Predictive models for occurrence of expansive intracranial hematomas and surgical evacuation outcomes in traumatic brain injury patients in Uganda: A prospective cohort study

Background: Hematoma expansion is a common manifestation of acute intracranial hemorrhage (ICH) which is associated with poor outcomes and functional status. Objective We determined the prevalence of expansive intracranial hematomas (EIH) and assessed the predictive model for EIH occurrence and surgical evacuation outcomes in patients with traumatic brain injury (TBI) in Uganda. Methods We recruited adult patients with TBI with intracranial hematomas in a prospective cohort study. Data analysis using logistic regression to identify relevant risk factors, assess the interactions between variables, and developing a predictive model for EIH occurrence and surgical evacuation outcomes in TBI patients was performed. The predictive accuracies of these algorithms were compared using the area under the receiver operating characteristic curve (AUC). A p-values of < 0.05 at a 95% Confidence interval (CI) was considered significant. Results A total of 324 study participants with intracranial hemorrhage were followed up for 6 months after surgery. About 59.3% (192/324) had expansive intracranial hemorrhage. The study participants with expansive intracranial hemorrhage had poor quality of life at both 3 and 6-months with p < 0.010 respectively. Among the 5 machine learning algorithms, the random forest performed the best in predicting EIH in both the training cohort (AUC = 0.833) and the validation cohort (AUC = 0.734). The top five features in the random forest algorithm-based model were subdural hematoma, diffuse axonal injury, systolic and diastolic blood pressure, association between depressed fracture and subdural hematoma. Other models demonstrated good discrimination with AUC for intraoperative complication (0.675) and poor discrimination for mortality (0.366) after neurosurgical evacuation in TBI patients. Conclusion Expansive intracranial hemorrhage is common among patients with traumatic brain injury in Uganda. Early identification of patients with subdural hematoma, diffuse axonal injury, systolic and diastolic blood pressure, association between depressed fracture and subdural hematoma, were crucial in predicting EIH and intraoperative complications.


Introduction
Traumatic brain injury (TBI) is common around the world, burdening economies with substantial healthcare costs in treatment and long-term neuropsychological disability 1 .Globally, TBI affects at least 69 million individuals and accounts for 5 million deaths annually 2 .Of these, 93% occur in low-and middle-income countries (LMICs) [3][4][5] .The highest cases of TBI result from road tra c accidents (RTA) 4 .
In Uganda, mortality among TBI patients is 4.2 times higher than in the US 6,7 .TBI is a major cause of prolonged sequelae, death in the young population, and a social burden as well [8][9][10][11] .TBI encompasses numerous types of insults to the brain, including scalp injury, skull fractures, surface contusions, diffuse axonal injury (DAI), hypoxic-ischemic damage, meningitis, and intracranial hemorrhage 12 .Among the most serious consequences is expansive intracranial hematoma (EIH).EIH refers to evidence of increased hematoma volume of over 33% within the intracranial vault or absolute hematoma growth over 6mL from an initial scan with varying consequences 13 .Intracranial hemorrhages can be classi ed in three ways: (1) based on time, classi ed as hyperacute, acute, subacute, and chronic; (2) based on size, including small, big, and massive; and (3) based on location, classi ed as epidural, subdural, intracerebral, intraventricular and subarachnoid 14 .
Currently, there is no effective treatment strategy to prevent EIH development in patients with TBI [15][16][17][18][19] .There is no clear effective therapy for EIH, thus also affecting the time of interventions making them experience varying prognoses, which at times can be poor 20,21 .Although there have been advances in quick and accurate diagnostic imaging procedures, the related morbidity and mortality following EIH are persistently high 13 .EIH is a critical predictor of poor prognosis in TBI patients, with varying incidences of progression of hematoma between 13 to 38% 22 .It is necessary to assess the risk of EIH in patients with intracranial hematoma (IH) as early as possible during hospitalization to avoid unnecessary morbidity and poor prognosis.The subdural hematoma (SDH), skull fracture, subarachnoid hemorrage (SAH) have been identi ed as risk factors for EIH among IH patients using logistic regression 12,[23][24][25][26] .Some studies propose different prognostic models, for example, the ICH score, to assess which patients would bene t from surgical treatment and which from a conservative one.A problem they present is that they do not take into account the evolutionary nature of intracranial hematomas 14 .
The prospect of modeling has gained attention and excitement because of its advantages in processing large and high-dimensional medical data.It might offer a practical tool to enhance resource allocation in remote settings with limited resources.Accurate predictive models for EIH occurrence and unfavorable surgical evacuation outcomes could be derived from early biological markers of injury severity.Currently, no studies have investigated the accuracy of modeling algorithms in predicting EIH and surgical evacuation outcomes in patients with IH in Uganda.Therefore, this study was conducted to identify relevant risk factors, assess the interactions between variables, and establish a predictive model for EIH occurrence and neurosurgical outcomes in TBI patients at MNRH.

