Refining outcome prediction after traumatic brain injury with machine learning algorithms

Outcome after traumatic brain injury (TBI) is typically assessed using the Glasgow outcome scale extended (GOSE) with levels from 1 (death) to 8 (upper good recovery). Outcome prediction has classically been dichotomized into either dead/alive or favorable/unfavorable outcome. Binary outcome prediction models limit the possibility of detecting subtle yet significant improvements. We set out to explore different machine learning methods with the purpose of mapping their predictions to the full 8 grade scale GOSE following TBI. The models were set up using the variables: age, GCS-motor score, pupillary reaction, and Marshall CT score. For model setup and internal validation, a total of 866 patients could be included. For external validation, a cohort of 369 patients were included from Leuven, Belgium, and a cohort of 573 patients from the US multi-center ProTECT III study. Our findings indicate that proportional odds logistic regression (POLR), random forest regression, and a neural network model achieved accuracy values of 0.3-0.35 when applied to internal data, compared to the random baseline which is 0.125 for eight categories. The models demonstrated satisfactory performance during external validation in the data from Leuven, however, their performance were not satisfactory when applied to the ProTECT III dataset.

Association of the bleeding time test with aspects of traumatic brain injury in patients with alcohol use disorder

BACKGROUND-AIM

Traumatic brain injury (TBI) and alcohol use disorder (AUD) can occur concomitantly and be associated with coagulopathy that influences TBI outcome. The use of bleeding time tests in TBI management is controversial. We hypothesized that in TBI patients with AUD, a prolonged bleeding time is associated with more severe injury and poor outcome.

MATERIAL AND METHODS

Moderate and severe TBI patients with evidence of AUD were examined with bleeding time according to IVY bleeding time on admission during neurointensive care. Baseline clinical and radiological characteristics were recorded. A standardized IVY bleeding time test was determined by staff trained in the procedure. Bleeding time test results were divided into normal (≤ 600 s), prolonged (> 600 s), and markedly prolonged (≥ 900 s). Normal platelet count (PLT) was defined as > 150,000/μL. This cohort was compared with another group of TBI patients without evidence of AUD.

RESULTS

Fifty-two patients with TBI and AUD were identified, and 121 TBI patients without any history of AUD were used as controls. PLT was low in 44.2% and bleeding time was prolonged in 69.2% of patients. Bleeding time values negatively correlated with PLT (p < 0.05). TBI patients with markedly prolonged values (≥ 900 s) had significantly increased hematoma size, and more frequently required intracranial pressure measurement and mechanical ventilation compared with those with bleeding times < 900 s (p < 0.05). Most patients (88%) with low platelet count had prolonged bleeding time. No difference in 6-month outcome between the bleeding time groups was observed (p > 0.05). Subjects with TBI and no evidence for AUD had lower bleeding time values and higher platelet count compared with those with TBI and history of AUD (p < 0.05).

CONCLUSIONS

Although differences in the bleeding time values between TBI cohorts exist and prolonged values may be seen even in patients with normal platelet count, the bleeding test is a marker of primary hemostasis and platelet function with low specificity. However, it may provide an additional assessment in the interpretation of the overall status of TBI patients with AUD. Therefore, the bleeding time test should only be used in combination with the patient's bleeding history and careful assessment of other hematologic parameters.

Mechanisms of blast induced brain injuries, experimental studies in rats

Traumatic brain injuries (TBI) potentially induced by blast waves from detonations result in significant diagnostic problems. It may be assumed that several mechanisms contribute to the injury. This study is an attempt to characterize the presumed components of the blast induced TBI. Our experimental models include a blast tube in which an anesthetized rat can be exposed to controlled detonations of explosives that result in a pressure wave with a magnitude between 130 and 260 kPa. In this model, the animal is fixed with a metal net to avoid head acceleration forces. The second model is a controlled penetration of a 2mm thick needle. In the third model the animal is subjected to a high-speed sagittal rotation angular acceleration. Immunohistochemical labeling for amyloid precursor protein revealed signs of diffuse axonal injury (DAI) in the penetration and rotation models. Signs of punctuate inflammation were observed after focal and rotation injury. Exposure in the blast tube did not induce DAI or detectable cell death, but functional changes. Affymetrix Gene arrays showed changes in the expression in a large number of gene families including cell death, inflammation and neurotransmitters in the hippocampus after both acceleration and penetration injuries. Exposure to the primary blast wave induced limited shifts in gene expression in the hippocampus. The most interesting findings were a downregulation of genes involved in neurogenesis and synaptic transmission. These experiments indicate that rotational acceleration may be a critical factor for DAI and other acute changes after blast TBI. The further exploration of the mechanisms of blast TBI will have to include a search for long-term effects.