Circadian rhythms of vital signs are associated with in-hospital mortality in critically ill patients: A retrospective observational study

ABSTRACT Vital signs have been widely used to assess the disease severity of patients, but there is still a lack of research on their circadian rhythms. The objective is to explore the circadian rhythms of vital signs in critically ill patients and establish an in-hospital mortality prediction model. Study patients from the recorded eICU Collaborative Research Database were included in the present analyses. The circadian rhythms of vital signs are analyzed in critically ill patients using the cosinor method. Logistic regression was used to screen independent predictors and establish a prediction model for in-hospital mortality by multivariate logistic regression analysis and to show in the nomogram. Internal validation is used to evaluate the prediction model by bootstrapping with 1000 resamples. A total of 29,448 patients were included in the current analyses. The Mesor, Amplitude, and Peak time of vital signs, such as heart rate (HR), temperature, respiration rate (RR), pulse oximetry-derived oxygen saturation (SpO2), and blood pressure (BP), were significant differences between survivors and non-survivors . Logistic regression analysis showed that Mesor, Amplitude, and Peak time of HR, RR, and SpO2 were independent predictors for in-hospital mortality in critically ill patients. The area under the curve (AUC) and c-index of the prediction model for the Medical intensive care unit (MICU) and Surgical intensive care unit (SICU) were 0.807 and 0.801, respectively. The Hosmer-Lemeshow test P-values were 0.076 and 0.085, respectively, demonstrating a good fit for the prediction model. The calibration curve and decision curve analysis (DCA) also demonstrated its accuracy and applicability. Internal validation assesses the consistency of the results. There were significant differences in the circadian rhythms of vital signs between survivors and non-survivors in critically ill patients. The prediction model established by the Mesor, Amplitude, and Peak time of HR, RR, and SpO2 combined with the Acute Physiology and Chronic Health Evaluation (APACHE) IV score has good predictive performance for in-hospital mortality and may eventually support clinical decision-making. Abbreviations: ICU: Intensive care unit; MICU: Medical intensive care unit; SICU: Surgical intensive care unit; HR: Heart rate; RR: Respiration rate; SpO2: Pulse oximetry-derived oxygen saturation; BP: Blood pressure; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; APACHE: Acute Physiology and Chronic Health Evaluation; bpm: beats per min; BMI: Body mass index; OR: Odd ratio; CI: Confidential interval; IQR: Interquartile range; SD: Standard deviation; ROC: Receiver operating characteristic; AUC: area under the curve; DCA: Decision curve analysis; IRB: Institutional review board.


Introduction
Many physiological and cellular processes cycle with time, with a period of roughly 24 hours between one peak and the next, and these circadian rhythms are the basis of "permissible homeostasis" (McKenna et al. 2017). Animal experiments have shown that the circadian rhythm of blood pressure and heart rate is significantly affected by feeding or light cycles, and the circadian rhythm of core temperature is also affected by different light intensities (Takasu et al. 2002;van den Buuse 1999). Patients in intensive care units (ICU) often experience circadian rhythms changes due to abnormal lighting, noise, and altered feeding schedules (Jobanputra et al. 2020). Circadian rhythms of the core temperature are present but altered in patients in the ICU, with the degree of circadian abnormality of the core temperature correlating with the severity of the disease (Gazendam et al. 2013). A recent study explored the circadian rhythms of vital signs in critically ill patients through three large retrospective clinical databases, and quantitative evaluation of these rhythms may help provide prognostic information for critically ill patients (Davidson et al. 2021).
An increasing number of studies have shown that circadian rhythms play a significant role in the onset and progression of the disease. TimeSignature analysis of peripheral blood gene expression profiles indicates that the gene expression rhythms of core clock genes rapidly become abnormal during critical illness (Maas et al. 2020). The environment of the intensive care unit (ICU), mechanical ventilation, medication, and critical condition have been reported to be important factors of sleep disturbance, causing severe abnormalities in sleep and circadian rhythms, which may increase mortality (Boyko et al. 2017;Telias and Wilcox 2019). Integrating the chronobiological approach into the daily practice of treating critical illness, monitoring circadian rhythms in the clinical environment, and alerting doctors when circadian rhythms seriously deteriorate may help improve outcomes for critically ill patients (McKenna et al. 2018).
The cosinor method is an essential tool for analyzing time series data in chronobiology. It is often used to analyze biological rhythm data, providing concise and intuitive estimates for periodic data through Mesor, Amplitude, and Acrophase (Cornelissen 2014;Doyle et al. 2021). In this study, we analyze the circadian rhythms of vital signs in critically ill patients using the cosinor method. This retrospective study attempts to develop a prediction model and shows it through a nomogram. The model is used primarily to predict early mortality in the hospital in critically ill patients, so we selected the continuous recording of vital signs from critically ill patients during the first full calendar day after admission to the ICU. This may be the first time that circadian rhythm parameters of vital signs have been used to establish a prediction model for critically ill patients.

