Design

All six included trials (Cottenceau 2011; Francony 2008; Harutjunyan 2005; Jagannatha 2016; Kumar 2019; Patil 2019), were conducted as parallel, RCTs. Five involved two arms and one (Patil 2019), three arms.

Sample sizes

Sample sizes tended to be small, ranging from 20 (Francony 2008), to 120 (Patil 2019). Data from 287 participants overall are included within this review, but as described below, 9% of these participants did not have diagnoses of traumatic brain injury. At protocol stage, we had estimated that 474 people were required to have a 90% chance of detecting, as significant at the 5% level, a decrease in death from 27% in the control group to 15% in the experimental group (Lu 2005); this target was manifestly not met within this review.

Only one trial ‐ the smallest included within the review (Francony 2008, n = 20), which did not collect data beyond 120 minutes of infusion, reported undertaking a power calculation, using the following assumptions: “The study population size for the trial was calculated assuming a 40% ± 15% ICP [intracranial pressure] in the HSS group and a 20% ± 15% ICP reduction in the mannitol group, according to previous studies ….Based on the formula for a normal distribution and assuming a two‐sided type I error of .05 and a power of .80, ten patients in each of the two groups were required” (Francony 2008, p 796).

No other trial reported undertaking a sample size calculation a priori, and all other trials (with the exception of Patil 2019), describe their small sizes as limitations. The authors of Jagannatha 2016 (n = 38), and Kumar 2019 (n = 30), both explicitly reported small trial size underneath the ‘limitations’ sections of their papers; likewise, investigators involved in Cottenceau 2011 (n =47), comment that “although the number of included patients was significantly larger than the number of patients included in most other comparable studies, the figures remained smaller than what would be needed to draw definite conclusions, especially when differences observed between subgroups are analyzed” (Cottenceau 2011, p. 210). Authors of Harutjunyan 2005 (n =32) also mention “the small patient population of each group” as a limitation, especially in the context of the “heterogeneity in the underlying neurological illness” (p R530).

Setting

All trials took place in ICUs, usually based within university hospitals. Three trials were conducted in India (Jagannatha 2016; Kumar 2019; Patil 2019), one trial each in France (Francony 2008), and Germany (Harutjunyan 2005), whilst the sixth trial included participants from both France and Israel (Cottenceau 2011). Trials were published between 2005 and 2019; the earliest recruitment period appears to have commenced in 2002 (Francony 2008), and the latest in 2015 (Patil 2019).

Participants

Inclusion and exclusion criteria varied amongst trials. Two were restricted solely to participants with traumatic brain injury severe enough to require intracranial pressure monitoring (Cottenceau 2011; Jagannatha 2016); a third (Francony 2008), required that they were stable patients (with or without traumatic brain injury) with sustained elevated intracranial pressure of 20 mm Hg for 10 minutes, not related to procedural pain (resulting in a participant group of whom 85% (17/20) had severe traumatic brain injury and the remainder of whom had suffered strokes). The fourth included trial required that participants have severe brain damage (Glasgow Coma Scale (GCS) < 8) with cerebral oedema, resulting in a participant group of whom only 31% (10) had a diagnosis of traumatic brain injury (the remainder had subarachnoid haemorrhage (9); brain infarctions (7); intracerebral haemorrhage (4) or “other” (2) (Harutjunyan 2005)). The inclusion criteria for Patil 2019 were that participants be screened by CT “to eliminate the need for surgery, then included if they had a GCS ≤8, and had sustained elevated ICP of >20 mm Hg for more than 5 minutes”. The sole paediatric trial (Kumar 2019), required that participants have severe traumatic brain injury, defined as a score of ≤8 on the Pediatric GCS, and present within 24 hours of trauma.

