This monocentric cohort study was conducted over 10 years, between January 2012 and December 2022, in the trauma critical care unit of Lapeyronie University Hospital (Level-I Regional Trauma Centre – Montpellier, France – OcciTRAUMA network). All trauma patients suspected to have sustained severe trauma by a prehospital medical team were directly admitted to the emergency room of this unit, following the French guidelines for prehospital medical triage [13]. Retrospective identification of the study population was allowed by the trauma registry of our institution. For this retrospective analysis, patients consent was not considered as necessary by our local ethical committee.

The present study included all patients experiencing IH following a TBI. ICP was defined as elevated when its value exceeded 22 mmHg for at least 5 min with no stimulus, despite the optimization of sedation, correction of secondary brain insults and requiring additional specific therapy to control it.

Minors, patients with incomplete medical records, and those who prematurely died in the 48 h following admission or underwent an early decompressive craniectomy were excluded. Patients who had EVD were also excluded.

During the study period, all TBI patients benefited from standardized management according to international guidelines [14]. All patients undergoing ICP monitoring with an intraparenchymal probe were sedated and mechanically ventilated. Maintaining normocapnia and normoxia was a central goal during their initial management. Insulin was administered to control serum glucose levels between 7 and 10 mmol/L. Patients were positioned supine with a 30-degree head-up tilt. Cerebral perfusion pressure (CPP) was maintained between 60 and 70 mmHg according to guidelines through vasoactive support with norepinephrine and, if necessary, plasma volume expansion with crystalloids [15]. The CPP could be adjusted via regular dynamic autoregulation assessments if necessary. For all patients, a cerebral CT scan was systematically performed on admission and 48–72 h after the injury to monitor lesion progression. Additional CT scans could be performed more closely based on clinical evolution or in cases of high risk of haemorrhage or complications.

ELD was introduced in 2012 as a rescue procedure for IH. It was thus used to reduce ICP to physiological levels when all conventional therapeutic procedures failed and in the absence of CT scan contraindications (i.e., no discernible basal cisterns, midline shift > 10mm, presence of tonsillar herniation or significant mass lesion). Timing of ELD insertion was decided according to a local standardized protocol described in the Additional file 1. Since January 2018, a unit protocol was established to determine ELD indications, promote its use, and standardize the timing of its placement. This protocol was based on a CT scan classification according to several criteria, such as the amount of sub- and supratentorial CSF reserves, the presence of extra-axial haematomas or haemorrhagic contusions, and the obliteration of the basal cisterns. This classification integrated seven grades to allow the identification of the benefit-risk ratio for CSF drainage via the lumbar route. Homogeneous or inhomogeneous IH mechanisms and the potential for CSF drainage determined the possibility of ELD and its recommended timing (Fig. 1). To ensure the safety of procedure and to exclude a potential contraindication, a CT scan performed within the last 12 h and a coagulation status were necessary.

Classification of intracranial CSF reserve. ELD: External lumbar drainage, CSF: Cerebrospinal fluid, CT: Computed tomography, ICP: Intracranial pressure, IH: Intracranial hypertension

To avoid risks of cerebral hypotension and downward herniation caused by an excessive initial CSF withdrawal, our protocol recommended several safety measures associated with ELD placement. Firstly, ELD should be performed in the lateral position to limit pressure gradient between the spinal and cranial subarachnoid spaces. Secondly, ELD catheter should be inserted quickly into the needle after obtaining CSF reflux to reduce the drained volume. Thirdly, once the patient was repositioned with head elevation, a careful CSF withdrawal was achieved when the patient was repositioned in supine position, around 1 ml/minute, until ICP reaches the target level. Fourthly, a continuous bilateral pupillary examination was performed during initial CSF subtraction and then every hour. Thereafter, nurse protocol to control CSF drainage during critical phase was applied once the catheter was connected to a sterile collection bag allowing to collect, to measure and to set the pressure gradient of drainage. The drainage system was initially positioned 15 cm above the tragus to maintain safe and continuous drainage and was subsequently adjusted depending on the ICP targets. The target range for ICP was typically set between 10 and 20 mmHg. If the ICP remained above 20 mmHg, the drainage pressure gradient was increased by lowering the drainage level by -5 cm H2O. Conversely, if the ICP dropped below 10 mmHg, the drainage was immediately stopped. When ICP increased to 15 mmHg, the drainage system was opened again and pressure gradient was reduced by increasing the drainage level by + 5 cm H2O. In case of pupillary changes, the drainage was immediately stopped, and a brain CT scan was quickly performed. The lumbar CSF flow and pressure were monitored by nurses hourly to avoid the risk of overdrainage. If the CSF drainage rate was more than 10 ml per hour, the level of drainage was increased by + 5 cm H2O. Weaning from ELD could be envisaged when the ICP remained within normal values for at least 12 h continuously with a + 20 cm H2O drainage level. A clamping test for 24 h was then attempted and a CT scan performed to confirm the absence of ventricular dilatation. ELD could be removed in the case of good clinical evolution.