Methods
The current study collected data from the eICU Collaboration Research Database v2.0, and the eICU Collaboration Research Database is a large public database that covers routine data on 200,859 patients admitted to intensive care units in the United States in 2014 and 2015, including vital signs, diagnostic information, treatment information, demographic information, and disease severity (Pollard et al. 2018). The author obtained access to the database (certification number: 40608375) and was responsible for data extraction.

Ethics statement
Data collection was carried out under the ethical standards of the institutional review board of the Massachusetts Institute of Technology (no. 0403000206) and with the Declaration of Helsinki of 1964 and its later amendments. This study analyzed a publicly available anonymized database with pre-existing institutional review board (IRB) approval.

Data collection
Patients were excluded for the following reasons: (1) absence of hospital vital status records; (2) absence of admission record of the MICU or SICU admission recording; (3) absence and inaccuracy of consecutive vital signs recording; (4) Vital signs recording < 6 or no circadian rhythms of vital signs during the first full calendar day in the ICU. The following variables were extracted from the eICU database: demographic parameters (age, sex, ethnicity, and body mass index (BMI)), consecutive parameters of vital signs (HR, temperature, RR, SpO2, and BP) (Vital signs were recorded at 5-minute intervals in the vitalPeriodic table of eICU database), diagnostic parameters (pulmonary disease, cardiovascular disease, gastrointestinal disease, renal disease, neurologic disease, endocrine disease, hematology disease, burn, neoplastic disease, infectious disease), treatment parameters (enteral nutrition, mechanical ventilation, surgery), APACHE IV score, in-hospital stay time, in-ICU stay time, and in-hospital mortality.

Outcomes
The primary outcome was in-hospital mortality. APACHE IV score is widely used to assess the severity and prognosis of critically ill patients, and it is based on variables within 24 hours of admission to the ICU and has good discrimination and calibration for predicting in-hospital mortality (Balkan et al. 2018;Zimmerman et al. 2006). Therefore, we also constructed a prediction model based on the APACHE IV score combined with the circadian rhythm parameters of vital signs to estimate clinical severity and prognosis in critically ill patients.

