The effect of conscious disorder on the 30-day mortality of sepsis

Background Conscious impairment is related to worse prognosis in patients with sepsis. we aimed to evaluate the impact of conscious disorder on the 30-day mortality of patients with sepsis and identify the remediable risk factors for it. Methods Data were retrospectively retrieved from a prospective intensive care database. Conscious disorder was assessed by the Glasgow Coma Score (GCS). The impact of conscious dysfunction on the 30-day mortality of sepsis and its potentially modiable risk factors were evaluated using Cox proportional hazard regression and logistic regression, respectively.


Introduction
Sepsis-associated encephalopathy (SAE) is de ned as diffused brain dysfunction occurring during the process of sepsis without evidence of an intracranial infection [1]. Patients with sepsis complicated by encephalopathy often suffer from increased short-term mortality [2], and long-term symptoms, such as memory de cits, learning disabilities, and depressive-like symptoms, which can then seriously lower patients' quality of daily life [3][4][5]. Thus, exploring the recti able risk factors contributing to the development of brain dysfunction during sepsis, and then eliminating them is of great signi cance under the condition of treatment absence. Two landmark studies focused on this eld are from Eidelman LA et al [6], and Sonneville R et al [7]. In the study of Eidelman LA et al, they demonstrated that hepatic and renal dysfunction as well as bacteremia were associated with an increased incidence of SAE. Sonneville R et al found that acute renal failure, staphylococcus aureus infection, and common metabolic disturbances are potentially modifable factors contributing to SAE. However, these two studies had limitations. In the study of Eidelman LA et al, they contained only 50 patients with sepsis diagnosed by Sepsis 1.0 [8], and among them, only 31 patients were diagnosed as SAE. The relatively old de nition of sepsis and low sample size may not well represents the characteristics of patients with sepsis and SAE. In the study of Sonneville R et al, they de ned SAE as GCS < 15 and/or delirium during sepsis, but fail to exclude etiological factors other than sepsis that may cause impaired consciousness [9], which may lead to higher numbers of SAE cases and then bias the analysis of risk factors.
Although several factors are considered to potentially contribute to the process of SAE [10][11][12][13][14][15][16][17], the de nite pathogenesis is still unclear, making the diagnosis of SAE an excluding process by recognizing brain dysfunction rst and then excluding the most common etiological factors causing it [18][19][20]. However, as brain dysfunction in the diagnosis of SAE contains several pathological changes (mainly include disorder of consciousness, delirium, and the changes of ultrastructure or electrical activity without clinical manifestations), it is impossible for one retrospective study to include the whole SAE cohort due to the limitation of databases (none of the available intensive care databases collect patients' data on all of these pathological changes). Consequently, except for designing prospective study containing all of these pathological changes, another viable solution based on the existing clinical databases may be to conduct retrospective studies focusing on different asepcts of these pathological changes, and then to integrate the results of them.
Therefore, in this retrospective study by a large clinical database, we evaluate the effect of conscious disorder, one asepct of SAE, on the 30-day mortality of patients with sepsis and identify the remediable risk factors for it by using the newest de nition of sepsis and implementing a strict exclusion criteria. The results of the study may facilitate clinicians to make timely treatment decision and reduce the incidence of SAE.

Data source
We conducted an observational study by retrieving data from the Medical Information Mart for Intensive Care (MIMIC III) open source clinical database, which contains de-identi ed health-related data of over forty thousand patients who received treatment in critical care units of the Beth Israel Deaconess Medical Center between 2001.06 and 2012.10 [21,22]. The database is continuously updated and the newest version (MIMIC-III v1.4) was released on 2 September 2016, which enhanced data quality and provided a large amount of additional data. We use the MIMIC-III v1.4 in our study and all data in it was classi ed into 26 tables recording various individual information, such as demographics characteristics, treatment measures, nursing notes and laboratory tests. Besides, it contains survival outcome data obtained from the hospital and laboratory health record systems reporting the in-hospital mortality, or from the Social Security Administration Death Master File recording the out-of-hospital survival data. The MIMIC III database can be freely utilized after successful application and ethical approval from the Institutional Review Boards of both Beth Israel Deaconess Medical Center (Boston, MA, USA) and the Massachusetts Institute of Technology (Cambridge, MA, USA). Since all data are de-identi ed in this database to remove patient information, the requirement for individual patient consent is not indispensable.

