Predicting Mortality in Traumatic Brain Injury Patients Through Decision Tree Machine Learning

Introdução: O traumatismo cranioencefálico (TCE) continua sendo um desafio significativo para a saúde global, com altas taxas de mortalidade e morbidade. A prevenção é essencial, apesar dos avanços no tratamento. Estudos epidemiológicos sobre o TCE, especialmente no Brasil e na América Latina, são limitados. A Inteligência Artificial (AI) e as Árvores de Decisão (DT) oferecem métodos promissores para a previsão de desfechos, como demonstrado por pesquisas recentes do CDC e estudos na Europa e nos Estados Unidos. Objetivo: Este estudo analisou prontuários médicos de pacientes com TCE atendidos no Hospital de Clínicas da Universidade Federal do Triângulo Mineiro (HC-UFTM), Brasil, entre janeiro de 2007 e dezembro de 2017. Métodos: Uma abordagem quantitativa e retrospectiva foi utilizada para desenvolver e validar um modelo de AI que prevê desfechos fatais. A coleta de dados incluiu informações demográficas, clínicas e resultados de exames de imagem. Algoritmos de aprendizado de máquina, em particular as DT, foram aplicados. Os métodos principais envolveram extração de características, correção do desequilíbrio de classes com técnicas de reamostragem e avaliação de desempenho do modelo por meio de validação cruzada. Resultados e Discussão: A melhor sensibilidade foi 0,73 ± 0,04 usando o método NearMiss (profundidade máxima 6, critério de entropia). A especificidade mais alta, 0,97 ± 0,02, foi alcançada com o método No Resampling (profundidade máxima 3, critério de entropia). Para o Valor Preditivo Positivo (VPP), o melhor resultado foi 0,80 ± 0,10, também com o método No Resampling (profundidade máxima 3, entropia). A maior precisão foi 0,84 ± 0,02 com No Resampling (profundidade máxima 4, entropia). Conclusão: As DTs mostram-se promissoras na previsão de mortalidade em pacientes com TCE. No entanto, é necessária validação externa para confirmar sua aplicabilidade clínica.

Introduction: Traumatic brain injury (TBI) remains a significant global health challenge, with high rates of mortality and morbidity. Prevention is key, despite advances in treatment. Epidemiological studies on TBI, especially in Brazil and Latin America, are limited. Artificial Intelligence (AI) and Decision Trees (DTs) offer promising methods for outcome prediction, as evidenced by recent research from the CDC and studies in Europe and the United States. Objective: This study analyzed medical records from TBI patients treated at the Clinics Hospital of the Universidade Federal do Triângulo Mineiro (HC-UFTM) in Brazil, between January 2007 and December 2017. Methods: A retrospective, quantitative approach was employed to develop and validate an AI model predicting fatal outcomes. Data collection included demographic and clinical information, along with imaging results. Machine learning algorithms, particularly the DT Classifier, were applied. Key methods involved feature extraction, class imbalance correction using resampling techniques, and cross-validation for model performance evaluation. Results and Discussion: The best sensitivity was 0.73 ± 0.04 using the NearMiss method (max depth 6, entropy criterion). The highest specificity, 0.97 ± 0.02, was achieved with the No Resampling method (max depth 3, entropy). For Positive Predictive Value (PPV), the top score was 0.80 ± 0.10, also with the No Resampling method (max depth 3, entropy). The highest accuracy, 0.84 ± 0.02, came from the No Resampling method (max depth 4, entropy). Conclusion: DTs show promise in predicting mortality in TBI patients. However, external validation is necessary to confirm their clinical applicability.

1. Ghajar J. Traumatic brain injury. Lancet. 2000;356(9233):923-9. http://doi.org/10.1016/S0140-6736(00)02689-1. PMid:11036909.

2. Menon DK, Schwab K, Wright DW, Maas AI. Position statement: definition of traumatic brain injury. Arch Phys Med Rehabil. 2010;91(11):1637-40. http://doi.org/10.1016/j.apmr.2010.05.017. PMid:21044706.

3. Ferreira AMM, Santos JS, Marim AA, et al. Prognostic value of serum creatinine in traumatic brain injury. J Bras Neurocir. 2021;32(3):229-41. http://doi.org/10.22290/jbnc.v32i3.1976.

4. Kamp MA, Tahsim-Oglou Y, Steiger HJ, Hänggi D. Traumatic brain injuries in the ancient Egypt: insights from the Edwin Smith Papyrus. J Neurol Surg A Cent Eur Neurosurg. 2012;73(4):230-7. http://doi. org/10.1055/s-0032-1313635. PMid:22271381.

5. Capizzi A, Woo J, Verduzco-Gutierrez M. Traumatic brain injury: an overview of epidemiology, pathophysiology, and medical management. Med Clin North Am. 2020;104(2):213-38. http://doi.org/10.1016/j. mcna.2019.11.001. PMid:32035565.

6. Magalhães ALG, de Souza L, Faleiro RM, Teixeira AL, Miranda A. Epidemiology of traumatic brain injury in Brazil. Rev Bras Neurol. 2017;53(2):15-22. http://doi.org/10.1590/0004-282x-anp-2021-0035. PMid:35476074.

7. Fernandes RNR, Silva M. Epidemiology of traumatic brain injury in Brazil. Braz Neurosurg. 2013;32(3):136-42. http://doi. org/10.1055/s-0038-1626005.