The one trial that focused on children under 16 years (Kumar 2019), did not report an overall mean for age, but provided ranges of ages. Participants were very young; 18 out of 30 were under five years of age, and one child was 22 months old. Within the trials largely focused on adults, minimum age criteria ranged from 15 (Jagannatha 2016), to 16 (Cottenceau 2011), to 18 years (Francony 2008; Harutjunyan 2005); the mean ages of included participants varied from around 30 (Jagannatha 2016) to 47 years (Harutjunyan 2005). In the four trials that reported on gender, male participants constituted the majority.

In terms of severity, participants ranged from means of GCS scores on admission of 7 (SD 2) to 8 (SD 2) in the two groups within the Francony 2008 trial (n = 20), to the most severely affected participants within the review, those within the Jagannatha 2016 trial (n = 38; median GCS scores = 4 in the hypertonic saline group (range 4 to 5) and 5 in the mannitol group (range 4 to 6). Authors of the two ‘mixed’ trials, which did not focus on traumatic brain injury, did not report the source of traumatic brain injuries (Francony 2008; Harutjunyan 2005); authors of the traumatic brain injury‐only trials (Jagannatha 2016; Cottenceau 2011; Kumar 2019; Patil 2019), did. Where reported, road traffic accidents accounted for the majority of traumatic brain injuries, followed by falls, assaults and ‘other’. Three trials explicitly categorised the nature of lesions, for example, extradural haematoma, subdural haematoma, contusions, intraventricular haemorrhage, diffuse axonal injury (Cottenceau 2011; Jagannatha 2016; Kumar 2019).

Interventions

Outcomes

The length of follow‐up (from one hour post infusion, to six months) impacted on the range of outcomes chosen. Four of six trials assessed mortality; one at end of stay in ICU, one at discharge from hospital and two at six months. We cannot disaggregate mortality data from the overall neurological outcome (GOS categories) in two trials (Cottenceau 2011;Kumar 2019).

All trials measured intracranial pressure, although in different ways and by different means.

Neurological outcome, even when assessed by a single instrument (GOS) used within three of six trials, did not make comparison straightforward. At six months, two trials reported GOS results dichotomised in the traditional way, grouping death, persistent vegetative state and severe disability together as a ‘poor’ outcome (Cottenceau 2011; Jagannatha 2016), a third trial dichotomised a poor outcome as death or persistent vegetative state, versus all other outcomes (Kumar 2019). Patil 2019 was short term in nature and so measured only the GCS at baseline and one hour after a successful infusion.

No trial appeared systematically to assess adverse effects.

Each trial is described in more detail in the Characteristics of included studies table and a separate table listing data reported by each trial at baseline and within the trial is also supplied ( Table 1).

Studies excluded for design features

We excluded 16 trials in total due to features of their design. We excluded five trials due to their cross‐over design (Battison 2005; Bourdeaux 2011; Huang 2014; Polushin 2009; Sakellaridis 2011). Searches of the Chinese literature identified 11 trials, which had to be excluded due to unreliable data or methods (incorrect ‘T’ values, questionable sequence generation), which we could not clarify through contact with trial authors (Huang 2015; Jin 2018; Li 2018; Liang 2013; Liu 2018; Mei 2016; Ni 2018; Shu 2015; Zhang 2014; Zhang 2015; Zhang 2018).

Termination

We excluded four registered trials because they were terminated, largely due to difficulties in recruitment (NCT01028339; NCT01108744; NCT01111682; NCT01215019.