Study population and data extraction
PgAdmin (version 4.1, Bedford, USA) was used to run structure query language (SQL) and then to extract data from the MIMIC III database. Seven tables were occupied in our study, including diagnoses_icd, icustays, chartevents, patients, labevents, microbiologyevents and prescriptions. We included adult patients (> 17 years of age) with a diagnosis of sepsis de ned by Third International Consensus De nitions for Sepsis (Sepsis 3.0): (1) Patients with infection con rmed by the positive results of microbial cultivation, and (2) the Sequential Organ Failure Assessment (SOFA) score ≥ 2 [23]. Excluded were patients (1) with primary brain injury (traumatic brain injury, ischemic stroke, hemorrhagic stroke, epilepsy or intracranial infection); (2) with pre-existing liver or kidney failure affecting consciousness; (3) with severe burn and trauma; (4) receiving cardiac resuscitation recently; (5) with chronic alcohol or drug abuse; (6) with hypertensive encephalopathy; (7) with severe electrolyte imbalances or blood glucose disturbances, including hyponatremia (< 120 mmol/l), hyperglycemia (> 180 mg/dl) or hypoglycemia (< 54 mg/dl); (8) dying or leaving within 24 hours since ICU admission; (9) without an evaluation of GCS. Eligible patients were included in the nal cohort for investigation (Additional le 1: Figure S1).For the nal cohort, we retrospectively collected the following data from the database: (1) demographic data including age, gender and ethnicity; (2) 30-day outcome; (3) comorbidity conditions as coded and de ned by the International Classi cation of Diseases, Ninth Revision (ICD-9) code; (4) mean value of vital signs during the rst 24 hours of ICU stay; (5) the rst laboratory data since ICU admission; (6) site of infection and type of micro-organism; (6) Medical intervention, including vasopressor adoption, mechanical ventilation and the use of antibiotic, sedative and analgesic agents. The severity of illness and organ failure was assessed by modi ed forms of the simpli ed acute physiology score (SAPS II) and the sepsisrelated organ failure assessment (SOFA) on the rst day of ICU stay, respectively [24,25]. The modi ed forms are SAPS II and SOFA score excluding the component of central nervous system. Besides, each component of the SOFA score was extracted to assess the function of each organ and a score > 2 was de ned as acute organ failure [18].

Sepsis-associated impairment of consciousness
We de ned sepsis-associated impairment of consciousness in the study as sepsis accompanied by the Glasgow Coma Score (GCS) < 15 on the rst day of ICU admission. GCS was con rmed as an excellent tool for the assessment of conscious dysfunction [26], and for neurological evaluation of critically ill patients [27]. For sedated or postoperative patients, we adopted GCS measured before sedation or surgery.

Statistical analysis
The distribution of variables was assessed by Shapiro-Wilk tests. Normally distributed variables, non-normal variables and categorical data were expressed as mean ± standard deviation (SD), median [interquartile ranges] and number (percentages), respectively.
The study population was divided into two groups (survivors and victims) based on the 30-day outcome since ICU admission. Normally distributed variables were compared between survivors and victims by unpaired Student' t test and non-normal variables by Mann-Whitney U test. The Chi-Squared test was adopted to assess the differences in categorical variables between the two groups.
Cox proportional hazard regression was used to screen risk factors independently associated with the 30-day mortality of sepsis. Speci cally, univariate analysis was rstly conducted and variables with a p value < 0.1 were then entered into the multivariable analysis, where variables with p < 0.05 were picked out to calculate the change of hazard ratio (HR) from the univariate regression (uHR) to multivariate regression (mHR) [HR-change=(uHR-mHR)/uHR × 100%]. Then, variables with HR-change < 10% were considered as the independent risk factors for the 30-day mortality of sepsis. Besides, the rank-hazard plot was used to visualize the relative importance of the independent risk factors in the Cox proportional hazard model [28].
As GCS was considered as an independent risk factor in the Cox regression, we divided the study cohort into four groups based on the GCS score, that is group with (1)  where patients with GCS 9-12 and GCS 3-8 had signi cantly lower survival rate than patients with GCS 12. Then, the risk factors independently associated with GCS ≤ 12 were recognized by logistic regression by a process similar to Cox regression. Speci cally, univariate analysis was rstly conducted and variables with a p value < 0.1 were entered into the multivariable analysis, where variables with p < 0.05 were picked out to calculate the change of odds ratio (OR) from the univariate regression (uOR) to multivariate regression (mOR) [OR-change=(uOR-mOR)/uOR × 100%].
Then, variables with OR-change < 10% were considered as independent risk factors. Sensitivity analyses were conducted in the same cohorts after exclusion of patients with a history of neurological diseases (mainly those with neurodegenerative diseases and cognitive impairment).
Statistical analyses were performed using R software (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria). Missing values were addressed with multiple imputation in the process of regression analysis and survival analysis. The imputation technique involves creating multiple copies of the data and replacing missing values with imputed values through a suitable random sample from their predicted distribution. We used the "mice" package of R to obtain 5 imputation datasets. A two-tailed p value < 0.05 was considered statistically signi cant.

