Effect of Continuous Hypertonic Saline Infusion on Clinical Outcomes in Patients with Traumatic Brain Injury: A Retrospective Causal Analysis

Purpose: Intracranial pressure (ICP) control has long been recognized as an important requirement for traumatic brain injury (TBI) patients. Nevertheless, the long-term effect of hypertonic saline (HTS) remains unknown. The aim of this study was to elucidate the effect on clinical outcomes in TBI patients admitted to intensive care unit (ICU) settings. Methods: We retrospectively identied moderate to severe TBI patients from two public databases named Medical Information Mart for the Intensive Care (MIMIC)-IV and eICU Collaborative Research Database (eICU-CRD). A marginal structural Cox model (MSCM) was used, with time-dependent variates designed to reect exposure over time during the ICU stay. A trajectory modeling, based on intracranial pressure evolution pattern, allowed identication of subgroups. Results: Overall, in our cohort of 1955 eligible patients, 130 (6.65%) received HTS. MSCM indicated that the HTS was signicantly associated with better Glasgow coma score [(GCS): hazard ratio (HR) 1.19, 95% condence interval (95% CI) 1.01-1.40, p = 0.041], higher infection complications [eg. urinary tract infection (HR 1.88, 95% CI 1.26-2.81, p = 0.002)], and increased ICU LOS (HR 2.02, 95% CI 1.71-2.40, p < 0.001). A protective effect from GCS by the HTS was found in the subgroup with medium and low ICP. Conclusion: Our study revealed no signicant difference in the all-cause mortality rates between patients receiving HTS or not. Increased occurrence rates of infection and electrolyte imbalance were inevitable outcomes caused by continuous HTS infusion. Although the study suggested the slight benecial effects including better neurological outcome, the results warrant further validation.


Introduction
Traumatic brain injury (TBI), a severe condition observed in trauma patients, is estimated to jeopardize 69 million individuals in 2019 worldwide, possessing a high mortality rate [1]. And it is reported that nearly 33% of patients with TBI die in hospital [2]. Thanks to prompt surgical and medical management, the mortality has steadily reduced in the past few decades, however, the incomplete recovery still contributes to TBI survivors with varying degrees of neurological disability, speci cally, adverse sequelae and changes in quality of life [3][4][5].
Intracranial pressure (ICP) represents a major predictor of neurological deterioration in patients with TBI, with elevated ICP being associated with poor neurological outcome [6]. Hypertonic saline (HTS) is regarded as the mainstay for treatment of elevated ICP in TBI. HTS agent might exert an early effect on ICP by decreasing blood viscosity and hematocrit, improving cerebral blood ow (CBF), and oxygen delivery, which in turn reduce cerebral blood volume (CBV) and ICP [7]. Thus, there may be a role for HTS administration to avoid progressive cerebral edema, subsequent neurological deterioration and secondary cognitive disorder.
However, there are concerns that HTS use may trigger acute kidney injury (AKI), hypernatremia and hyperchloremia, which, further, were independently associated with in-hospital mortality in patients with moderate-severe TBI (msTBI) [8,9]. And the fourth edition of the Brain Trauma Foundation's Guidelines states that there is insu cient evidence to support the application of any speci c hyperosmolar drug for patients with msTBI [10]. Other guidelines do not elucidate the direct effect on clinical outcome associated with the use of HTS or just are based on low-quality evidence [11,12]. Furthermore, the predominance of literature focused on severe TBI, whereas milder forms of TBI remain unclear.
In particular, the role of HTS administration in patients with msTBI, de ned as Glasgow Coma Score (GCS) < 12 remains uncertain, with physiologic studies showing bene t and clinical studies suggesting harm [13,14]. Although one retrospective study incorporated HTS as the time-dependent variable by conventional Cox analysis, its credibility was limited by residual confounding, inappropriate accounting for dynamic interaction between time-varying treatment and confounders [15].
It is noted that these publications were limited by the single hospital sampling, limited variables, insu cient statistic power, and a lack of effective control for time-varying confounders. In fact, the treatment with HTS, a long-term dynamic variable that changes over time, depends on sodium and chloride. However, there are concerns that performing a randomized controlled trial (RCT) may not be ethical, thus, we emulated an analysis of a hypothetical trial through the use of observational longitudinal data to achieve causal inference. Thus, this present article aimed to investigate the causal effect of HTS on the clinical outcomes in TBI patients from two large intensive care unit (ICU) databases. Additionally, with an aim to explore the heterogeneity of the data, a trajectory modeling based on longitudinal/dynamic ICP evolution pattern during ICU stay was achieved.

Participant selection
Inclusion criteria were patients with msTBI. People with an age of less than 16 years old, ICU stays less than 48 h, osmotherapy use prior to ICU admission, cervical spinal cord injury, and bolus injection were excluded from the study. Moreover, for patients with ICU admissions more than once, only data of the rst ICU admission of the rst hospitalization were included in the analysis.

