The influence of biological rhythms on the initial onset of status epilepticus in critically ill inpatients and the study of its predictive Model

ABSTRACT This study aims to explore the relationship between the circadian rhythms of critically ill patients and the incidence of Status Epilepticus (SE), and to develop a predictive model based on circadian rhythm indicators and clinical factors. We conducted a diurnal rhythm analysis of vital sign data from 4413 patients, discovering significant differences in the circadian rhythms of body temperature, blood oxygen saturation, and heart rate between the SE and non-SE groups, which were correlated with the incidence of SE. We also employed various machine learning algorithms to identify the ten most significant variables and developed a predictive model with strong performance and clinical applicability. Our research provides a new perspective and methodology for the study of biological rhythms in critically ill patients, offering new evidence and tools for the prevention and treatment of SE. Our findings are consistent or similar to some in the literature, while differing from or supplementing others. We observed significant differences in the vital signs of epileptic patients at different times of the day across various diagnostic time groups, reflecting the regulatory effects of circadian rhythms. We suggest heightened monitoring and intervention of vital signs in critically ill patients, especially during late night to early morning hours, to reduce the risk of SE and provide more personalized treatment plans.


Introduction
Status Epilepticus (SE) is characterized as a medical condition where epileptic seizures persist for more than five minutes or occur successively without complete restoration of consciousness within a five-minute span (Glauser et al. 2016;Trinka et al. 2015).Recognized as a frequent neurological emergency, SE can result in neuronal damage, cognitive dysfunction, and even mortality (Sánchez and Rincon 2016).The worldwide incidence of SE varies, ranging from 5.2 to 61.4 cases per 100,000 individuals annually, with a mortality rate spanning from 7.3% to 34.0% (Trinka and Kalviainen 2017).SE's etiology is multifaceted, encompassing infections, intoxications, metabolic disorders, brain tumors, traumatic brain injuries, ischemic or hemorrhagic strokes, and genetic or congenital diseases (Chen et al. 2007;Sutter et al. 2015).The primary therapeutic objective for SE involves promptly halting seizures, preventing neuronal damage and cerebral edema, stabilizing vital signs, and addressing the underlying cause (Glauser et al. 2016).Nonetheless, diagnosing and treating SE present considerable challenges, including variable diagnostic criteria, non-standardized treatment protocols, drug intolerance or insensitivity, and unforeseen complications (Hirsch and Gaspard 2013).Thus, identifying efficient predictive factors, assessment tools, and developing precise predictive models are essential for enhancing the diagnosis, treatment, and prognosis of SE patients.
Recent studies have increasingly focused on the connection between the circadian rhythms of critically ill patients and diseases.Circadian rhythms are self-sustained physiological or behavioral fluctuations that typically occur within a 24 hr period (Cederroth et al. 2019;Patke et al. 2020), influenced by external "timekeepers" or zeitgebers such as the light/dark cycle.These rhythms are regulated by internal biological clocks (Korf 2022;Raible et al. 2017), notably the suprachiasmatic nucleus (SCN) of the hypothalamus, and are evident in various physiological processes including the sleep-wake cycle, body temperature, and hormone secretion (Van Drunen and Eckel-Mahan 2021).Recent studies have highlighted that disruptions in circadian rhythms significantly elevate the risk of seizures, particularly in intensive care settings (Jin et al. 2020).The circadian rhythm, which serves as an internal clock regulating numerous physiological processes, is intricately linked to an increased incidence of status epilepticus.Research further indicates that targeted circadian interventions, such as optimizing lighting and sleep schedules, can effectively mitigate the severity and frequency of seizures (Wang et al. 2023).These insights underscore the critical importance of incorporating circadian rhythm management into clinical treatments, especially for critically ill patients who are at a heightened risk of seizures.
Examining the relationship between diurnal rhythms and SE in patients is of great importance for understanding the pathogenesis of SE, predicting risks and outcomes, and developing personalized treatment plans (Gourineni and Zee 2005;Logan and McClung 2019).Despite this, there is a lack of research on the diurnal rhythms of critically ill patients, particularly those with SE.This is due to the nervous system dysfunction associated with SE, which can disrupt the stability and synchrony of the biological clock and consequently alter diurnal rhythms.Such alterations may be linked to the occurrence, progression, and prognosis of SE, potentially inducing or exacerbating seizures or influencing their type, frequency, and duration (Oldham et al. 2016;Rao et al. 2021).This study's aim is to explore the relationship between the biological rhythms of critically ill patients and SE and to create a predictive model integrating circadian rhythm indicators with clinical factors.This research utilized an innovative approach, assessing patients' circadian rhythm status using various biological parameters (e.g., heart rate, blood pressure, body temperature, blood oxygen saturation) and merging these with clinical data (like age, GCS [Glasgow Coma Scale] score, comorbidities) to formulate a predictive model for SE risk.