Characteristics of participants
There were a total of 16602 patients with infection in the database according to the results of microbial cultivation, and 7918 patients with a SOFA score≥2 were diagnosed as sepsis. After screening by the exclusion criteria, a total of 4446 patients were included into the nal cohort and conscious dysfunction (GCS<15) was observed in 2126 (47.8%) patients.
The baseline characteristics and the characteristics at ICU admission of all participants were exhibited in Table 1 and  Table 2. Patients who died (victims) within 30-day since ICU admission were older than those who survived (survivors), with a median age of 78 [67,86] in victims and 68 [55, 78] in survivors. Patients were predominantly male in both victims and survivors. Comorbidities showed that victims had higher proportion of cardiovascular diseases, chronic liver disease and coagulopathy. Organ failure and the state of illnesses were more severe in victims, with higher modi ed SOFA and SAPSII score, and higher frequency of coagulation, brain, cardiovascular and renal dysfunctions in victims than that in survivors. Victims had higher proportion of pulmonary and/or fungal infection than survivors. Besides, several vital signs and laboratory indexes were different between victims and survivors, with higher level of respiratory rate, lactate, AST, bilirubin, potassium, RDW and WBC, and lower level of arterial blood pressure, temperature, PO 2 and hemoglobin in victims, respectively.
Medical interventions are described in Additional le 2 Table S1. Victims had longer ICU stay time and shorter hospital stay time than survivors. The proportion of vasopressor adoption and mechanical ventilation were higher in victims than that in survivors. Moreover, victims had higher frequency use of carbapenems, vancomycin, benzodiazepines and analgesic drugs than survivors.  Table 3. One variable with a HR-change<10% demonstrated that the impact of it on the 30-day mortality of patients with sepsis was not signi cantly in uenced by the other variables in the multivariate Cox regression, and can be considered as the independent risk factor for the 30-day mortality of sepsis (Additional le 3: Figure S2). Then, the relative importance of independent risk factors were visualized by the rank-hazard plot (Figure 1). Results revealed that among the risk factors which were negatively associated with the 30-day mortality of sepsis, GCS score was the most important one with the highest relative hazard at a range of 3-13 except for that of Mean SpO 2 at the lower limit. Survival analysis of patients with sepsis according to the GCS score As GCS has been demonstrated to be an independent risk factor for the 30-day mortality of sepsis, Kaplan-Meier's survival analysis was then conducted to visualize the 30-day survival according to different GCS score of participants ( Figure 2). Results showed that the 30-day survival had no difference between septic patients with normal consciousness (GCS=15) and those with mild impairment of consciousness (GCS 13-14) (log-rank p=0.28), whereas septic patients with moderate impairment of consciousness (GCS 9-12) or coma (GCS 3-8) had higher 30-day mortality than both of them (log-rank p<0.0001). Besides, the 30-day survival had no difference between septic patients with moderate impairment of consciousness and those with coma (log-rank p=0.48). These results indicated that it was GCS 3-12 that contribute to the increased 30-day mortality of patients with sepsis and was the independent risk factor for it.
Risk factors for sepsis-associated moderate impairment of consciousness and coma The OR (95%CI) of variables with p value<0.05 in multivariate logistic regression were exhibited in Table 4. One variable with a OR-change<10% demonstrated that its effect on the incidence of GCS≤12 in patients with sepsis was not signi cantly in uenced by the other variables in the multivariate logistic regression, and can be considered as the independent risk factor contributing to the moderate impairment of consciousness and coma in sepsis (Figure 3). Sensitivity analysis was conducted in patients without a history of neurological diseases and the results were exhibited in Additional le 4: Figure S3. Then, we divided the continuous variables of the independent risk factors into three groups based on their reference range and the proportion of patients with GCS≤12 in each group were calculated and visualized in Figure 4.