Study design
A prospective cohort study conducted between the June 2021 and June 2023 to predict EIH occurrence and surgical evacuation outcomes in TBI patients.

Study settings
The study was conducted at Mulago National Referral Hospital (MNRH), Kampala, Uganda.MNRH, the largest public hospital in Uganda located around 5 kilometers from the city center.It serves about 75% of injured victims in Kampala and surrounding districts 27 .The Accident and Emergency department attends to approximately 64 urgent surgical evacuations of intracerebral hemorrhage in patients with acute TBI at MNRH every month 28 .TBI patients were recruited on admission at the Accident and Emergency Department, then followed up in the operative theatres and the neurosurgery department.

Study population
All TBI patients aged 18 years or older with a radiologically con rmed brain computerized tomography scan that showed an intracranial bleed were eligible to participate in the study.Purposive sampling was used to recruit participants.

Study participant selection criteria
Participants were TBI patients aged 18 years and above, post resuscitation GCS of 4 to14, with evidence of EIH on two CT scans (increase in hematoma volume > 33% or absolute hematoma growth > 6ml from the initial scan) exclusively, eligible for cranial surgery and enrolled in the study within 24 hours of initial presentation to hospital.A written signed informed consent from the patient or their next of kin was obtained.Patients with (1) known pre-thrombocytopenia, (2) a history of coagulation disorders, and (3) used anticoagulants, (4) pregnancy, (5) and those with the inability to consent before surgical intervention were excluded from the study.EIH was de ned by an increase in acute intracranial traumatic hematoma volume > 33% or absolute hematoma growth > 6ml from the initial scan within 72 hours of injury during the study period 29

Study procedures and variables
Patients were recruited from their admission at the Accident and Emergency Department, followed up in the operative theatres, postoperatively in the neurosurgery ward, and neurosurgical outpatient clinics for up to 6 months for the occurrence of complications.Trained research assistants, using a Research Electronic Data Capture (REDCap) system, collected pertinent demographic, clinical, laboratory, and radiological information.Patient demographic characteristics recorded included age, gender, occupation, residence type, geographic regions, and matrimonial state.Outcome's data were obtained at enrollment (baseline) and then at 1 day, 30-, 90-, and 180-days post-discharge.Clinical outcomes were recorded during these outpatient visits.The principal outcomes included death from any cause and complications (clinically apparent spasticity, cerebrospinal uid leakage, bleeding diastasis, infection, and pneumonia).
Other parameters of concern included type of cranial surgery (burr holes, craniotomy, craniectomy), evolution of intracranial hematomas over time (within 24, > 24 but < 48, and over 72 hours), timing to surgery (early, late and delayed), and site of surgery (frontal, fronto-parietal, fronto-parieto-temporal, parieto-temporal, occipito-parietal, temporal, parietal, occipital, interhemispheric, or posterior fossa).Early surgical evacuation occurs when an expansive mass lesion is removed early (within 24 hours) after an acute TBI 30 .Late surgical evacuation occurs when the expansive mass lesion is removed later (within 25 to 72 hours) after an acute TBI 30,31 .Delayed surgical evacuation occurs when an expansive mass lesion is removed 72 hours after an acute TBI.