Data source
The research utilized the eICU Collaborative Research Database, a comprehensive, multicenter Intensive Care Unit (ICU) database established through a collaboration between the Laboratory for Computational Physiology at the Massachusetts Institute of Technology and Philips Healthcare (Pollard et al. 2018).Spanning 208 hospitals across the United States and encompassing the years 2014 to 2015, the database contains information on 200,859 patient admissions.Available on the PhysioNet website, the eICU database is publicly accessible, though it requires registration and the successful completion of online training and testing.The study was conducted under the institutional training collaboration initiative number 11366502.The eICU database serves as a vital resource in critical care medicine, offering extensive data that supports the study and enhancement of patient treatment.

Clinical diagnosis sources and basis in the database
The primary outcome variable in this study was the incidence of SE, a critical neurological emergency demanding immediate recognition and intervention.Identification of SE cases was based on diagnostic information from the admissiondx (admission diagnoses) and diagnosis (diagnoses during ICU stay) tables, as well as criteria defined by the APACHE IV system.Specifically, SE was identified as a comprehensive convulsive or nonconvulsive epileptic seizure lasting over five minutes or repeated seizures over five minutes without a return to baseline consciousness between episodes (Trinka et al. 2015).The occurrence of other complications during the ICU stay, such as infections, cardiovascular events, respiratory failure, and renal insufficiency, was recorded based on diagnoses in the admissiondx and diagnosis tables.The sources of vital signs and laboratory test results during the ICU stay, such as heart rate, pulse, temperature, blood pressure, blood gas, electrolytes, and coagulation, were also extracted from the vitalaperiodic, vitalperiodic, and lab tables in the eICU database.

Data quality assurance and processing steps
To ensure data quality and validity, we processed the raw data through several steps: Selection of Subsets: We selected data tables relevant to our variables of interest, including "admissiondx," "diagnosis," "apacheApsVar" (APACHE IV assessment indicators), and "apachePredVar" (APACHE IV predictive indicators), merging them into one dataset.
Duplicate Removal: We checked for and removed duplicate patient records, retaining only the first instance.
Handling Missing Values: We addressed missing values differently based on variable type.For numerical variables, we employed multiple imputation, and for categorical variables, we used mode imputation.
Data Consistency and Error Handling: We corrected or removed inconsistent or erroneous data.For example, we standardized date formats to YYYY-MM-DD, harmonized diagnosis names to ICD-10 codes, and removed outliers (such as records with ages below 0 or above 120).
Data Transformation: Based on the definitions of the APACHE IV system, we transformed status epilepticus and other complications into binary variables (0/1).We also standardized or log-transformed some numerical variables based on their distribution.
Patients were selected according to the study's inclusion and exclusion criteria, as detailed in the study flow chart (Figure 1) and supplementary material (Figure S1).These steps resulted in a clean, well-organized, and effective dataset, laying the groundwork for subsequent data analysis and modeling.

