In emergency situations, collecting accurate clinical information from trauma patients can be challenging, and accurate data are often difficult to obtain. Accurate assessment of the risk of progression to traumatic intracranial hemorrhage (ICH) is essential, especially for patients who are relatively stable or have minimal traumatic ICH compared with those deemed to require immediate emergency surgical intervention within the TBI cohort.13,14Furthermore, precise details regarding the mechanism of head injury often prove difficult to ascertain.15.
The aim of this study is to develop a model to predict the short-term prognosis of patients with traumatic brain injury. This model emphasizes the use of clear and readily available information obtained in the emergency department. Specifically, it relies on initial head CT scan data and physical examination findings, which are readily available and readily accessible in the emergency department.
Types of traumatic intracranial hemorrhage
Previous literature has examined the analysis of various traumatic ICH types.16Alvarez-Sabin et al. reported the phenomenon of delayed traumatic cerebral hemorrhage.17However, there is a lack of studies demonstrating that different ICH types affect the frequency of ICH progression. Furthermore, a systematic clinical analysis of the impact of each ICH type on patient outcomes has yet to be performed. The type of ICH characterized as petechiae in this study has also been referred to as “blooming” or showing a “salt-and-pepper appearance” in previous studies.16,18Pathologically, this phenotype exhibits severe symptoms of traumatic subarachnoid hemorrhage resulting from progressive microvascular rupture and subsequent hemorrhage extending into the brain parenchyma. We hypothesized that these pathological differences would result in the poorest prognosis of the PH type, which was confirmed by feature importance analysis of the XGboost model.
Fractured skull and injuries from counterattack
The clinical significance of rebound head injury, characterized by the occurrence of brain injury contralateral to the point of impact, has been suggested as a potential indicator of the severity of head trauma.19This view is based on the understanding that anti-Koop injuries carry a higher risk of complications, such as brain swelling and bleeding, compared with injuries that occur only at the site of impact (Koop injuries).9.
In this study, the incidence of countercoupe ICH in patients with occipital bone fractures was observed to be 17.9%, higher than that in patients with skull fractures at other sites (frontal bone fractures 3.7%, temporal bone fractures 7.2%, parietal bone fractures 3.7%). Thus, the frequency of frontal lobe ICH was significantly increased in patients with an initial impact on the occipital bone. This trend may be related to brain contusions occurring on structures such as the irregular surface of the anterior cranial fossa of the skull and the anterior clinoid process. This could explain the general association of frontal lobe countercoupe ICH with TBI involving occipital bone impact.9.
In this study, we successfully developed an algorithm that can predict individual prognosis using CT findings and clinical information. By integrating both clinical and radiological factors, such as countercoup injury and certain types of ICH, we were able to predict ICH progression in patients with mild to moderate traumatic brain injury (TBI) with high accuracy.
The proposed XGBoost model showed an average accuracy of 91% in predicting ICH progression, outperforming the logistic regression model which achieved an AUC of 0.82. This improved performance highlights the effectiveness of the XGBoost model in predicting ICH progression and highlights the benefits of applying advanced machine learning techniques over traditional statistical methods for clinical prediction. Furthermore, our analysis validated the great utility of the SHAP values obtained from the XGBoost model in assessing individual ICH progression risk. The incorporation of SHAP values enhances the visualization of individual risk factors, providing clinicians with an important tool to interpret the impact of different predictors on ICH progression at an individual level. This capability allows for more accurate and customized clinical decision-making.
To the best of our knowledge, this study is the first attempt to develop a machine learning model to predict ICH progression using CT scan imaging data. We hope that our findings will aid in the early identification of patients at risk for ICH progression, thereby aiding in treatment decisions and monitoring strategies. This approach may reduce the risk of complications and improve overall outcomes in traumatic brain injury (TBI) patients.
Study limitations
The current study has several limitations. First, the limited number of patients in each age group prevented us from analyzing the risk of ICH progression in different age groups. Second, we did not consider the potential influence of variables such as current medication use and underlying diseases on ICH progression in patients with TBI. Because it is difficult to obtain a complete medical history from patients who present to the emergency room with traumatic brain injury, our study focused primarily on factors that are quickly and easily available in the emergency room, especially radiological factors, to investigate their association with ICH progression. Although we investigated the history of antiplatelet and anticoagulant use, only a small proportion of patients (27 of 650, 4.2%) were confirmed to have used these medications. This limited number of patients was insufficient to establish a statistical correlation with ICH progression. This likely reflects the unreliability of the initial medical history taking and suggests that patients who were receiving antiplatelet or anticoagulant therapy may have presented with a more severe ICH and therefore may have been excluded from this study because these patients would have required immediate surgical intervention.
Third, our machine learning model was developed using data from a single institution, highlighting the need for future research to perform general validation of the model using external datasets.
Future studies aim to improve the accuracy of the algorithm to predict TBI progression. To improve the predictive potential of the current machine learning algorithm, it is important to collect more comprehensive personal information from patients' medical records. In addition, future studies should investigate factors that influence the need for surgery in patients exhibiting ICH progression, especially the changes in Glasgow Coma Scale (GCS) after follow-up and the need for subsequent surgical intervention. Such analyses are expected to have great clinical significance.