Data collection
In this study, the data were extracted from MIMIC-IV and eICU-CRD including age, gender, race, weight, body mass index (BMI), admission type, smoking history. Coexisting disorders were also collected based on the recorded International Classi cation of Diseases (ICD)-9 and ICD-10 codes. Then, the Charlson comorbidity index (CCI) was calculated from its component variables. Lastly, we extracted data containing laboratory parameters, injury details, multiple scoring systems, medication use, ICU interventions, and neurosurgical interventions on the rst day of ICU admission. Laboratory variables of HTS, sodium and chloride were measured during the entire ICU stay. For patients with multiple measurements, the minimum daily value of serum sodium and chloride was included in the analysis owing to the fact that it was related to the greatest severity of illness.

Primary and secondary outcomes
The primary outcomes were all-cause mortality, GCS value on the day of discharge. Secondary outcomes incorporated AKI, infection complications [Urinary tract infection (UTI), pneumonia, sepsis], electrolyte imbalance (de ned as hyperchloremia, hypernatremia), ICU length of stay (LOS) and hospital LOS. Then, a trajectory modelling based on ICP evolution pattern during ICU stay was established.

Statistical analysis
Values were presented as the means with standard deviations (if normal) or medians with interquartile ranges (IQR) (if non-normal) for continuous variables, and total numbers with percentages for categorical variables. Proportions were compared using χ² test or Fisher exact tests while continuous variables were compared using the t test or Wilcoxon rank sum test, as appropriate.

Marginal structure Cox model (MSCM)
In this study, HTS, a time-dependent variable, was dichotomized as "any dose of HTS exposure versus none" on a daily basis. The daily use of HTS was predicted by both time-xed covariates and timevarying confounders including sodium and chloride to facilitate casual interference. In the rst step, the probability of receiving HTS at each follow-up day was estimated by both time-xed and time-varying covariates, and then inverse probability of treatment weighting (IPTW) was calculated. In the second step, covariates were balanced across the population to attempt to emulate RCT at the time of the HTS exposure period [19]. Then, a causal estimate of the treatment's effect on the study outcome was estimated. Details of MSCM can be seen at electronic supplemental material (ESM).

Grouped based trajectory modeling
Group-based trajectory modeling (GBTM) is an established analytical technique used to identify hourly ICP clusters following a similar progression of changes over time on a given variable [20]. Details on GBTM are also reported in the ESM.

Sensitivity Analysis
To increase the robustness of our ndings, pre speci ed subgroup analyses strati ed by age, gender, BMI, ICP, blood sodium level, blood chloride level, severity of injury, CCI and injury severity score (ISS) were performed. Moreover, the patterns assumed missing to be completely at random, so multiple imputation approach was used to iterate the original data (ESM table S1) [21]. Ultimately, unmeasured confounding may bias the estimates from this observation study, thus, the E-value was also computed to further evaluate the robustness of the ndings [22,23]. The E-value indicated how strongly an unmeasured confounder would have to be associated with both the HTS use and the outcomes of interest to reduce the observed effect to the null, conditional on the measured covariates.
Statistical signi cance was considered to be at two-sided p < 0.05. All analyses were performed with R

Trajectory modeling analysis
According to the trajectory modeling, ve subgroups of patients with distinct ICP evolution pattern during their ICU stay were identi ed (Fig. 3). The ICP patterns of the different subgroups can be seen as follows: With regard to the percentage of HTS infusion, the difference between the ve subgroups was not statistically signi cant. In contrast, there were fewer death events and better neurological outcomes in the subgroup with the low and medium ICP level. Owing to the fact that 92% patients were in the group 2 and 3, thus, the further subgroup analyses were conducted in these two subgroups.