Analysis method
The aim of this research was to explore the relationship between the circadian rhythms of critically ill patients and SE, using statistical analyses performed with R software (version 4.3.1)and Python software (version 3.11).The data were randomly divided into a training set and a validation set in a 7:3 ratio, with the training set being used for feature selection and model building, and the validation set for evaluating the model's performance.Continuous variables were described using median and interquartile range (IQR) (Q1, Q3) due to their non-normal distribution.The Wilcoxon rank-sum test was employed for group comparisons, and categorical variables were described by count and proportion and compared using the Chi-square test or Fisher's exact test.The Mean Cosine Similarity method was utilized to assess the consistency of patients' diurnal rhythms.This approach, often applied in time-series data comparison, calculates the cosine angle between different time series to gauge their similarity (de Vos et al. 2022).In this study, it was applied to circadian rhythm data of patients, with 24 hr circadian rhythm data for each patient being extracted and modeled to derive Figure 1.A road map to assess the relationship between circadian rhythms and status epilepticus.The roadmap outlined in the article describes the analysis process undertaken by the researchers.Analysis samples and a vast array of variables were collected from the eICU database.These variables were categorized into four key areas: circadian rhythm indicators, demographic information, laboratory data, and details pertaining to complications as well as the clinical scoring system.Subsequently, two major analyses were conducted to evaluate the relationship between these variables and the incidence of Status Epilepticus.
rhythm parameters such as mesor, amplitude and peak time.All hypothesis tests were two-tailed, and a p-value below 0.05 was considered statistically significant.

Relationship between diurnal rhythm and SE
The CircaCompare R package was used to perform single cosine transformations on vital sign parameters (Parsons et al. 2020).The single cosine transformation, a fitting method based on the cosine function, effectively extracts the diurnal rhythm characteristics of time-series data (Marler et al. 2006).The goal was to evaluate the differences in diurnal rhythms of vital signs between the SE and non-SE groups and their association with SE.Parameters of diurnal rhythms, such as the mesor, amplitude, and peak time, were estimated and compared (Hou et al. 2022).The mesor is the rhythmadjusted average of a vital sign, reflecting its overall level throughout the day.Amplitude is the distance from the mean level to the peak of the diurnal rhythm, indicating the range of fluctuation of the vital sign over a day.Peak time is the time when the diurnal rhythm of the vital sign reaches its peak, reflecting the phase of the rhythm.Furthermore, vectors were constructed using the diurnal rhythm parameters of each group, and Euclidean distance and Root Mean Square Error (RMSE) were calculated between these vectors.The Euclidean distance measures the absolute disparity between the two groups' rhythmic patterns, while RMSE offers an average magnitude of these variations across the diurnal cycle.These measures were derived from the diurnal rhythm parameters, providing insights into the rhythmic stability and physiological variations between the SE and non-SE groups.The distributions of these diurnal rhythm parameters, along with the Euclidean distance and RMSE values, were visually represented to facilitate comparative analysis (Alfakih 2018;Karunasingha 2022).For SE patients, ANOVA was used to test interaction effects between diurnal rhythm indicators and the time of onset, with results displayed in box plots.

Model performance evaluation
The model's classification performance was assessed by plotting the ROC curve and calculating the AUC.The model's calibration, a crucial factor, was evaluated by examining the consistency between its predictions and actual outcomes using 1000 bootstrap resamplings to generate calibration curves.The Clinical Impact Curve (CIC) was used to assess the model's clinical applicability, understanding its impact on clinical decisions and practical value.The model's generalizability was tested using validation set data as new input, applying the Hosmer-Lemeshow test to assess the model's fit.This step was crucial in determining the model's potential to generalize to new sample sets while maintaining robust generalizability.

Baseline clinical characteristics
The study analyzed data from 4413 patients, comprising 117 individuals with Status Epilepticus (SE) and 4296 without SE.Notable differences in baseline clinical characteristics were identified between these groups.Patients with SE tended to be younger, had lower blood potassium levels, higher blood sodium levels, and lower GCS scores.Additionally, the SE group exhibited increased incidences of conditions like pulmonary infection, meningitis, diabetes, stroke, coma, depression, hypocalcemia, hypoglycemia, heart failure, and myocardial infarction.No significant disparities were noted in other clinical parameters such as RBC (Red Blood Cells), WBC (White Blood Cells), MCH (Mean Corpuscular Hemoglobin), Hgb (Hemoglobin), Hct (Hematocrit), Lymphs (Lymphocytes), platelets, PT (Prothrombin Time), glucose, bicarbonate, etc.These baseline characteristics are detailed in Table 1.