Miscellaneous (chiefly ineligible intervention or comparator)

We excluded nine trials for other reasons. We excluded two trials (Jiang 2018; Yang 2019), because they compared differing doses of hypertonic saline, with no eligible control; one trial (Jafari 2018), because they administered hypertonic saline to both groups, with furesimide as an added treatment; and one trial (Du 2017) because there was no defined trigger for starting hyperosmotherapy to reduce ICP, and in addition, mannitol was administrated Q8h (every 8 hours). Following email communication with the trial author (Ichai 2013), we excluded Ichai 2009 because they used sodium lactate with the aim of decreasing the raised intracranial pressure; although sodium lactate is a hyperosmolar solution, its effects cannot be attributed to a classical osmotic effect. It differs fundamentally from sodium chloride because lactate is a metabolisable anion; this means that even with comparable osmolarity in vitro, sodium lactate becomes two times less hypertonic than equiosmotic sodium chloride in the body. We excluded one trial (Hong 2017), because of ineligible population. They administered hypertonic saline but not as a result of intracranial pressure monitoring. One trial appeared to meet the inclusion criteria, but was a nontherapeutic investigation that only considered coagulation (Wang 2017). We excluded one trial (Upadhyay 2010), for multiple reasons: quasi‐randomisation; too low a proportion of participants with traumatic brain injury (only 7% traumatic brain injury). We excluded one ongoing trial (Roquilly 2017), because its comparator, ‘standard care’, is unlikely to include alternative intracranial pressure‐lowering agents.

One trial (Vialet 2003). awaits classification (see also Characteristics of studies awaiting classification) due to unreconciled discrepancies in numbers of participants reported within the paper.

Random sequence generation

All included trials reported acceptable methods of generating randomisation sequence (either computer‐generated random‐number tables, or sealed, opaque envelopes) and so we judged them to be at low risk of bias.

Allocation concealment

Three trials (Cottenceau 2011; Francony 2008; Patil 2019), concealed allocation by using sealed and opaque envelopes and we judged them to be at low risk of bias. Harutjunyan 2005 and Kumar 2019 did not describe the method of allocation concealment, and the assessment for both is thus ‘unclear’. The trial author of Jagannatha 2016 told us in a personal communication that, “we did not do allocation concealment”, and so we judged it to be at high risk of bias.

Participants

We judged all the trials to have a low overall risk of bias for blinding of participants; all of the people recruited to the trials had a brain injury severe enough to have a GCS score of 8 or lower, were sedated, and so were unaware of the treatment they were receiving.

Treating physicians

Overall, we judged the included trials to have a high risk of bias for blinding treating physicians. In Francony 2008, it was not possible to blind administration because the two treatments were of different volumes. Cottenceau 2011 explained in 2015 through correspondence that the team was well aware of the regimen at the acute phase. Harutjunyan 2005, Jagannatha 2016, Kumar 2019 and Patil 2019 did not describe blinding and it is unlikely to have been ensured.

Outcome assessors

We judged trials to have varying risk of bias in blinding outcome assessors. Cottenceau 2011 told us in a personal communication that, “as for the outcome, it was assessed in both centers by blinded medical staff during follow‐up visits or by phone calls issued by blinded personnel”. Kumar 2019 reported using an assessor unaware of treatment status. In Francony 2008, Harutjunyan 2005 and Jagannatha 2016, objective outcomes such as laboratory measures or death are unlikely to have been affected by problems of blinding. We judged the above trials at low risk of bias for blinding outcome assessors. We judged risk of bias for blinding outcome assessors in Patil 2019 as high risk, as they do not mention blinding and all outcomes were assessed in the short‐term.

Mortality

Three out of six included trials reported on our primary outcome (mortality); timings varied and meta‐analysis was feasible between only two trials.

Poor outcome (measured by the dichotomised Glasgow Outcome Scale)

At protocol stage, we planned to report this outcome in the conventional manner, that is, by converting GOS scores into a dichotomous outcome where ‘death or disability’ signifies death, persistent vegetative state and severe disability; and a ‘good outcome’ includes moderate disability and good recovery. Two trials reported data suitable for pooling for this outcome at six months (Cottenceau 2011; Jagannatha 2016), although the uncertainty due to missing data from Jagannatha 2016 referred to above remains a concern. We thus reported findings in a similar way (available data, ‘best‐case’ and ‘worst‐case’).