The primary demographic data, Glasgow Coma Scale (GCS) on admission, cranial CT scan findings, and initial treatments, were documented for each patient. The Abbreviated Injury Scale (AIS) was calculated for each anatomical area, including the head, face, thorax, abdomen, extremities, and skin. All surgical interventions were collected. The ICP status was also recorded before and after implementing lumbar drainage. All CT scans were analyzed retrospectively for the present study by a radiologist blinded to the clinical outcome. All admission CT scan were analyzed to determine the Marshall score. Moreover, patients were classified according to our classification of the intracranial CSF reserve presented in Fig. 1. Seven grades are described, considering the presence of an intracranial injury, as well as the volumes of the basal cisterns and the lateral ventricles.

The occurrence and duration of IH phenomena were defined. The delay and duration of the ELD procedure were obtained in medical reports, as well as the initial volume of CSF drained. Any complications associated with lumbar drainage, such as infections, catheter occlusion, pupillary status, and cerebral complications, were also collected based on nurse ICU sheets and medical reports. All interventions to reduce the ICP, including measures to control body temperature, osmotherapy, and barbiturate administration, were also documented. Additionally, the duration of mechanical ventilation, sedation, and the length of stay in the ICU were collected. Finally, the neurological recovery of all patients was determined at ICU discharge and 6 months after the trauma, using the Glasgow Outcome Scale (GOS). This assessment was obtained from medical records.

The threshold used for the treatment of intracranial hypertension was 22 mmHg as recommended by the latest edition of Brain Trauma Foundation [14].

TBI severity was classified according to the GCS: mild TBIs had a GCS score of 14–15, moderate TBIs a GCS score of 9–13, and sTBI a GCS score of 3–8.

The critical CSF reserve represented the grade of our intracranial CSF reserve classification at the onset of IH.

The neurological outcomes were categorized based on the GOS [16]. Patients with GOS 4–5 were considered as having a “good outcome”, while those with GOS 1–3 were classified as having an “unfavourable outcome”.

The studied patients were initially divided into two groups, ELD and no ELD. Their clinical characteristics were initially presented and compared. Quantitative data were expressed as means (standard deviation [SD]) or median (interquartile range [IQR]) and compared using Student or Mann–Whitney U tests. Qualitative data were expressed as numbers (percentages) and compared using chi-square or Fisher tests.

Since the ELD and no ELD groups were not randomized and cannot be considered comparable, a propensity score was determined for each patient to account for potential confounding risk factors and reduce the expected selection bias. A logit score was thus established by multivariable logistic regression predicting ELD placement. This logistic model included the following variables determined a priori: age, diabetes insipidus, initial Marshall score, critical CSF reserve, osmotherapy, worst GCS, injury severity score (ISS), Simplified Acute Physiology Score (SAPS II), Revised Trauma Score (RTS), chest AIS, abdominal AIS, head AIS, face AIS, extremities AIS, skin AIS, mean arterial pressure, presence of tension pneumothorax on admission, or initial transfusion volume. This propensity score was compared between two groups as a quantitative variable and was also used as an adjustment variable in multivariable analysis to predict IH control and outcome.

Kaplan–Meier analysis was used to compare the cumulative incidence of normalization of ICP between the ELD and no ELD groups to study the delay in reaching IH control. Considering that early death was a possible clinical outcome and informative censoring, the Fine and Grey method was applied, classifying all dead patients as patients with persistent IH. The comparison between the groups was performed using the Grey test, which incorporates this competitive risk. Additionally, Cox proportional hazard regression analysis was conducted to assess whether an ELD was associated with faster normalization of ICP. The crude hazard ratio (HR) of ELD was provided with its 95% confidence interval (CI). Multivariable adjustment was then performed for potentially confusing risk factors determined a priori: age, initial Marshall score, initial and critical CSF reserve, GSC, ISS score, Simplified Acute Physiology Score, Revised Trauma Score, Trauma Injury Severity Score, chest/abdominal/extremities AIS score, mean arterial pressure, pneumothorax, initial transfusion, ventilation before admission, and the propensity score.

The influence of ELD on the final neurological recovery was ultimately assessed using a multivariable logistic regression analysis to predict the 6-month GOS (good outcome). Odds ratios (ORs) for ELD were thus provided with their 95% CI. Similarly to the previous analysis, multivariable adjustment was conducted using the same potentially confusing risk factors determined a priori, including the propensity score.

All statistical analyses were conducted using the software SAS online. A p-value below 0.05 was considered statistically significant.