Statistical data analysis
Non-parametric continuous data were summarized using the median and interquartile range, whereas continuous parametric data were shown using the mean and standard deviation.Categorical variables were expressed as frequencies and percentages.To assess for the risk factors of expansive hematoma, the chi-square test was used at bivariate to assess for variables that had an independent association with the outcome.
We used logistic regression model and selected variables and those of clinical signi cance by a stepwise backwards selection based on model AICs.The explanatory variables were categorized into demographical features, neurological assessment, baseline laboratory characteristics, baseline blood pressure characteristics, and neuroimaging patterns.Only the rst orders of the variables were included at this step.Then, implemented the model selection again by including all the interaction terms of the selected variables in the previous step.The model selection procedure is the same and based on model AICs.Eventually, we pooled all the selected variables together and manually picked the most relevant variables.All the variables and their interaction terms were included in this model which was later under the same screening process as previous models.The study evaluated the model performance using AUC, accuracy, and ve-fold cross-validation.
The study adapted this metric-based and manual mixed screening procedure, because of the large number of candidate variables collected in the data.The metric-based screening was to eliminate statistically irrelevant factors from the data, while the manual screening was based on neurological knowledge and the purpose of having an interpretable model.It was likely that some variables were missed, but we believe that this procedure was both feasible and suitable.
The data were analyzed using regression models that t the response type: logistic for binary, ordinal logistic for ordinal variable, and linear for continuous.Variables with p-values < 0.05 remained in the model and were considered to be signi cant risk factors for intra and post-operative complications, including death among adult patients with expansive hematomas following traumatic brain injury.Then, variables that "best" t the outcome (i.e., EIH) were selected by choosing the model with the lowest AIC during a backward stepwise selection.Finally, all selected variables were pooled from each category to t the nal logistic regression model.The nal model was selected based on the same criteria, but the predictive performance on the validation dataset was undesirable.Therefore, the model was manually tweaked to achieve better accuracy, AUC in both validation dataset and ve-fold cross-validation.Levels of p < 0.05 were considered statistically signi cant.

Study Flow of the Participants
During the study period, a total of 1500 patients were screened (Fig. 1).Out of these, 21.6% (n = 324) were enrolled into the study.In a cohort of 324 patients with intracranial hematomas, 59.3% (n = 192) had expansive intracranial hematomas indenti ed on CT scan (Fig. 1).Of the 324 patients with intracranial hematomas, 192 (59.3%) had expansive hematomas identi ed on CT scan resulting in a proportion of 0.59 (95% CI: 0.54 to 0.65).

Demographic characteristics
In this study, a majority of the participants with EIH were signi cantly older than the no EIH group (42.3 ± 17.9 vs. 30.5 ± 14.0), most of the patients were male 261 (80.6%) and majority 152 (46.9%) were motorcyclists known as boda riders.The results show that most of the patients 184 (56.8%) were from rural residence (Table 1).

Postoperative clinical Outcomes
The respective inpatient postoperative risks were 10.2% for death, 58.0% for intraoperative complications, 15.7% for early posttraumatic seizures (PTS), 11.7% for coma, 11.7% for brain oedema, 4.9% for wound infection, 4% for nutrition de cit, 3.4% for pneumonia, 2.8% for CSF leakage, 1.2% for subdural hygroma (Table 3).However, 282(88.4)alive until the end of the study.Among these patients, 208 (73.8%) were discharged with favourable condition (de ned by GOS of > 3), 74 (26.2%) were discharged with unfavourable condition (de ned by GOS of < 3)..9months of follow-up.The probability of survival decreased as the follow-up time increased, especially within the rst months of follow-up according to the Kaplan-Meier survival curve for time to death for TBI patients with intracranial hematomas after surgery (Fig. 3).

Models that can predict intraoperative complications after neurosurgical evacuation in traumatic expansive intracranial hematoma patients
Holding the rest of variables constant, undergoing craniectomy would make the odds of getting intraoperative complication become 8.5317969 times (p = 0.03578).Holding the rest of variables constant, undergoing craniotomy would make the odds of getting intraoperative complications become 1.5244901 times (p = 0.56229).Holding the rest of variables constant, having delay surgery (over 72 hours from the accident) would make the odds of getting intraoperative complications become 2.8757974 times (p = 0.00229).Holding the rest of variables constant, having early surgery (within 24 from the accident) would make the odds of getting intraoperative complications become 1.6229086 times (p = 0.27600).Holding the rest of variables constant, undergoing frontoparietal decompression would make the odds of getting intraoperative complications become 3.6784446 times (p = 0.00531).
Holding the rest of variables constant, undergoing fronto-parieto-temporal decompression would make the odds of getting intraoperative complications become 13.329105 times (p = 0.00027).Holding the rest of variables constant, undergoing parieto-temporal decompression would make the odds of getting intraoperative complications become 2.3774298 times (p = 0.04745).Holding the rest of variables constant, undergoing occipito-parietal decompression would make the odds of getting intraoperative complications become 5.2172998 times (p = 0.06236) (Table 4) and (Fig. 4).

