.4)	0.46 (0.29–0.72)	0.001
Temperaturemax (37.5–37.9)	0.42 (0.28–0.63)	<0.001	Temperaturemax (37.5–37.9)	0.41 (0.26–0.63)	<0.001
Temperaturemax (38.0–38.4)	0.40 (0.27–0.60)	<0.001	Temperaturemax (38.0–38.4)	0.41 (0.27–0.64)	<0.001
Temperaturemax (38.5–38.9)	0.46 (0.31–0.69)	<0.001	Temperaturemax (38.5–38.9)	0.46 (0.30–0.72)	0.031
Temperaturemax (>39.0)	0.60 (0.40–0.89)	0.010	Temperaturemax (>39.0)	0.62 (0.40–0.96)	0.031
			Age	1.03 (1.02–1.03)	<0.001
			CRRT	6.08 (4.37–8.46)	<0.001
			Ventilation	0.70 (0.54–0.90)	0.006
			Glucocorticoid	1.68 (1.37–2.07)	<0.001
			Muscle relaxants	1.68 (1.24–2.28)	0.001
			Cardiac arrest	5.08 (3.70–6.98)	<0.001
			Brain injury	2.19 (1.74–2.77)	<0.001
			Metastatic cancer	3.76 (2.73–5.16)	<0.001

Continuous variables are displayed as mean (standard deviation) or median (first quartile–third quartile); categorical variables are displayed as count (percentage).

CI, confidence interval; CRRT, continuous renal replacement therapy; OR, odds ratio; temperature₀, initial temperature after intensive care unit admission; temperature_max, maximum temperature during ICU stay; temperature_min, minimum temperature; temperature_mean, mean temperature.

Table 5.

Adjusted Odds Ratios Using Temperature_max As the Design Variable in Acute Respiratory Distress Syndrome Patients With and Without Infection

Infectious cohort			Noninfectious cohort
Variable	OR (95% CI)	p	Variable	OR (95% CI)	p
Temperaturemax (36.0–36.4)	0.99 (0.27–3.67)	0.990	Temperaturemax (<36.0)	10.77 (0.86–134.68)	0.065
Temperaturemax (36.5–36.9)	Ref		Temperaturemax (36.0–36.4)	8.39 (1.32–53.12)	0.024
Temperaturemax (37.0–37.4)	0.47 (0.26–0.84)	0.011	Temperaturemax (36.5–36.9)	Ref
Temperaturemax (37.5–37.9)	0.49 (0.28–0.86)	0.012	Temperaturemax (37.0–37.4)	0.43 (0.20–0.96)	0.039
Temperaturemax (38.0–38.4)	0.44 (0.25–0.76)	0.003	Temperaturemax (37.5–37.9)	0.29 (0.13–0.62)	0.002
Temperaturemax (38.5–38.9)	0.39 (0.22–0.68)	0.001	Temperaturemax (38.0–38.4)	0.36 (0.17–0.77)	0.008
Temperaturemax (>39.0)	0.43 (0.25–0.74)	0.002	Temperaturemax (38.5–38.9)	0.51 (0.23–1.14)	0.101
Age	1.02 (1.02–1.03)	<0.001	Temperaturemax (>39.0)	1.12 (0.51–2.47)	0.778
CRRT	4.55 (3.18–6.51)	<0.001	Female	1.52 (1.11–2.10)	<0.001
Glucocorticoid	1.44 (1.13, 1.82)	0.003	Age	1.03 (1.02–1.04)	<0.001
Muscle relaxants	1.45 (1.03–2.05)	0.036	CRRT	11.76 (4.77–28.96)	<0.001
Cardiac arrest	3.35 (2.22–5.04)	<0.001	Glucocorticoid	1.97 (1.27–3.05)	0.003
Brain injury	1.41 (1.04–1.90)	0.026	Muscle relaxants	2.53 (1.34–4.79)	0.004
Metastatic cancer	3.82 (2.61–5.59)	<0.001	Cardiac arrest	10.12 (6.02–17.02)	<0.001
			Brain injury	4.47 (3.02–6.62)	<0.001
			Metastatic cancer	2.88 (1.51–5.50)	0.001

Continuous variables are displayed as mean (standard deviation) or median (first quartile–third quartile); categorical variables are displayed as count (percentage).