Within Cottenceau 2011: 17 of 22 (77.3%) people in the hypertonic saline group and 14 of 25 (56%) people in the mannitol group died or had a severe disability at the end of the follow‐up period. In Jagannatha 2016, trial authors report that, “GOS scores at 6 months, dichotomized as unfavourable (GOS 1–3) and favorable (GOS 4–5)” are regarded as “comparable (p = 0.21)” (Jagannatha 2016, p. 71). Twelve of 18 (67%) people in the hypertonic saline group and 16 of 20 (80%) people in the mannitol group died or had a severe disability at six months. These findings are difficult to interpret in the light of the missing data on mortality referred to above. Pooling produces a result in which we have low confidence due to missing data as well as acknowledged baseline imbalance in Cottenceau 2011 (RR 1.09, 95% CI 0.82 to 1.44; I² = 67%; 2 trials, 85 participants; Analysis 1.3). We have also presented best‐ and worst‐case scenarios in Figure 5.

Kumar 2019 included solely children aged under 16 years. They also reported GOS, but dichotomised differently, grouping all children who survived (regardless of severity of disability) against those who died or remained in a persistent vegetative state. We could not, therefore, pool findings with those above. Results suggest no clear difference between groups (RR 0.72, 95% CI 0.14 to 3.61; 1 trial, 25 participants; Analysis 1.4).

Uncontrolled intracranial pressure

Only two trials (Francony 2008; Harutjunyan 2005) reported ‘uncontrolled ICP’ and they each had a different definition of ‘uncontrolled ICP’. In addition, the unit of analysis of Harutjunyan 2005 is ‘episodes’, while Francony 2008 is ‘participants’, so we could not incorporate them in a meta‐analysis.

All six trials reported the effect of HTS or mannitol on ICP. Three trials (Francony 2008; Jagannatha 2016; Kumar 2019) reported the mean magnitude of ICP reduction, with standard deviations. The other three trials (Cottenceau 2011; Harutjunyan 2005; Patil 2019) reported the initial ICP prior to, and immediately following, the administration by infusion of the study medication , in the form of means and ranges. We could not pool data for this outcome due to manifest heterogeneity in timings and modes of assessment, and other issues (including non‐normally distributed data, differing populations (adults and children, non‐TBI participants)). Therefore, we report results for this outcome narratively.

Cottenceau 2011: absolute ICP measurement values were reported as 12.2 ± 6.1 mmHg and 13.9 ± 7.8 mmHg in the hypertonic saline group versus 10.5 ± 6.8 mmHg and 13.6 ± 7.5 mmHg in the mannitol group after 30 minutes and 120 minutes of the infusion, respectively. The mean magnitude of intracranial pressure reduction after 30 minutes and 120 minutes of the infusion in the hypertonic saline groups were 1.70 mmHg higher (1.99 lower to 5.39 higher) and 0.30 mmHg higher (4.09 lower to 4.69 higher), respectively. Trial authors reported that, “both HTS [hypertonic saline] and MTL [mannitol] effectively and equally reduced ICP [intracranial pressure] levels with subsequent elevation of CPP [cerebral perfusion pressure] and CBF [cerebral blood flow], although this effect was significantly stronger and of longer duration after HTS … Further, effect of HTS on ICP appeared to be more robust in patients with diffuse brain injury” (Cottenceau 2011, p 2003).

Francony 2008: intracranial pressure was reported as being reduced by 45% ± 19% of baseline values (‐14 ± 8 mmHg) and by 32% ± 12% of baseline values (‐10 ± 4 mmHg ) in the mannitol group versus 35% ± 14% (‐10 ± 5 mmHg) and by 23% ±10% (‐6 ± 3 mmHg) in the HTS group at 60 minutes and at 120 minutes after the start of the study medication infusion, respectively). Only one person from the hypertonic saline group was a low responder to osmotherapy (with a reduction in intracranial pressure of < 20% of baseline values at 60 minutes after the start of infusion). Trial authors found both interventions to be effective for this outcome but added pretreatment factors need to be considered (e.g. serum sodium and haemodynamics).