Models to predict mortality after neurosurgical evacuation among traumatic expansive intracranial hematoma patients
Holding the rest of variables constant, undergoing craniotomy would make the odds of dying become 10.094291 times (p = 0.000623).Holding the rest of variables constant, undergoing craniotomy would make the odds of dying become 10.094291 times (p = 0.000623).Holding the rest of variables constant, undergoing decompressive craniectomy would make the odds of dying become 1.9833354 times (p = 0.421221) (Table 5) and (Fig. 5).Health related quality of life of intracranial hematomas after TBI who undergoing surgical evacuation increases overtime.Health-related quality of life of intracranial hematomas after TBI, assessed 6 months post-surgical evacuation, was higher for patients with no EIH (Quality of Life after Brain Injury score: 88.0(8.6)vs 85.0(9.5),P = 0.010) (Table 6) and (Fig. 6).Holding the rest of variables constant, undergoing craniotomy would make the odds of having unfavorable quality of life become 94.632408 times (p = 0.000642).Holding the rest of variables constant, undergoing decompressive craniectomy would make the odds of having unfavorable quality of life become 2.2630195 times (p = 0.639813) (Table 7) and (Fig. 7).

Models to predict unfavorable functional outcomes after neurosurgical evacuation in traumatic expansive intracranial hematoma patients
Holding the rest of constant, undergoing decompressive craniectomy would make the odds of having unfavorable functional outcomes become 2.2412587 times (p = 0.023181).Holding the rest of variables constant, undergoing craniotomy would make the odds of having unfavorable functional outcome become 1.0187434 times (p = 0.000824) (Table 8).

Discussion
This study assessed the burden, identify relevant risk factors, the interactions between variables, and establish a predictive model for EIH occurrence and surgical evacuation outcomes among TBI patients at Mulago National Referral Hospital.The application of these models can assist in decision-making for individual TBI patients with risk of EIH.Several predictive models for intracranial hematomas have been described by regulatory bodies, most of which focus on the quality of hospitalization 32 .The evolutionary character of cerebral bleeding and effect of timing from injury (as observed in practice and time of presentation at the unit) to cranial surgery on outcomes are not taken into account by the majority of models 33 .Furthermore, the models were established on small samples, many were methodologically awed, and few were validated in external populations 33 .Few others were neither presented in a clinically practical way, nor were they established in populations in LMIC, where 93% of TBI occur 34 .
Furthermore, patients and surgeons frequently are faced with di cult decisions making in regard to the management of EIH.Weighing in patient characteristics when several options (e.g., age, history of toxic substances, heathy ASA state, imaging ndings, evolution of hematoma over time, blood pressure parameters, type of surgery, conservative, golden hour, site of surgery) are available is currently done in an arbitrary way.The development of EIH-speci c models would be the cornerstone of de ning quality in surgical health care delivery 32 .

Models that can predict expansive intracranial hematomas occurrence
Considering the feature importance in our developed random forest algorithm-based model, the top ve features were age, systolic blood pressure, diastolic blood pressure, subdural hematoma (SDH), diffuse axonal injury (DAI), skull fracture, and an interaction between skull fracture and SDH and were found to be indepently associated with EIH in the nal model (Table 2).These ndings concur with previous studies which reported that SDH, skull fracture are risk factors to EIH 12,[23][24][25][26] .Some studies have linked the evolution of intracranial hematomas with factors such as age greater than 61 23,24,[35][36][37] and elevated admission systolic blood pressure 23,38,39 .This study did not nd subarachnoid hemorrage(SAH) among risk factors as reported by Allison et al. 24 .The average area under the receiver curve (AUC) from a vefold cross-validation was 0.833, while the average accuracy was 73.4%.The AUC of the current study seven-point models disagree with models developed by Cepata et al (0.72) 40 , but higher than a simple four-point predictive score reported by Allison et al.(0.77) 24 .