CI, confidence interval; CRRT, continuous renal replacement therapy; OR, odds ratio; temperature₀, initial temperature after intensive care unit admission; temperature_max, maximum temperature during ICU stay; temperature_min, minimum temperature; temperature_mean, mean temperature.

However, an excessive increase in temperature_max was not associated with the 28-day mortality risk in the noninfectious cohort (p > 0.05).

Minimum value of temperature and prognosis of ARDS patients

To investigate the potential effect of hypothermia on the prognosis of ARDS patients, the relationship between temperature_min value and clinical outcomes was determined. There was a dramatic decrease in the 28-day mortality rate for temperature_min values from <32.0°C to 36.9°C, with the lowest mortality rate occurring in the 36.0–36.9°C subgroup. Patients whose temperature_min value exceeded 37.0°C had a higher 28-day mortality rate than those whose temperature_min ranged from 34.0°C to 36.9°C (Fig. 3A).

FIG. 3.

The relationship between temperature_min value and 28-day mortality in patients with ARDS. (A) The polynomial regression and bar graphs. (B) The crude OR and (C) the adjusted OR of 28-day mortality relative to 36.0–36.9°C for different categories of temperature_min during ICU stay. ARDS, acute respiratory distress syndrome; CI, confidence interval; ICU, intensive care unit; OR, odds ratio.

A line graph was used to present the dramatic relationship between temperature_min value and the 28-day mortality risk detected by the logistic regression model. Temperature_min values ranging from 36.0°C to 36.9°C were used as reference. As shown in Figure 3B, hypothermia_min was associated with increased risk of 28-day death in patients with ARDS, with the OR increasing stepwise from the 35.0°C to 35.9°C subgroup (OR: 1.4, 95% CI: 1.16–1.69, p < 0.001) to <34.0°C subgroup (OR: 4.15, 95% CI: 2.99–5.74, p < 0.001). Furthermore, a temperature_min value ≥37.0°C was determined to be an independent risk factor for 28-day death (OR: 2.83, 95% CI: 1.59–5.03, p < 0.001). A similar tendency was found in the adjusted model (Fig. 3C), but no statistically significant difference was found in the 35.0–35.9°C subgroup compared with the reference subgroup.

Discussion

Some previous limited data have suggested a potential relationship between body temperature change and the prognosis of patients with ARDS (Schell-Chaple et al., 2015). Although an association between temperature and several diseases has been widely detected, little is known about the relationship between body temperature and mortality in patients with ARDS. In the present study, there was a U-shaped relationship between body temperature during the ICU stay and 28-day all-cause mortality. Hyperthermia_max was associated with a decreased 28-day mortality in the infectious cohort. The increase in the temperature_max value was significantly associated with the risk reduction of 28-day death, with temperatures ranging from 36.5°C to 36.9°C as a reference. The results obtained from the noninfection cohort were not the same as those found in the infectious cohort. A mild increase in temperature_max value was detected as an independent protective factor against 28-day death in the noninfectious cohort. However, its protective effect was lost when temperature_max value exceeded 38.5°C. Hypothermia_min is associated with poor clinical outcomes in ARDS patients.

However, a temperature_min value above 37.0°C was also considered an independent risk factor. We believe that these interesting phenomena may be closely related to the activation of the immune system. Overall, the present work reveals an important relationship between body temperature and clinical prognosis in patients with ARDS. We expect that our findings could have implications for the treatment and management of patients with ARDS in the future.