Statistical analysis
The circacompare R software package was used to perform the single cosinor transformation for the parameters of vital signs to estimate and statistically support differences in the Mesor, Amplitude, and Peak time between circadian rhythms (Parsons et al. 2020). Mesor is a rhythm-adjusted mean of vital signs. When Mesor increases, the patient's vital signs increase throughout the day. Amplitude is the distance from the highest point of the circadian rhythms of vital signs to Mesor. When Amplitude increases, the fluctuation range of one-day vital signs increases. Peak time is when the circadian rhythms of vital signs reaches the highest point.
Continuous variables were shown as mean, standard deviation, or median and interquartile range (IQR). Categorical variables were displayed as numbers and percentages. To compare the baseline characteristics of survivors and nonsurvivors in patients with critical illness, the Wilcoxon rank sum test and independent sample T-test were used for continuous variables when appropriate; the chi-square test was used for categorical variables. To investigate the impact of circadian rhythms on in-hospital mortality, compared the group mean and cosinor differences of the Mesor, Amplitude, and Peak time of vital signs between survivors and non-survivors in critically ill patients. In addition, the logistic regression was used to compare each circadian rhythms variable of vital signs (Covariate: APACHE IV score) in relation to outcomes with their respective lowest risk or highest risk considered as the reference. Finally, one prediction model was established by APACHE IV score and Mesor, Amplitude, and Peak time of HR, RR, and SpO2. The prediction model was shown through the nomogram. Predictor lines were drawn upward to confirm the points received from the nomogram. The sum of these points was located on the Total Points axis; a line was drawn downward to project on the bottom scales, which determined the risk of in-hospital mortality. After that, the prediction model was internally validated by bootstrapping with 1000 resamples. The Hosmer-Lemeshow test was used to assess the goodness of fit of the prediction model. The receiver operating characteristic (ROC) curve, AUC was used to evaluate the discriminative power of the model. The calibration curve was used to evaluate the predictive accuracy and conformity of the model, and calibration was assessed by bootstrapping with 1000 resamples (Huang et al. 2020). The DCA reflected the net benefit of the model for patients.
R software (version 4.0.2, www.rproject.org) was used for statistical analysis, and p < .05 was considered statistically significant.

Demographic and clinical characteristics
A total of 29,448 patients were included in the current analyses, including 21,657 patients whose HR have circadian rhythms, 2,338 patients whose temperature have circadian rhythms, 18,280 patients whose RR have circadian rhythms, 19,736 patients whose SpO2 have circadian rhythms, and 5,193 patients whose BP have circadian rhythms (Figure 1 and Supplementary Table  S1). The baseline characteristics of critically ill patients admitted to MICU or SICU are shown in Table 1. In MICU and SICU, compared with the survivors during hospitalization, the non-survivors were associated with older age, lower BMI and In-hospital stay time, and higher APACHE IV score and in-ICU stay time. Compared with the MICU, the SICU was associated with more cardiovascular disease, gastrointestinal disease, renal disease, neurologic disease, endocrine disease, hematology disease, burn, neoplastic disease, enteral nutrition, mechanical ventilation, and surgery and less pulmonary disease in survivors, while more gastrointestinal disease, renal disease, neurologic disease, endocrine disease, hematology disease, burn, Infectious disease, enteral nutrition, mechanical ventilation, and surgery and less cardiovascular disease in nonsurvivors (Supplementary Table S2).

Mesor, Amplitude, and Peak time of vital signs in relation to the outcomes
Logistic regression analysis showed that the circadian rhythms of vital signs were an independent predictor of in-hospital mortality in critically ill patients, according to the following results. After adjusted to the APACHE IV score, in the MICU, Mesor and Amplitude of HR per bpm increase were associated with a 1.02-fold increase in the risk of inhospital mortality. Moreover, the risk was highest in patients who had Peak time of HR reached between 06:00 and 12:00. Similar in the SICU, Mesor and Amplitude of HR per bpm increase were also associated with higher in-hospital mortality (1.01-fold and 1.03-fold increase, respectively). As to the rhythm of temperature, the risk of in-hospital mortality was lowest in patients who had Peak time of temperature reached between 18:00 and 24:00 in the MICU, and Amplitude of temperature per one centigrade increase was associated with a 1.15-fold increase in the risk of in-hospital mortality in the SICU. Mesor and Amplitude of RR per min increase were associated with higher in-hospital mortality (1.06-fold and 1.09-fold, respectively) in the MICU. In terms of RR rhythm, the risk was lowest in patients who had Peak time of RR reached between 18:00 and 24:00, and Mesor of RR per min increase was associated with a 1.04-fold higher in-hospital mortality in the SICU. Mesor and Amplitude of SpO2 were also associated with the risk of inhospital mortality (0.90-fold and 1.12-fold higher per one percent increase respectively in the MICU; 0.90-fold and 1.14-fold higher per one percent increase respectively in the SICU). Moreover, the risk was highest in patients who had Peak time of SpO2 reached between 06:00 and 12:00. As to the rhythm of BP, Mesor of SBP and DBP per mmHg increase was associated with a 0.98-fold increase in the risk of in-hospital mortality. The risk was lowest in patients who had Peak time of SBP and DBP reached between 12:00 and 18:00 in the SICU (18:00 and 24:00 in the MICU) ( Table 2).