Models that can predict intraoperative complications of EIH in TBI patients
Undergoing craniectomy, having a delayed surgery over 72 hours, undergoing frontoparietal, frontoparietotemporal and parietotemporal decompression were found to be independently associated with intraoperative complications in the nal model (Table 4).Holding the rest of variables constant, undergoing craniotomy would make the odds of getting intraoperative complications become 1.5244901 times (p = 0.562) while undergoing decompressive craniectomy would make the odds of getting intraoperative complications become 8.5317969 times (p = 0.03578).The fact that patients run the danger of a sizable number of complications after DC and that it may further degrade quality of life is undeniable, even if one were to argue that the technique only reduces mortality at the expense of increasing the proportion of the seriously crippled 41 .The Monro Kellie doctrine's space restrictions are intended to be circumvented through decompressive craniectomy (DC), which disturbs the cerebral blood and CSF ow dynamics.The timing of the resulting di culties, which occur days to months following the operation, can be predicted to help in managing them 41 .This nding is in line with a study conducted by Hawryluk et al which con rmed that DC remains a controversy 30 .Choosing to perform a DC is still challenging and the overall bene ts should be balanced against the outcomes and complications on a case by-case basis.DC causes serious complications including meningitis, subdural hygroma, hydrocephalus and increased reoperation rate.Morbidity related to the surgical evacuation was also examined.Intraoperative complications occurred in 58.0% of our patients; the high intraoperative complications con rmed substantial challenges in surgical intervention in LMIC which may impact on the long-term survival and outcomes.Early posttraumatic seizures (PTS) were the most common complication and was observed in 15.7% of patients.PTS are a serious consequence among TBI patients with EIH globally and most especially in low developed countries like Uganda.PTS are commonly associated with severe TBI and the reported incidences vary greatly 42 .In the present study, coma and brain oedema were observed in 11.7% among patients.Furthermore, wound infection (4.9%) and pneumonia (3.4%), CSF leakage (0.3%) were reported in the analysis of 324 patients (Table 4).These results differ from a case series conducted in India where infection (9.1%) and CSF leakage (9.1%) were most reported 43 .

Models to predict intraoperative mortality after neurosurgical evacuation traumatic expansive intracranial hematoma patients
The overall mortality rate in this study was 10.2%, which is lower than the 20% overall mortality rate seen in a prospective research in China on patients who underwent early versus late craniectomy after a traumatic brain injury 44 .The mortality rate observed in this study agrees with the ndings obtained from a study conducted in Uganda (9.6%) 6 .Undergoing craniotomy was found to be independently associated with intraoperative mortality in the nal model (Table 5).Holding the rest of variables constant, undergoing craniotomy would make the odds of dying become 10.094291 times (p = 0.000623) while undergoing decompressive craniectomy would make the odds of dying become 1.9833354 times (p = 0.421221) (Table 7).Even if DC was associated with high odds of getting intraoperative complications, however, it was also linked with reduced mortality in this study.These ndings were supported by DECRA and RESCUEicp results which reported that the reduced mortality rate and higher rates of complication as a result of DC [45][46][47] .This procedure has been demonstrated to reduce ICP and to minimize days in the ICU.A large frontotemporoparietal DC (not less than 12 x 15 cm or 15 cm diameter) is recommended over a small frontotemporoparietal DC for reduced mortality and improved neurologic outcomes in patients with severe TBI 48,49 .
Models to 18 months unfavorable quality of life after neurosurgical evacuation among traumatic expansive hematoma patients Undergoing craniotomy and having delayed EIH progression were found to be independently associated with 18 months unfavorable quality of life after neurosurgical evacuation in the nal model (Table 7).
Holding the rest of variables constant, undergoing craniotomy would make the odds of having unfavorable quality of life become 94.632408 times (p = 0.000642) while undergoing decompressive craniectomy would make the odds of having unfavorable quality of life become 2.2630195 times (p = 0.639813).(Table 7).Health related quality of life of intracranial hematomas after TBI who undergoing surgical evacuation increased overtime.Health-related quality of life of intracranial hematomas after TBI, assessed 6 months post-surgical evacuation, was higher for patients without EIH (Quality of Life after Brain Injury score: 88.0(8.6)vs 85.0(9.5),P = 0.010) (Table 7).This nding agrees with previous study which demonstrated that EIH are the most dangerous form of intracranial hematoma that occurs following traumatic brain injury (TBI) 38,50 .In addition, EIH appears to be associated with a high rate of neurological deterioration in patients with TBI 20,21 .What is not contested is that EIH patients face the risk of a large number of complications after the operation and that can further compromise their quality of life.
Models to predict unfavorable functional outcomes neurosurgical evacuation in traumatic expansive intracranial hematoma patients However, 88.4% (282) patients survived from intracranial hematoma TBI.Among these patients, 73.8% (208) were discharged with favourable condition (de ned by GOS of > 3) while 26.2% (74) were discharged with unfavourable condition (de ned by GOS of < 3).These ndings are better when compared with a study conducted at MNRH which reported 71.7% patients were alive at the end of the study 51 .Undergoing craniotomy, craniectomy and having delayed EIH progression were found to be independently associated with 18 months unfavorable functional outcomes after neurosurgical evacuation in the nal model (Table 8).Holding the rest of variables constant, undergoing decompressive craniectomy would make the odds of having unfavorable functional outcomes become 2.2412587 times (p = 0.023181) while undergoing craniotomy would make the odds of having unfavorable functional outcome become 1.0187434 times (p = 0.000824).(Table 8).This result is in line with previous reports where DC has been established to reduce mortality only at the expense of increasing the proportion of the severely disabled 49 .
The present study had limitations including, small sample size, more variables of 824 than the sample size of the study participants and missing values.This was expected and the research team increased 10% of possible dropout or missing data.Thus, the model selection was done to achieve better interpretability of the ndings.The model used a 70% and 30% split for the train and test data.A seed was set so that the ndings were reproducible.However, the ndings show that the selected models were subject to changes when the seed changes, meaning a person who uses a different con guration for the 70%, 30% is likely to get different models in the positive direction.And these ndings could be due to the small sample size used in the study.In addition, there were not enough samples to t a large model with all these factors and their interaction and then do model selection.