ARDS is an acute hypoxemic respiratory failure seen in critically ill patients with high mortality. Dynamic monitoring of progression and prognosis prediction are essential to improve the clinical outcomes of patients with ARDS. Although several prognostic factors and prediction models for ARDS have been suggested in previous studies, they were not practicable in everyday clinical practice, especially in resource-limited settings, because of the requirements for detection and analysis approaches (Zhang and Ni, 2015; Zhao et al., 2017). In contrast, some routine vital signs may be feasible options for therapeutic monitoring and clinical outcome prediction under such environmental conditions. Body temperature is a significant human vital sign that can be monitored in hospitalized patients. The advantages of temperature monitoring include its noninvasive nature, low cost, and ease of performance. The relationship between body temperature and clinical outcomes has been investigated in several diseases, such as neurological injury, cardiac arrest, and sepsis (Diringer et al., 2004; Kushimoto et al., 2013; Laupland et al., 2012; Weinstein et al., 1983).

However, few studies that have focused on the relationship between body temperature and ARDS are still limited. In a prospective cohort study, including 450 patients with acute lung injury, Netzer et al. (2013) reported that hypothermia was independently associated with increased in-hospital mortality (relative risk: 1.68, 95% CI: 1.06–2.66, p = 0.03), but fever was not associated with mortality (all p > 0.05). Another study by Schell-Chaple et al. (2015) found that baseline body temperature was negatively related to the 90-day mortality of ARDS patients, and the odds of 90-day death were reduced by 15% with every 1°C increase (OR: 0.85, 95% CI: 0.73–0.98, p = 0.03). Patients with a temperature ≥39.5°C had the best outcome. This was an intriguing argument, but it seemed counterintuitive. Fever may reflect the natural defense mechanism of the host against injury and infection; however, hyperthermia may have a detrimental effect on patients. This counterintuitive conclusion can be explained by an insufficient sample size. Our results, which were obtained from a larger sample size, showed a U-shaped relationship rather than a linear relationship between body temperature and mortality in patients with ARDS.

The lowest mortality rate was observed in the middle parts of the temperature trace, rather than in the extremely high-temperature condition. In accordance with our results, a recent study also demonstrated a U-shaped relationship between the first 24-day mean body temperature and 1-year mortality in postcardiac surgery patients from the ICU (Xu et al., 2021).

In addition, we observed some interesting findings. First, the prognostic value of temperature_max value in the infectious and noninfectious cohorts was not fully identical. Second, the elevated temperature_max value was considered as the protective factor, while the elevated temperature_min was associated with poor outcomes with temperatures ranging from 36.0°C to 37.0°C as a reference. We speculated that these alterations might be associated with an altered inflammatory response of the host immune system against pathogens. The immune system has developed as a host defense system against a variety of diseases, especially infectious diseases. This procedure can lead to changes in body temperature, which is mainly characterized by fever.

Fever exerts both physiological and pathological effects. Different lines of evidence suggest that the increase in endothelial and epithelial injury caused by hyperthermia would result in worse lung injury by enhancing neutrophil extravasation and vascular hyperpermeability, both of which are also considered to be the mechanism of ARDS (Nagarsekar et al., 2012; Tulapurkar et al., 2012). As a defensive response, fever can inhibit the growth of pathogenic organisms and enhance the activity of the immune system to protect the body against infection and diseases. It has been reported that the effect of hyperthermia is different in the infection and noninfection cohorts (Lee et al., 2012; Young et al., 2012). However, whether infection acts as an interactive factor between body temperature and mortality in ARDS patients has not been investigated. Hyperthermia_max was associated with higher 28-day mortality rates in the infectious cohort; however, no difference was found in the noninfectious cohort. Interestingly, the inflection point of the polynomial regression curve of the infectious cohort was higher than that of the noninfectious cohort.

This could be explained by the fact that patients without infection did not benefit from the antipathogenic effect of fever; however, it was only affected by the detrimental effects of hyperthermia.