Prediction model of in-hospital mortality for critically ill patients
Based on the results of the logistic regression analysis, we selected the APACHE IV score and the Mesor, Amplitude, and Peak time of HR, RR, and SpO2 to construct a prediction model by multivariate logistic regression analysis and to show in the nomogram ( Figure 3 and Supplementary Table S9). In the prediction model, to ensure that the included vital signs had complete circadian rhythm parameters, although some circadian rhythm parameters did not show significance, they were still included. For each critically ill patient, higher total points indicated a higher risk of in-hospital mortality. For example, in MICU, if a patient whose APACHE IV score is 53, Mesor of HR is 97.10 bpm, Amplitude of HR is 11.30 bpm, Peak time of HR reached 18:00-24:00, Mesor of RR is 33.28 /min, Amplitude of RR is 1.68 /min, Peak time of RR reached 00:00-06:00, Mesor of SpO2 is 93.85%, Amplitude of SpO2 is 1.19%, and Peak time of SpO2 reached 06:00-12:00, then the corresponding score of he will be approximately 22.5, 11.5, 2, 0, 13, 0.65, 4, 5, 1.15, and 0.65, respectively. The total score is approximately 60.45, indicating an approximately estimated risk of in-hospital mortality of 11% for this case. There is no significant change in the prediction model by bootstrapping with 1000 resamples (Supplementary Table S9). The Hosmer-Lemeshow test P-values of the MICU and SICU prediction models were 0.076 and 0.085, respectively (Supplementary Table  S10), demonstrating a good fit of the prediction model. The AUC of MICU and SICU were 0.807 and 0.801, respectively (Figure 4), reflecting the excellent discriminative power of the prediction model. Meanwhile, the prediction model had good accuracy and conformity, and the calibration curve was close to the ideal diagonal line (Supplementary Figure S1). Furthermore, DCA shows that the net benefit of the prediction model is better than the APACHE IV score (Supplementary Figure S2). These results demonstrated that our prediction model had significant potential for clinical decision-making.