Circadian rhythm patterns in biological parameters
Single cosine transformations were performed on rhythm indicator parameters for both groups, as shown in Figure 2. The SE group had significantly lower temperature mesor and peak time values than the non-SE group (p < 0.05), indicating more unstable diurnal temperature rhythms and earlier peak times.The SaO2 mesor in the SE group was significantly higher (p < 0.05), suggesting an elevated diurnal rhythm in blood oxygen saturation, possibly reflecting a compensatory mechanism for hypoxemia.The SE group also had significantly higher systemic diastolic mesor (p < 0.05), While it is noted that there may be some overlap between the two pairs of variables, namely Pulmonary infections and Infections as well as Head infections and CNS infection, it is essential to emphasize that all four variables hold significant importance in the study of the onset of status epilepticus.Each of these variables can contribute valuable insights into the complex interplay of factors that may lead to the development of this condition.Pulmonary infections and infections in general can reflect the role of systemic infections in provoking seizures, while head infections and CNS infections specifically highlight the potential impact of central nervous system involvement.By examining these variables collectively, we can gain a comprehensive understanding of their collective influence on the occurrence of status epilepticus.indicating increased vascular resistance.The amplitude of Systemic systolic in the SE group was significantly higher (p < 0.05), pointing to greater variability and potential instability in the cardiovascular system.
The study used Euclidean distance and Root Mean Square Error (RMSE) to assess the consistency of diurnal rhythms in different vital sign indicators between SE and non-SE patients.As shown in Supplementary Material Figure S2, lower values in Euclidean distance for temperature, heart rate, respiration, and SaO2 indicated smaller differences in diurnal rhythms between the two groups.Conversely, higher distances for diastolic and systolic pressure suggested less consistency in their rhythms.The RMSE results revealed that body temperature had the lowest prediction error, while systolic pressure had the highest, indicating the highest and lowest consistency in diurnal rhythm predictions between the groups, respectively.
The comparative analysis of circadian rhythms in physiological parameters between groups with and without conversion to status epilepticus revealed significant variations, as detailed in Table 2.The temperature mesor was notably lower in the conversion group (36.22 vs. 36.95,p < 0.001), indicating a distinct circadian disruption in patients who convert.Similarly, heart rate amplitude was significantly higher in the conversion group (49.09 vs. 42.2,p = 0.007), suggesting increased cardiovascular activity.Noteworthy differences were also observed in peak times, with heart rate, respiration, and Sao2 showing marked variations in timing, pointing to altered physiological rhythms in the conversion group.For instance, the heart rate peak time shifted dramatically from 2.32 hours in the no conversion group to 18.81 hours in the conversion group (p < 0.001).This shift, along with similar trends in other parameters such as systemic diastolic peak time and Sao2, underscores the impact of status epilepticus conversion on bodily functions.These findings underscore the potential of cosinor analysis in identifying critical physiological alterations associated with the conversion to status epilepticus, emphasizing its value in clinical assessments and interventions.
This study provides a detailed description of the distribution of circadian rhythm indicators in patients with SE during various time intervals of the day (00:00-06:00, 06:00-12:00, 12:00-18:00, 18:00-24:00).Box plots (see Figure 3) reveal significant differences in the median values of body temperature, heart rate, SaO2 and systolic pressure during the night to early morning period (00:00-06:00) compared to other time intervals, significantly reflecting the variations in circadian rhythm indicators across different times of the day.Moreover, the median values, interquartile ranges (IQR), and outliers for each rhythm indicator are clearly presented in the box plots, with the whisker length indicating variability beyond the quartiles.Validated by Analysis of Variance (ANOVA), these results demonstrate the strong link between circadian rhythm indicators and the start time of SE, thus affirming the utility of these indicators for predicting and modeling the disease.This breakthrough provides a strong basis for further biostatistical modeling, increases our The data columns present the mean values for both groups -those without conversion ("No Conversion") and those with conversion ("Conversion").P-values are used to assess the statistical significance of differences between the two groups, with p < 0.05 indicating statistical significance.An independent sample t-test was employed for the statistical analysis, and the significance levels for various parameters are provided in the table.
knowledge of the physiological rhythms of SE patients, and helps to more accurately identify and forecast SE episodes in clinical practice.