Elevated temperatures suggest better immune reserves for the fight against diseases. Therefore, patients with elevated temperatures have a better prognosis than those without elevated temperatures. However, a constantly high temperature (>37.0°C) would be an indicator of poor clinical outcomes, which can be derived from the relationship between the 28-day mortality rate and elevated temperature_min value. We believe that the persistently high temperature (>37.0°C) represents the continuous activation of the immune system and the nonsmooth process to fight against diseases. In addition, persistently high temperatures might also amplify side effects, for example, high metabolism and high oxygen consumption. Therefore, short-term and transient temperature elevation seemed to be a protective factor, while a constant temperature elevation may indicate a poor prognosis for patients with ARDS. Whether aggressive temperature-lowering intervention in patients with persistent hyperthermia affects the outcome of patients with ARDS would be an interesting issue to investigate. Thus far, the use of temperature-lowering interventions can yield different outcomes in patients with hyperthermia.

Although it is well accepted that antipyretic therapies are crucial for critically ill patients with hyperthermia, the value of antipyretic treatments remains ambiguous. An encouraging result from a previous study showed that external cooling could significantly lower the 14-day mortality in febrile patients with septic shock (Jaitovich and Sporn, 2013). However, other clinical trials have reported that fever control with antipyretic medicine did not influence the 90-day mortality of ICU patients (Young et al., 2015), and they may even lead to higher mortality (Lee et al., 2012). It should be noted that the methods of antipyretic intervention and whether patients have an infectious disease have an impact on the efficacy of antipyretic therapy. In the present study, antipyretic drugs and physical cooling were not included as confounders because of the difficulty in collecting relevant records. Although the effect of temperature-lowering interventions in critically ill patients is still not well understood, our findings may be influenced by it.

Although the relationship between hyperthermia in critically ill patients with ARDS and outcomes is not well understood, the impact of hypothermia on the mortality of patients with ARDS is essentially consistent with that of previous studies and our present findings. Hypothermia is associated with increased risk of death. Moreover, the relationship between hypothermia and increased mortality has been reported in both groups of patients with and without infection (Capuzzo et al., 2006; Kushimoto et al., 2013; Young et al., 2012). Although the mechanism underlying ARDS-related hypothermia remains unclear, iatrogenic hypothermia should not be ignored. The application of muscle relaxants, glucocorticoids, and mechanical ventilation can decrease body temperature and lead to hypothermia. Body temperature control is frequently applied in patients with cardiac arrest and brain injury (O'Leary et al., 2016; Panchal et al., 2020).

In our logistic regression model, the relationship between hypothermia and mortality remained robust after adjusting for confounding factors. Interestingly, temperature_min above 37.0°C was also considered an independent risk factor of 28-day mortality, which meant that persistent body temperatures above 37.0°C appeared to be harmful. We speculated that persistently high temperatures may amplify the side effects of hyperthermia and lead to patient mortality. It may also indicate a continuous, but not smooth, process to fight the disease and pathogens.

Interestingly, several preclinical studies have shown that therapeutic hypothermia could effectively improve gas exchange, correct the imbalance of anti-inflammatory and proinflammatory cytokines, and significantly improve the outcome of acute lung injury (Xia et al., 2016). In some case reports, therapeutic hypothermia has shown a protective effect in patients with acute respiratory failure (Hayek et al., 2017). These findings indicate that therapeutic hypothermia and spontaneous hypothermia might have different effects in patients with ARDS. In the present study, we could not distinguish between therapeutic and spontaneous hypothermia because no relevant data on therapeutic hypothermia were provided from the MIMIC-III database, which might have influenced our results. Thus, the effect of hypothermia on the prognosis of ARDS patients and whether there is a potential therapeutic significance of induced hypothermia in ARDS management is worthy of further study.