Conclusion
We developed 5 proposed models including ( Figure 1

Table 1
Baseline demographic characteristics of patients with EIH In the nal model, age, systolic blood pressure, diastolic blood pressure, subdural hematoma (SDH), diffuse axonal injury (DAI), skull fracture, and an interactive term of skull fracture and SDH were selected.Odds of having EIH increased to 1.045 times for one increment in systolic blood pressure.Odds decreased to 0.942 times for one increment in diastolic blood pressure.Having SDH increased the odds of EIH to 6.286 times, while diffuse axonal injury increased it to 4.024 times.Given a patient has skull fracture, SDH decreased the odds of EIH to 0.0676 times (Table2).Key codes : 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1----Statistical signi cance at 95% CI Figure2has two curves, one for training data, one for testing data.Model demonstrated good discrimination with areas under curve for EIH occurrence (0.833) while the average accuracy was 73.4%.

Table 3
Baseline complications occurrence for adult traumatic brain injury patients with intracranial hematomas

Table 4
Model coe cients for intraoperative complications occurrence following neurosurgical evacuation among adult TBI with EIH

Table 7
Model coe cients for unfavorable quality of life following neurosurgical evacuation among adult TBI with EIH

Table 8
Model coe cients for unfavorable functional outcomes after neurosurgical evacuation in adult TBI with EIH ) model for unfavorable functional outcomes which have 15 coe cients each to predict earlier individualized estimates of EIH occurrence and surgical evacuation outcomes based on preoperative conditions at MNRH and similar settings across the region.These proposed models have demonstrated good discrimination with AUC for EIH occurrence (0.833) while the average accuracy was 73.4%, intraoperative complications (0.67), mortality (0.36), and 18 months unfavorable quality of life after neurosurgical evacuation in TBI patients (0.61).: Expansive intracranial hematoma, GCS: Glasgow coma scale, GOS: Glasgow outcome scale, HIC: High-income countries, LMIC: Low-and middle-income countries, Midline shift <1cm, MDS: Midline shift ≥1 cm, MNRH: Mulago National Referral Hospital; OBC: Open basal cistern, PR: Prevalence risk; QoLIBRI: quality of life after brain Injury; RCT: randomized clinical trials, TBI: Traumatic Brain Injury; TEH: Traumatic expansive hematoma, SAH: Subarachnoid hemorrhage, SDH: Subdural hematoma, SK: Skull fracture, MAP: Mean arterial pressure 1) model with 8 coe cients for EIH occurrence; (2) model for intraoperative complications; (3) model for death occurrence; (4) model for unfavorable quality of life and (5EIHFigures Page 26/32