Feature importance
RFE and four other machine learning algorithms (LR, RF, XGBoost, and ANN) identified the ten most important variables, including temperature mesor, stroke, age, hypoglycemia, potassium, GCS, myocardial infarction, hyponatremia, temperature peak time, and SaO2 mesor.The performance of each algorithm was evaluated using ROC curves and AUC, as shown in Supplementary Material Figure S3.Integrating AUC scores and clinical guidelines, these ten variables were chosen to build the final model, encompassing several circadian rhythm indicators and clinical factors.

Construction of a predictive nomogram
A predictive nomogram (Figure 4) was developed to offer an intuitive assessment of SE risk.This graph offers an intuitive interpretation of the key clinical features and their association with SE risk.Specifically, the distribution of age spanned from 10 to 90 years, with blood potassium levels varying from 1 to 11 mmol/L, and GCS scores from 3 to 15.The observed data range for temperature rhythm (temperature mesor) extended from 32.5°C to 39.5°C, while the distributions of temperature peak time and arterial oxygen saturation rhythm (SaO2 mesor) were also precisely depicted.The Total Points section summarized the combined effect of all features, directly correlating with Disease Risk.This predictive tool aims to provide a precise, visual assessment for clinical use, facilitating effective determination of a patient's risk of developing SE.SaO2), systemic diastolic, and systemic systolic are segmented into time groups: 00:00-06:00, 06:00-12:00, 12:00-18:00, and 18:00-24:00.Each boxplot illustrates the median, interquartile range (IQR), and the outliers for each parameter.The extension of the whiskers indicates variability outside the upper and lower quartiles, providing insights into the circadian variation of vital signs.Significant differences among the time groups were identified, as indicated by ANOVA testing, suggesting significant differences (p < 0.05) in the distribution of body temperature, heart rate, blood oxygen saturation, and systemic systolic during the 00:00-06:00 time period compared to other time intervals.These variations may be associated with the diurnal rhythm changes in patients.

Performance and internal validation of the nomogram
To evaluate the credibility and reliability of the model, comprehensive performance and internal validation analyses were conducted.Firstly, the model's discriminative ability, or its capacity to correctly classify diseased and non-diseased patients at different thresholds, was measured using the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve.