Harutjunyan 2005: a total of 53 episodes of raised intracranial pressure exceeding 20 mmHg from 15 people in the mannitol group required infusion of trial medication. For four of these episodes (7.5%), intracranial pressure was uncontrolled within an average of 8.7 (4.2 to 19.9) minutes. 57 episodes of increased intracranial pressure occurred in the 17 people in the hypertonic saline group. For two of these episodes (3.5%), intracranial pressure was uncontrolled within 6.0 (1.2 to 15.0) minutes. Trial authors in this trial concluded hypertonic saline to be more effective for increased intracranial pressure than mannitol but cautioned that the benefit in this trial might be explained by “local osmotic effects” as there had been no differences in baseline haemodynamics.

Jagannatha 2016: trial authors here report that mean fall in intracranial pressure following a dose of hyperosmolar agent was 8.9 ± 8.4 mmHg in the mannitol group and 10.1 ± 8.7 mmHg in the hypertonic saline group, based on a comparison of 488 episodes of raised intracranial pressure across six days. The mean fall in intracranial pressure following a dose of hyperosmolar agent in the hypertonic saline group was 1.20 mmHg higher (‐0.37 lower to 2.77 higher) than with mannitol. Trial authors reported that this was significant and also that the percentage time for which intracranial pressure remained below a threshold of 20 mmHg on day 6 was significantly higher for hypertonic saline than mannitol. They also reported that cerebrospinal fluid drainage was performed on 41 ± 38 occasions in the mannitol group and 45 ± 31 occasions in the HTS group (p= 0.73).

Kumar 2019: trial authors report that most episodes of raised intracranial pressure were first managed by cerebrospinal fluid drainage which was effective in “almost more than two thirds of episodes”. Thereafter, if not managed and requiring administration of a hyperosmolar agent, the mean (SD) reduction in intracranial pressure was −5.67 (SD 3.9) in the hypertonic saline group; − 7.13 (SD 2.9) in the mannitol group (Kumar 2019, p 1003). Trial authors report that the difference was not statistically different.

Patil 2019: all findings in this trial relate to a single episode of raised intracranial pressure, after which assessment was carried on for a maximum of one hour (assessments were made at 10, 30 and 60 minutes). Investigators reported data in the form of means and ranges (rather than standard deviations); they then report findings in percentage improvements. All three hyperosmolar agents (hypertonic saline, mannitol and mannitol plus glycerol) were reported as effective, but hypertonic saline was slightly superior, effecting a greater change in reducing intracranial pressure, and more quickly, than other agents, while at a lower dose. The “maximum change in ICP [intracranial pressure] occurred after the bolus dose of 3% HTS [hypertonic saline]”; the “maximum decrease in ICP was produced by 3% HTS (60%), followed by the 10% mannitol plus 10% glycerol combination group (57%) and then 20% mannitol (55%). When the 3 groups were compared, 3% HTS required the lowest dose, that is, 1.4 mL/kg, followed by the 10% mannitol plus 10% glycerol combination group, that is, 1.7 mL/kg, and then the 20% mannitol group, that is, 2 mL/kg.” The time required to reduce intracranial pressure below 15 mm Hg was 16 minutes (range 6 to 39 minutes) in the 3% hypertonic saline group, 23 minutes in the mannitol group (range, 10 to 70 minutes) and 19 minutes in the 10% mannitol plus 10% glycerol combination group (range, 7 to 50) minutes.

A rebound phenomenon during treatment

None of the trials reported on this outcome systematically, although it is mentioned in passing as potentially affecting those treated with mannitol in one trial (Jagannatha 2016).

Pulmonary oedema during treatment

None of the trials reported data on pulmonary oedema during treatment.

Acute renal failure during treatment

None of the trials reported data on acute renal failure during treatment.