Discussion
To the best of our knowledge, this is the rst, multicenter, longitudinal study indicating the effect of continuous HTS infusion on in-hospital outcome among msTBI patients while addressing simultaneously the time-varying nature of this type of exposure. We found that the utility of HTS was not independently associated with mortality, but with better neurological outcomes, increased infection complications, DVT, hyperchloremia, hypernatremia, LOS, and sedation duration. By characterizing the ICP trajectories of patients, we also identi ed that HTS was associated with better neurological outcome in the medium and low ICP subgroups.
Of note, cerebral edema develops from several pathologic mechanisms following TBI, leading to cerebral herniation and a rapid worsening of prognosis [24,25]. Nevertheless, the availability of neurosurgical interventions in most medical facilities is limited, means of inducing a hyperosmolar environment including HTS are currently proposed [26]. Accordingly, the effects of HTS on TBI patients need to be further explored.
As expected, in this passage, both hypernatremia and hyperchloremia are both causes and results of HTS. Similarly, previous studies have also concluded that if not properly controlled, continuous infusion of HTS may bring hypernatremia and hyperchloremia, accompanied by the increased in-hospital mortality [27,28].
Moreover, in this study, continuous HTS infusion did signi cantly improve neurological outcome as assessed by the GCS value on the day of discharge. Relatively more severe TBI (moderate: 747 vs severe: 1208) patients have increased the power to indicate the impact of HTS owing to the fact that the risk of intracranial hypertension or brain edema is higher in this population [29]. The precise regulatory mechanism remains to be further elucidated. Perhaps, the e cacy of HS in brain edema resulting from TBI was closely associated with the downregulation of aquaporin-4 (AQP4), the restoration of brain blood barrier (BBB) integrity and the suppression of in ammatory factors including Interleukin (IL-1β), tumor necrosis factor (TNF-α), NF-κB [30].
Furthermore, our study provided evidence that continuous HTS infusion was not signi cantly associated with mortality. It is worth noting that our passage was able to elucidate that the mortality described was related to an underlying medical condition itself, not other confounders, including hypernatremia or hyperchloremia. Congruently, a 2021COBI RCT published in JAMA found that there was no signi cant difference in 6-month mortality between the HTS group and control group [31]. Likewise, Tan SK et al [15] found that HTS was not associated with hospital mortality in patients with severe TBI. Yet one systematic review concluded that HTS was associated with a reduction of in-ICU mortality [32]. Given the heterogeneity of the included population, we planned a subgroup analysis to account for this possibility. Speci cally, restricting the analysis to the subgroup of obesity patients did demonstrate a higher mortality. In this regard, studies conducted by Brown CV et al, Chabok SY et al had similar results to ours [33,34]. Further, this adverse effect seems to be due to age, lower admission blood pressure, and more associated chest injury, rather than a direct result of the obese state, however, this speculation needs further validation [33].
In addition, our ndings added additional evidence to previous studies suggesting that continuous HTS infusion did not result in increased AKI, suggesting no harm to the kidney [35,36]. An important factor may be that, in ICU, patients were kept in euvolemia or mild hypervolemia, despite supraphysiologic serum sodium and serum osmolarity.
Nevertheless, a concerning nding was the association between continuous HTS infusion and increased infection and LOS. Previous clinical trials concerning continuous HTS infusion and infection were limited with mixed results; some reported increased infection [35] while others did not [37,38]. Physiologically, high sodium levels have demonstrated suppressive effects on leukocyte activation and could theoretically impair the immune system, resulting in higher infection rates [39]. As hypothesized, previous studies have suggested that increased LOS may be due to multiple complications, especially, hypernatremia and infection. Further studies might be needed to con rm the above assumptions.
To further explore the association of HTS on in-hospital outcomes and provide an insight into the mechanisms by which ICP level produces this effect, we characterized the ICP trajectories and estimated the impact of ICP burden on outcomes. Indeed, the heterogeneity of the included population in terms of ICP evolution pattern was evidenced. Although case mixes were different from one subgroup to the others, a protective effect was found in the low and medium ICP subgroups, which was consistent with the result of MSCM, adding the robustness to our ndings.
The strength of our study lied in a population-based longitudinal cohort from multi-center in US, a highquality data with granular temporal detail, a homogeneous population, accordingly, ensuring the robustness, reliability, generalizability of the ndings. Apart from this, to estimate the causal effect of HTS administration on in-hospital outcome, we employed MSCM, which helped align results from an observational study with those of actual randomized controlled trials. Moreover, by implementing multiple imputation to missing data, the analysis gave a relatively robust conclusion to reduce the estimation bias and improve validity. Unmeasured confounders were also treated by E-Value. When compelling randomized trials are not yet forthcoming, it is incrementally valuable that credible evidence for the effect of the HTS in TBI patients was supplied in our study.
This study had several limitations, consistent with those inherent to many large administrative database studies. First, based on electronic records of routine clinical practice, missing data and outliers were common. Second, we did not have data on functional outcomes after discharge, which was arguably an important indicator in this study. Second, as a retrospective study, unmeasured confounding was inevitable. Thus, to quantify the potential implications of it, E-Value sensitivity analysis was used, which demonstrated that the results obtained would be different only if we had omitted the major covariates, which was often not the case in our study. Third, instead of outcome adjudication, our outcomes were de ned by ICD-9 and ICD-10 diagnosis codes. Incorrect codes or misclassi cation bias inevitably exist, so the codes we used were all veri ed by previous articles [40,41].

Conclusions
In this cohort study which enabled a less bias estimation by causal inference methods, we found that HTS was not associated with reduced mortality but with better neurological outcome, accompanied by increased infection, and LOS in msTBI patients. Despite careful methodology involving MSM and trajectory modeling were employed, well-designed, prospective, multicenter simulation studies or RCTs are needed to further clarify the impact of HTS on clinical outcomes.
Abbreviations TBI traumatic brain injury; ICP:intracranial pressure; HTS:hypertonic saline; CBF:cerebral blood ow; CBV:cerebral blood volume; AKI:acute kidney injury; msTBI:moderate-severe traumatic brain injury; Flowchart of eligible participants. ICU intensive care unit, TBI traumatic brain injury, HTS hypertonic saline Weights distribution plot for the inverse probability weights that were used to adjust for confounding. ICU intensive care unit