8. Almeida CER, Sousa JL Fo, Dourado JC, Gontijo PAM, Dellaretti MA, Costa BS. Traumatic brain injury epidemiology in Brazil. World Neurosurg. 2016;87:540-7. http://doi.org/10.1016/j.wneu.2015.10.020. PMid:26485419.

9. Peterson AB, Zhou H, Thomas KE. Disparities in traumatic brain injury-related deaths: united States, 2020. J Safety Res. 2022;83:419-26. http://doi.org/10.1016/j.jsr.2022.10.001. PMid:36481035.

10. Brazinova A, Rehorcikova V, Taylor MS, et al. Epidemiology of traumatic brain injury in europe: a living systematic review. J Neurotrauma. 2021;38(10):1411-40. http://doi.org/10.1089/ neu.2015.4126. PMid:26537996.

11. Tomar GS, Singh GP, Lahkar D, et al. New biomarkers in brain trauma. Clin Chim Acta. 2018;487:325-9. http://doi.org/10.1016/j. cca.2018.10.025. PMid:30342876.

12. Albano C, Comandante L, Nolan S. Innovations in the management of cerebral injury. Crit Care Nurs Q. 2005;28(2):135-49. http://doi. org/10.1097/00002727-200504000-00006. PMid:15875444.

13. McCarthy J, Minsky ML, Rochester N, Shannon CE. A proposal for the Dartmouth summer research project on artificial intelligence, August 31, 1955. AI Mag. 2006;27(4):12. http://doi.org/10.1609/aimag. v27i4.1904.

14. Rosenblatt F. The perceptron, a perceiving and recognizing automaton Project Para. Buffalo: Cornell Aeronautical Laboratory; 1957.

15. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436-44. http://doi.org/10.1038/nature14539. PMid:26017442.

16. Kingsford C, Salzberg SL. What are decision trees? Nat Biotechnol. 2008;26(9):1011-3. http://doi.org/10.1038/nbt0908-1011. PMid:18779814.

17. Correia MM. Perfil do traumatismo cranioencefálico e sua associação prognóstica com a dosagem sérica de creatinina à admissão [dissertação]. Uberaba: Universidade Federal do Triângulo Mineiro; 2019. Available from: http://bdtd.uftm.edu.br/handle/tede/843. Accessed: 5/3/2024.

18. Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: synthetic minority over-sampling technique. J Artif Intell Res. 2002;16:321-57. http://doi.org/10.1613/jair.953.

19. Han H, Wang WY, Mao BH. Borderline-SMOTE: a new over-sampling method in imbalanced data sets learning. In: Huang DS, Zhang XP, Huang GB, editors. Advances in intelligent computing. Berlin: Springer; 2005. p. 878-87. (Lecture Notes in Computer Science; vol. 3644). http://doi.org/10.1007/11538059_91.

20. Tomek I. Two modifications of CNN. IEEE Trans Syst Man Cybern. 1976;6(11):769-72. http://doi.org/10.1109/TSMC.1976.4309452.

21. Mani I, Zhang I. kNN approach to unbalanced data distributions: a case study involving information extraction. Proceedings of Workshop on Learning from Imbalanced Datasets; 2003; Washington, D.C. Ottawa, ON: University of Ottawa; 2003. p. 1-7. (vol. 126). Available from: https://www.site.uottawa.ca/~nat/Workshop2003/jzhang.pdf. Accessed: 5/3/2024.

22. Lööv S. Comparison of undersampling methods for prediction of casting defects based on process parameters [thesis]. Uppsala: Uppsala University; 2021. Available from: https://www.diva-portal.org/smash/ record.jsf?pid=diva2:1597599. Accessed: 5/3/2024.

23. Batista GE, Prati RC, Monard MC. A study of the behavior of several methods for balancing machine learning training data. SIGKDD Explor. 2004;6(1):20-9. http://doi.org/10.1145/1007730.1007735.

24. Batista GE, Bazzan AL, Monard MC. Balancing training data for automated annotation of keywords: a case study. Wob. 2003;3:10-8.

25. John Lu ZQ. The elements of statistical learning: data mining, inference, and prediction. J R Stat Soc Ser A Stat Soc. 2010;173(3):693- 4. http://doi.org/10.1111/j.1467-985X.2010.00646_6.x.

26. Rau CS, Kuo PJ, Chien PC, Huang CY, Hsieh HY, Hsieh CH. Mortality prediction in patients with isolated moderate and severe traumatic brain injury using machine learning models. PLoS One. 2018;13(11):e0207192. http://doi.org/10.1371/journal.pone.0207192. PMid:30412613.

27. Mohd Noor NSE, Ibrahim H. Predicting outcomes in patients with traumatic brain injury using machine learning models. In: Jamaludin Z, Ali Mokhtar MN, editors. intelligent manufacturing and mechatronics. Singapore: Springer; 2020. p. 12-20. http://doi.org/10.1007/978-981- 13-9539-0_2.

28. Courville E, Kazim SF, Vellek J, et al. Machine learning algorithms for predicting outcomes of traumatic brain injury: a systematic review and meta-analysis. Surg Neurol Int. 2023;14:262. http://doi.org/10.25259/ SNI_312_2023. PMid:37560584.