Discussion
In this retrospective analysis by MIMIC III database, we found that mild disturbance of consciousness (GCS [13][14] was not an independent risk factor for the 30-day mortality of sepsis, while moderate disorder of consciousness (GCS 9-12) and coma (GCS 3-8) was independently associated with it. Then, the mean value of systolic pressure on the rst day of ICU admission, utility of cephalosporin, urinary infection, and the level of hemoglobin, neutrophil and serum sodium were identi ed to be the independent risk factors leading to the moderate impairment of consciousness and coma in patients with sepsis. After divided systolic pressure, hemoglobin, neutrophil and serum sodium into three groups based on their reference range, systolic pressure and/or serum sodium higher than the normal reference range (systolic pressure ≥ 140 mmHg, serum sodium ≥ 145 K/uL) was associated with increased incidence of moderate impairment of consciousness and coma, while the level of hemoglobin and neutrophil lower than the normal reference range (hemoglobin < 11 g/dL and neutrophil < 50% ) was related to decreased incidence of them. As reducing hemoglobin and neutrophil to a level lower than their normal reference range may bring adverse effect on patients, it is inappropriate to give intervention on both of them. Therefore, the risk factors that can be recti ed to reduce the incidence of moderate conscious impairment and coma include hypertension (systolic pressure ≥ 140), hyponatremia (serum sodium ≥ 145 K/uL), the utility of cephalosporin and urinary infection. To the best of our knowledge, this is the rst study to evaluate the effect of conscious impairment on the 30-day mortality of sepsis and explore the modi able risk factors contributing to sepsis-associated disorder of consciousness.
Hypertensive encephalopathy is a hypertensive emergency de ned by acute neurological dysfunction associated with an acute and severe increase in blood pressure (systolic pressure > 220 mmHg or diastolic pressure > 120 mmHg).
Patients suffering from hypertensive encephalopathy often have alterations in mental status, or accompanied by seizures, visual disturbance, or headache [29]. However, in our study, patients with hypertensive encephalopathy according to the ICD-9 code has been excluded from the study cohort, and the highest systolic pressure in the study cohort was 180 mmHg, less than the typically lower limit of systolic pressure (220 mmHg) that may cause hypertensive encephalopathy. All of these ruled out the possibility of hypertensive encephalopathy as the cause of conscious dysfunction in patients with sepsis accompanied by high systolic pressure. Indeed, studies have revealed that hypertension is an independent risk factor for delirium in ICU patients [30][31][32], in patients after cardiac or abdominal operation [33,34], and in patients at a general hospital [35]. As delirium is reported in 11-80% of patients with critically ill, and patients with delirium may clinically present as conscious disorder, we infer that the potential overlap of cohort between our study and the above ones may make hypertension an independent risk factor for the moderate impairment of consciousness and coma. Moreover, studies have demonstrated that hypertension is directly linked to memory impairment [36], diminished psychomotor speed [37] and inattention [38]. Therefore, if patients subject to intense physical or psychological stress (such as sepsis) in ICU, they may be more vulnerable to develop conscious impairment, but the exact mechanisms for the impairment of consciousness caused by hypertension is still unclear and need to be further studied.
The results from Sonneville R et al suggested that hypernatremia might play a role in the pathophysiology of SAE [7].
They included patients with either hypernatremia or hyponatremia in their study, but only hypernatremia was nally identi ed as an independent risk factor for SAE in the logistic regression analysis. Interestingly, in our study, although patients with hyponatremia were excluded from the study cohort, hypernatremia is still independently related to GCS ≤ 12 in the logistic regression analysis, further con rmed that hypernatremia might have negative effects on the consciousness of patients with sepsis and need to be avoided.
Previous studies have suggested that antibiotics may play a role in the pathophysiology of SAE [39], which was consistent with our conclusion that cephalosporin was connected with the moderate impairment of consciousness and coma. Cephalosporin has been widely regarded as a neurotoxic antibiotic [39]. Therefore, dose monitoring should be conducted during the systematic use of cephalosporin as the neurotoxic effect are related to the overdose of antibiotics [40]. However, the causal relationship between cephalosporin use and sepsis-associated conscious dysfunction remains to be demonstrated.
Urinary tract infection is widely regarded as a common medical comorbidity in patients with disorders of consciousness [41][42][43], however, its impact on spesis-associated impairment of consciousness was rarely reported. In the study of Sonneville R et al [7], they found that it was staphylococcus aureus, rather than infection source, that was associated with SAE, while Zhang et al draw inconsistent conclusions that except for staphylococcus aureus, patients with biliary tract infections and intestinal infections were more prone to SAE [18]. In our study, we found that urinary infection was independently linked to moderate impairment of consciousness and coma in septic patients.
Considering the similar exclusion criterion between the study of Zhang et al and ours, we tend to believe that infection site is a risk factor contributing to conscious dysfunction during sepsis. Therefore, clinical prevention of hospitalacquired urinary infection and active treatment towards urinary infection of various origins with appropriate antibiotics according to drug susceptibility results may be helpful to protect septic patients from the deterioration of consciousness and improve short-term prognosis Several strengths could be found in the present study. First, we use Sepsis 3.0 to de ne sepsis, which was rare in previous retrospective clinical studies. Second, the exclusion criteria of this study are relatively strict and comprehensive, which to the maximum extent ensure that conscious impairment is caused by sepsis instead of other pathogenic factors. Third, MIMIC III is a high-quality database including a large number of patients with sepsis of various origins. In this regard, our study may be the largest retrospective study focusing on the outcomes and risk factors of sepsis-associated disorder of consciousness. The large sample size of the study avoid over tting in Cox and logistic regression, and ensure the accuracy and reliability of the results.
This study has certain limitations: First, the retrospective nature of this observational study determined that unidenti ed confounding factors may affect the results of regression analysis. Second, we focused on sepsisassociated disorder of consciousness in this study, while the other components of SAE, including sepsis-associated delirium, sepsis-associated ultrastructure changes of brain and sepsis-associated electrical activity changes were not involved, further studies are essential to evaluate their effect on the prognosis of sepsis and explore their remediable risk factors if needed. Third, we did not assess the impact of conscious impairment on the long-term mortality of patients with sepsis. Fourth, the causal relationship between the identi ed risk factors and sepsis-associated disorder of consciousness can not be determined from this observational study.