Discussion
This study aimed to explore the correlation between the circadian rhythms of critically ill patients and SE, and to build a predictive model based on circadian rhythm indicators and clinical factors.By utilizing Mean Cosine Similarity and Single Cosine Transformation methods, the diurnal rhythms of vital signs were analyzed and compared across different patient groups.Significant distinctions in the diurnal rhythms of body temperature, blood oxygen saturation, and heart rate were observed in SE patients, which correlated with the occurrence of SE.Additionally, various machine learning algorithms were employed to identify essential variables and a predictive model with excellent performance and clinical relevance was constructed.This research provides fresh perspectives and methodologies for studying the circadian rhythms in critically ill patients and offers new evidence and tools for the prevention and treatment of SE.
The findings align with some existing studies.For instance, significant differences were found in the mesor and amplitudes of body temperature rhythms between SE patients and non-SE patients, linked to the severity and prognosis of SE (Rossetti 2011) Differences were also noted in blood oxygen saturation rhythms, associated with the etiology and prognosis of SE (Bateman et al. 2008;Moseley et al. 2012), and in blood pressure rhythms, related to SE types and outcomes (Viloria-Alebesque et al. 2023).These results imply that changes in vital sign rhythms in SE patients might reflect the physiological impact of SE (Smyk and Van Luijtelaar 2020) and could serve as risk factors or outcome indicators.
The study also offers new insights.For example, contrary to a study that found no significant differences in heart rate rhythms between SE and non-SE patients (Khan et al. 2018), this research observed significant differences, positively correlated with SE incidence.This disparity could be attributed to the larger sample size and more sensitive statistical methods used.Additionally, the study identified significant differences in the amplitude of respiratory rhythms between SE and non-SE patients, a novel finding not previously reported, suggesting that changes in respiratory rhythms might relate to SE occurrence or its impact on the respiratory system.
The research analyzed circadian rhythm indicators in patients with SE at different diagnostic times, finding significant diurnal variations.This underscores the critical role of circadian rhythms in the physiological regulation of SE patients.The results are consistent with current literature (Karoly et al. 2021), supporting the validity and effectiveness of using these indicators for disease prediction and modeling.Notably, the mesor of various rhythmic indicators such as temperature, heart rate, respiratory rate, blood oxygen saturation, diastolic and systolic pressure often peaked in the early morning hours (00:00-06:00).This period may be key for clinical monitoring and intervention to reduce SE risk and tailor treatment plans.Additionally, existing research supports the relationship between epilepsy and the circadian rhythms of sleep, with partial seizures often occurring at specific times, indicating a link between epilepsy onset and biological rhythms (Loddenkemper et al. 2011).Another study emphasizes the connection between diurnal variations in epilepsy seizures and the body's hormone secretion, which modulates neuronal cell membrane excitability (Taubøll et al. 2015).Additional findings in the field suggest that the interplay between circadian rhythms and epilepsy seizures may be governed by multiple regulatory layers, including neurotransmitters, ion channels, and gene expression (Catterall 2012;Cho 2012;Jaworski 2020;Re et al. 2020).
Furthermore, the study developed a predictive model incorporating circadian rhythm indicators and clinical factors, exhibiting high accuracy and consistency.Utilizing various machine learning algorithms, the research identified key variables, notably several circadian rhythm indicators and clinical factors.Significant predictors of SE were identified: mesor and peak times of temperature rhythms, amplitude of respiratory rhythms, and mesor of blood oxygen saturation and heart rate rhythms.These findings align with the results from univariate analysis.Moreover, age, blood potassium and sodium levels, and Glasgow Coma Scale scores also emerged as crucial predictors of SE, corroborating with baseline clinical characteristics analysis results.This suggests that the predictive model is founded on reliable data and sound assumptions, capturing the complex and multifactorial aspects of SE.A comprehensive predictive nomogram was also developed, visually delineating the contribution and significance of each variable in forecasting SE.This tool aims to offer a precise and user-friendly assessment method for clinical application, enhancing the evaluation of patients' risk of developing SE.
In the future, we propose extending our investigations across diverse geographical and cultural contexts to validate the universality of our findings.Such studies are crucial for determining whether the effects of biological rhythms observed are consistent across various settings or subject to environmental and cultural variations.Additionally, we recommend prospective longitudinal research to solidify the causal links suggested by our initial findings, by tracking changes in biological rhythms over time and assessing their response to specific interventions.This methodological approach would provide robust evidence of causality and enhance our understanding of how these rhythms influence health and behavior longitudinally.Furthermore, exploring less studied biological rhythms, such as ultradian and infradian rhythms, could uncover complex interactions and dependencies, providing new insights into biological temporal dynamics.Lastly, interdisciplinary research that examines the interactions between different biological rhythms, such as the interplay between circadian and seasonal rhythms in mood disorders, using advanced modeling techniques, could reveal intricate relationships crucial for developing targeted interventions and fostering a comprehensive approach to health and disease management.This holistic exploration of biological rhythms and their interactions will not only expand the current understanding but also pave the way for innovative therapeutic strategies.