Our findings may provide valuable guidance for physicians regarding treatment and management strategies for patients with ARDS in emergency department and ICU departments. Changes in body temperature appear to be potential prognostic markers for ARDS. First, there was a significant association between hypothermia and increased mortality in patients with ARDS, especially at temperatures lower than 35.0°C. Therefore, hypothermia should be avoided. However, therapeutic hypothermia was not within the scope of this study. Second, the effects of hyperthermia were more complicated. Short-term and transient temperatures above 37.0°C would be a favorable prognostic marker for better outcomes, but persistent temperatures above 37.0°C were a significant risk factor for patients with ARDS. Therefore, in the management of ARDS, attention should be paid to the degree and duration of the temperature changes. Moreover, more attention should be paid to the differential roles of hyperthermia in infectious and noninfectious patients.

Our findings will be beneficial for the management and treatment of patients with ARDS. It is important to note here that whether patients could gain substantial benefit through therapeutic temperature control is inconclusive. Further high-quality clinical studies are needed to confirm our findings. Although our conclusions have limited generalizability, they may serve to drive further research on temperature management and ARDS.

This study has several limitations, and caution must be exercised when interpreting the results. First, the imbalance of confounding factors might have led to an outcome. Although we performed a logistic regression model to adjust for confounders, the bias could not be completely eliminated. Second, due to the difficulty in collecting relevant records, antipyretic treatment and therapeutic hypothermia were not included as confounders, which may have adversely affected our results. Third, as a rapidly changing biomarker, temperature trajectories may have different prognostic values in patients with ARDS. Trajectory analysis is a better method for confirming the potential relationship than the maximum and minimum. High-quality prospective studies and trajectory analysis methods are required to confirm our findings.

Conclusions

Hypothermia was associated with increased mortality in patients with ARDS, whereas the effect of hyperthermia on the mortality of patients with ARDS was not fully consistent in the infection and noninfection subgroups. Short-term and transient temperatures above 37.0°C would be beneficial to patients with ARDS, but extreme hyperthermia and persistent temperatures above 37.0°C should be avoided. The relationship between body temperature and prognosis in patients with ARDS is worthy of further investigation.

Authors' Contributions

Y.F.: Conceptualization, methodology, software, formal analysis, and writing––original draft. Y.Z.: Software, formal analysis, and visualization. Q.L.: Formal analysis and funding acquisition. X.H.: Conceptualization and formal analysis. Y.L., C.J., L.H., and C.R.: Writing––review and editing. X.Z.: Conceptualization, writing––review and editing, supervision, project administration, and funding acquisition.

Ethics Approval and Consent to Participate

No additional ethics approval or written informed consent was required for the present study. Ethics approval was obtained from the IRB of the Beth Israel Deaconess Medical Center and MIT. The requirement for informed consent to participate was waived by the IRB because the personal information of the patients was hidden in the MIMIC-III database.

Data Availability Statement

The datasets used in the present study are publicly available in the MIMIC-III v1.4 database (https://mimic.physionet.org). The structured data used in the present study are presented in the Supplementary Data. More detailed information and further inquiries can be directed to the corresponding author.

Author Disclosure Statement

The authors declare that they have no affiliations with or involvement in any organization or entity with any financial interest in the subject matter or materials discussed in this article.

Funding Information

This study was supported by the National Nature Science Foundation of China, Grant/Award Numbers: 31371509 and 82100244; 2020 Li Ka Shing Foundation Cross-Disciplinary Research Grant, Grant/Award Number: 2020LKSFG20B; 2022 Characteristics and Innovation Grant for College of Guangdong Province, Grant/Award Number: 2022KTSCX040; China Postdoctoral Science Foundation, Grant/Award Numbers: 2022M712012; Shandong Province Medical and Health Science and Technology Development Project, Grant/Award Numbers: 202003040648; and Binzhou Medical University Scientific Foundation, Grant/Award Number: BYFY2020KYQD40; 2021 Science and Technology Special Fund of Guangdong Province, Grant/Award number: 210713156872672.