Discussion
In this study, we demonstrated significant differences in circadian rhythms parameters of vital signs, such as HR, temperature, RR, SpO2, and BP, between survivors and non-survivors in critically ill patients. Among them, circadian rhythms changes of HR, RR, and SpO2 may be predictors of in-hospital mortality in critically ill patients. Therefore, we first developed a prediction model for inhospital mortality in critically ill patients using Mesor, Amplitude, and Peak time of HR, RR and SpO2 combined with APACHE IV score. Patients in the ICU often experience circadian rhythms changes due to abnormal lighting, noise, altered feeding schedules, mechanical ventilation, sedation, and the critical illness itself, and circadian rhythms changes in response to disease and injury can affect patient recovery and outcome (Billings and Watson 2015;Jobanputra et al. 2020). Previous studies have found that circadian rhythms change in some vital signs, such as HR and BP, are associated with in-hospital mortality in critically ill patients (Beyer et al. 2021;Yang et al. 2022). Normalization of circadian rhythms in critically ill patients may be an essential part of critical care, but as far as we know, there is no prediction model for the prognosis of critically ill patients using circadian rhythm parameters of vital signs, so this may be the first time. There are significant differences in vital signs such as HR, respiration, BP, and SpO2 between different categories of ICU (Zaidi et al. 2019). Compared to MICU, patients in SICU are more likely to experience more harm due to surgery (Park et al. 2018). Considering that patients admitted to MICU and SICU may have different circadian rhythms due to different diagnoses and treatments, we explored the impact of their respective circadian rhythms of vital signs on in-hospital mortality and constructed their separate prediction models.
Vital signs have been widely used to assess the disease severity and mortality of patients (Spångfors et al. 2019), but there is still a lack of research on their circadian rhythms. A recent study found that in the mortality cohort of patients with COVID-19, the HR amplitude decreased and the RR amplitude increased days before death (van Goor et al. 2022). The circadian rhythms change of vital signs may alert physicians to impending disease deterioration and organ dysfunction, and according to the differences in the circadian rhythms of vital signs, critically ill patients may be further stratified according to the severity of the disease in advance to help physicians' clinical decision-making. Through the mean and cosinor differences of the group, we found significant differences in the circadian rhythms of vital signs, such as HR, temperature, RR, SpO2, and BP, between survivors and nonsurvivors in critically ill. A previous study has found that a higher Amplitude of BP is associated with lower in-hospital mortality in critically ill patients (Beyer et al. 2021). Among patients with head trauma admitted to the ICU, a lower mean temperature and a higher Amplitude predicted higher in-hospital mortality (Culver et al. 2020). These outcomes are similar to the mean and cosinor differences. Another study also found that greater Mesor and Amplitude of HR were associated with higher in-hospital mortality in patients with stroke and critically ill (Yang et al. 2022). In critically ill patients requiring oxygen therapy, SpO2 levels were associated with in-hospital mortality in a U-shape (van den Boom et al. 2020). After logistic regression, we screened out the circadian rhythm parameters of HR, RR, and SpO2 as possible predictors of in-hospital mortality in critically ill patients, and established a prediction model together with the APACHE IV score.
The APACHE IV score has good discrimination and calibration to predict in-hospital mortality, based on variables within 24 hours after admission to the ICU, and has been widely used to assess the severity and prognosis of critically ill patients (Balkan et al. 2018;Zimmerman et al. 2006). The latter study concluded that the APACHE IV score system showed satisfactory discriminative power and good calibration ability, but was not appropriate to be used as a single criterion for admission to the ICU (Choi et al. 2017). The precision of the APACHE IV score as a prediction model should be dynamic and regularly retested; if the accuracy decreases, it should be revised and updated (Zimmerman et al. 2006). Currently, we have not found that the variables of the APACHE IV score include circadian rhythm parameters. We combined the circadian rhythm parameters of HR, RR, and SpO2 with APACHE IV scores to construct a prediction model that still showed good discriminative power, accuracy, and conformity. These results indicate that our prediction model has significant potential for clinical decision-making. Visualizing the prediction model through the nomogram provides clinicians with a simple and intuitive practical prediction tool. This may be the first time the circadian rhythm parameters of vital signs combined with the APACHE IV score were used to construct a prediction model for the prognosis of critically ill patients, and in the future update and review of the APACHE IV score, the inclusion of circadian rhythm parameters of vital signs may also be considered.

Limitations
First, this study was retrospective and could not well set up proper intervention measures, so intervention factors were not considered. Second, to make the analysis comprehensive, we included as many vital sign parameters as possible. However, due to the missing database recording, samples for vital sign parameters, such as temperature and BP, compared to other vital sign parameters, were so small that their results were difficult to interpret. Third, more samples indicate more stable and significant outcomes, So we evaluated the effects only by internal validation instead of sacrificing samples for external verification. If better data and more samples are available in the future, we will seek further external verification.

Conclusions
We demonstrated significant differences in circadian rhythms of vital signs, such as HR, temperature, RR, SpO2, and BP, between survivors and non-survivors in critically ill patients. Among them, circadian rhythms changes of HR, RR, and SpO2 may be predictors of inhospital mortality in critically ill patients. Therefore, we developed a prediction model for in-hospital mortality in critically ill patients using Mesor, Amplitude, and Peak time of HR, RR, and SpO2 combined with APACHE IV score and shows it through nomogram, and which has good predictive performance for in-hospital mortality and may eventually support clinical decision-making. required training course (the Collaborative Institutional Training Initiative) and requesting access to the eICU Collaborative Research Database, researchers can seek to use the database. The author ZN Y obtained the access of the database (certification number: 40608375).