Strengths and limitations
The strengths of this study include that it is a large-scale data from the eICU Collaborative Research Database enhances the robustness and relevance of the findings.
Application of sophisticated machine learning algorithms for nuanced analysis of circadian rhythms.Groundbreaking exploration of the relationship between circadian rhythms and SE onset, broadening understanding in neurology.Development of a predictive model, offering practical utility in early detection and management of SE.The study acknowledges several limitations and areas for future research.Firstly, its retrospective nature may introduce selection bias and confounding factors.Although statistical methods were employed to mitigate confounders, the potential impact of unknown or unmeasured factors cannot be entirely excluded.Consequently, the findings necessitate validation and replication in larger and more diverse samples across multiple centers.Secondly, the research relied on 24 hours of circadian rhythm data post-admission, possibly not capturing the full extent of long-term changes and dynamics in patients' circadian rhythms.Future research could benefit from incorporating longer-term or more frequent circadian rhythm data to enhance the sensitivity and accuracy of findings.Lastly, the focus was solely on diurnal rhythms of vital signs, omitting other biological rhythms that might influence SE, such as individual variances and environmental factors.These aspects could significantly affect patients' circadian rhythms and the occurrence of SE, suggesting a need for more holistic research approaches in future studies.
In the discussion of potential future research directions, we propose extending our investigations across diverse geographical and cultural contexts to validate the universality of our findings.Such studies are crucial for determining whether the effects of biological rhythms observed are consistent across various settings or subject to environmental and cultural variations.Additionally, we recommend prospective longitudinal research to solidify the causal links suggested by our initial findings, by tracking changes in biological rhythms over time and assessing their response to specific interventions.This methodological approach would provide robust evidence of causality and enhance our understanding of how these rhythms influence health and behavior longitudinally.Future research could benefit from incorporating longer-term or more frequent circadian rhythm data to enhance the sensitivity and accuracy of findings.Lastly, the focus was solely on diurnal rhythms of vital signs, omitting other biological rhythms that might influence SE, such as individual variances, inhibition of natural sleep stages and changes in temperature and other circadian rhythms related to drugs and environmental factors.These aspects could significantly affect patients' circadian rhythms and the occurrence of SE, suggesting a need for more holistic research approaches in future studies.

Conclusion
In conclusion, the study establishes a strong link between the circadian rhythms of critically ill patients and SE occurrence.The stability and balance of these rhythms act as protective or risk factors for SE.The developed predictive model, combining circadian rhythm indicators and clinical factors, offers an accurate and intuitive tool for clinical assessment.These findings provide insights into the pathogenesis of SE and support clinical decision-making, opening new avenues for research into the circadian rhythms of critically ill patients and enhancing the diagnosis and treatment of SE.

Five
machine learning algorithms (RFE [Recursive Feature Elimination], LR [Logistic Regression], RF [Random Forest], XGBoost [eXtreme Gradient Boosting], and ANN [Artificial Neural Network]) were used to predict SE occurrence probability and select the most important variables from 58 clinical features.The performance of each algorithm was evaluated using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC).Based on AUC scores and clinical guidelines, 10 most important variables were chosen from those identified by RFE for the final model, including several circadian rhythm indicators and clinical factors.The final model achieved an AUC of 0.882, outperforming other algorithms.A line graph depicted the contribution and importance of each variable in predicting SE.

Figure 2 .
Figure 2. Comparative analysis of diurnal variations in biological rhythm indicators among patients with and without status epilepticus.This figure presents the 24 hr variations in six principal biological rhythm indicators (heart rate, respiratory rate, blood oxygen saturation (SaO2), systemic systolic, systemic diastolic, and body temperature) among patients with and without status epilepticus.The data points denote actual measurements, whereas the connecting lines illustrate the temporal trends of these indicators.A comparative analysis of the data from both patient groups reveals the influence of status epilepticus on biological rhythm indicators, offering vital insights for clinical diagnosis and treatment strategies.Each trend line in the graph is derived from smoothed observational values, aiming to more clearly depict the overall data trends.