Conclusions
In this retrospective analysis by MIMIC III database, we found that moderate impairment of consciousness (GCS 9-12) and coma (GCS 3-8) were independently related to the 30-day mortality of sepsis. Potentially remediable risk factors for patients with sepsis characterized by GCS ≤ 12 at ICU admission were hypertension (systolic pressure ≥ 140 mmHg), hypernatremia (serum sodium ≥ 145 K/uL), the utility of cephalosporin and urinary infection. Although these factors are likely to play a role in the pathophysiology of sepsis-associated disorder of consciousness, we can not draw the conclusion that medical interventions towards these remediable risk factors can surely reduce the incidence of moderate impairment of consciousness and coma. Further studies are needed to investigate the causal relationship between them. All data are de-identi ed in this database to remove patients' information, the requirement for individual patient consent is not indispensable.

Consent for publication
Not applicable Availability of data and materials The rank-hazard plot to visualize the relative importance of independent risk factors for the 30-day mortality of patients with sepsis.

Figure 2
The Kaplan-Meier's survival estimated of the 30-day survival probability of patients with sepsis. Patients were divided into 4 groups according to the GCS score, namely GCS=15 (normal consciousness), GCS 13-14 (mild impairment of consciousness), GCS 9-12 (moderate impairment of consciousness) and GCS 3-8 (coma). (a) The curve showed that septic patients with moderate impairment of consciousness or coma had higher 30-day mortality than those with normal consciousness or mild impairment of consciousness (log-rank p<0.0001). (b) the 30-day survival had no difference between septic patients with normal consciousness and those with mild impairment of consciousness (logrank p=0.28). (c) The 30-day survival had no difference between septic patients with moderate impairment of consciousness and those with coma (log-rank p=0.48).

Figure 3
The forest plot of the risk factors independently associated with sepsis-associated moderate impairment of consciousness and coma in the logistic regression.