Figure 3 .
Figure 3. Relationship between the onset time of status epilepticus and circadian rhythm indicators.Boxplot representation of circadian rhythm index across four times intervals for Status Epilepticus patients.The distribution of temperature, heart rate, respiration, blood oxygen saturation (SaO2), systemic diastolic, and systemic systolic are segmented into time groups: 00:00-06:00, 06:00-12:00, 12:00-18:00, and 18:00-24:00.Each boxplot illustrates the median, interquartile range (IQR), and the outliers for each parameter.The extension of the whiskers indicates variability outside the upper and lower quartiles, providing insights into the circadian variation of vital signs.Significant differences among the time groups were identified, as indicated by ANOVA testing, suggesting significant differences (p < 0.05) in the distribution of body temperature, heart rate, blood oxygen saturation, and systemic systolic during the 00:00-06:00 time period compared to other time intervals.These variations may be associated with the diurnal rhythm changes in patients.
Figure S4.A and Figure S4.B displayed the model's AUC curves for both training and test sets, achieving 0.826 (95% CI: 0.776-0.876)and 0.882 (95% CI: 0.811-0.953),respectively.Sensitivity and specificity were 0.808 and 0.723; Youden's index was 0.488, with a cutoff value of 0.227, indicating excellent discriminative ability.The model's calibration was further assessed, the congruence between predicted results and actual observations, visualized using calibration curves generated from 1000 bootstrap resamples (Figure S4.C).Ideally, the calibration curve should closely approximate the 45-degree diagonal line.The model's calibration curve closely matched the ideal, indicating good calibration.Additionally, the model's clinical utility was evaluated using the CIC, measuring its value in decision support.The CIC plot (Figure S4.D) illustrated the model's impact and performance on clinical decisions at various thresholds.The model significantly enhanced decision quality at different thresholds, providing effective guidance for clinical practice.Finally, to verify the model's generalizability, the Hosmer-Lemeshow test was used to evaluate consistency between the model on new data and training data.The test results indicated that the model fit well on new data, demonstrating robust generalizability.

Figure 4 .
Figure 4.The predictive line chart shows the relationship between disease risk and 10 related factors.The graph employs a predictive line chart to elucidate the relationship between disease risk and 10 pertinent factors.These factors, the paramount variables selected from 4413 patient data via the RFE machine learning algorithm, encompass temperature rhythm, stroke, age, hypoglycemia, serum potassium, Glasgow Coma Score (GCS), myocardial infarction, hyponatremia, temperature peak time, and arterial oxygen saturation rhythm (SaO2 mesor).The horizontal axis denotes the population percentage exhibiting each factor, while the vertical axis signifies disease risk.Distinct lines represent each factor, facilitating easy differentiation.The graph annotates the points corresponding to each factor under varying population percentages.The aggregate points, summarizing the collective impact of all aforementioned factors, correlate directly with disease risk.This predictive tool aims to furnish clinics with an accurate and intuitive assessment instrument, enhancing the efficacy of disease risk evaluation for patients.

Table 1 .
Baseline characteristics of patients with and without conversion to status epilepticus.All values are presented as medians (Q1, Q3); A p value below 0.05 denotes statistical significance; Abbreviations: RBC, Red blood cell count; WBC, White blood cell count; Monos, Monocyte percentage; MCH, Mean corpuscular hemoglobin; Hgb, Hemoglobin; Hct, Hematocrit; Lymph, Lymphocyte percentage; PT, Prothrombin time; CNS infection, Central nervous system infection; COPD, Chronic obstructive pulmonary disease.

Table 2 .
Comparison of average cosinor analysis of circadian rhythms in physiological parameters between groups with and without conversion to status epilepticus.
This table displays the results from the cosinor analysis of circadian rhythms in physiological parameters such as temperature, heart rate, respiration, and others, comparing groups with and without conversion to status epilepticus.