| Peer-Reviewed

A Neural Network Classifies Traumatic Brain Injury Outcomes: Glasgow Coma Triples Are Needed

Received: 14 September 2022     Accepted: 12 October 2022     Published: 11 November 2022
Views:       Downloads:
Abstract

The Glasgow Coma Score (GCS) is statistically dubious because its calculation assumes that (a) the diagnostic scores used to assess degree of consciousness are numerical and (b) there is an implied metric. The assessed diagnostic scores are, however, categorical and there exists no metric; hence, summing is neither permitted nor medically informative. Novel methods: In this paper, we statistically analyze the Glasgow Coma Triples (GCTs) of 162 patients (114 males; 48 females; aged 3–93 years) by using unsupervised machine-learning techniques: first, one-hot encoding; second, a dimension reduction autoencoder; and finally KDE (Kernel Density Estimation). Results: We find that this sequence can classify how the resulting segmentation (triage) results in (a) the dead patients clustering separately from the survivors, and (b) the survivors clustering into five groups with different hospital discharge outcomes: from those with GCT={1,1,1} to those with GCT={4,6,5}, albeit with varying trajectories. Conclusions: The use of machine learning techniques can uncover the medical progressions of TBI patients that are impossible to discover using conventional GCS analysis. We also find a triage for outcomes, including five clusters for surviving patients. Further research is needed to verify what medically determines these varying trajectories and their ranges in probabilities; using GCS cannot contribute to these extended investigations, however.

Published in Clinical Medicine Research (Volume 11, Issue 6)
DOI 10.11648/j.cmr.20221106.11
Page(s) 150-158
Creative Commons

This is an Open Access article, distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium or format, provided the original work is properly cited.

Copyright

Copyright © The Author(s), 2022. Published by Science Publishing Group

Keywords

Traumatic Brain Injury, Glasgow Coma Score, Kernel Density Estimation, Dimension Reduction, Feature Extraction, Triage, Unsupervised Machine Learning, Glasgow Coma Triples

References
[1] Teasdale G, Jennett B. (1974) Assessment of coma and impaired consciousness: a practical scale. Lancet 304, 81–84.
[2] Teasdale G, Jennett B, Murray L, Murray, G. (1983) Glasgow Coma Scale: to sum or not to sum. Lancet. 2, 678.
[3] Moore L, Lavoie A, Camden S, Le Sage N, Sampalis JS, Bergeron E, Abdous, B. (2006) Statistical validation of the Glasgow coma score. J Trauma. 60, 1238–1243.
[4] Iankova, A. (2006) The Glasgow Coma Scale clinical application in an emergency. Emergency Nurse 14 (8), 30–35. Doi: 10.7748/en2006.12.14.8.30.c4221.
[5] Green, S. M. (2011) Cheerio, Laddie! Bidding Farewell to the Glasgow Coma Scale Annals of Emergency Medicine 58 (5), 427–430. Doi: 10.1016/j.annemergmed.2011.06.009.
[6] Zuercher M, Ummenhofer W, Baltussen A, Walder, B. (2009) The use of Glasgow Coma Scale in injury assessment: a critical review. Brain Inj 23, 371–384.
[7] Gill M, Windemuth R, Steele R, Steven, G. (2005) A comparison of the Glasgow coma score to simplified alternative scores for the prediction of traumatic brain injury outcomes. Ann Emerg Med 45, 37–42.
[8] Gill M, Steele R, Windemuth R, Green, SM. (2016) A comparison of five simplified scales to the prehospital Glasgow Coma Scale for the prediction of traumatic brain injury outcomes. Acad Emerg Med 13, 968–973.
[9] Choi JH, Jakob M, Stapf C, Marshall RS, Hartmann A, Mast H. (2008) Multimodal early rehabilitation and predictors of outcome in survivors of severe traumatic brain injury. J Trauma 65, 1028–1035. doi: 10.1097/TA.0b013e31815eba9b.
[10] Rollnik JD, Bertram M, Bucka C, Hartwich Jöbges M, Ketter G, Leineweber B, Mertl-Rötzer M, Nowak DA, Platz T, Scheidtmann K, Thomas R, von Rosen F, Wallesch CW, Woldag H, Peschel P, Mehrholz J, Pohl M. (2016) Criterion validity and sensitivity to change of the Early Re Rehabilitation Index (ERI): results from a German multi-center study. BMC Res Notes 9, 356. doi: 10.1186/s13104-016-2154-8.
[11] Kelleher, J. D. (2019) Deep Learning. MIT Press, Cambridge, MA, USA.
[12] James G, Witten D, Hastie T, Tibshirani R. An Introduction to Statistical Learning with Applications in R. New York: Springer Science+Business (2015). 426 p. doi: 10.1007/978-1-4614-7138-7.
[13] Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. Data Mining, Inference, and Prediction. New York: Springer (2017). 745 p.
[14] Murphy KP. Machine Learning. A Probabilistic Perspective. Cambridge, MA: The MIT Press (2012). 1071 p.
[15] Mohri, M., Rozdamizadeh, A., Talwalkar, A. (2018) Foundations of Machine Learning (Adaptive Computation and Machine Learning Series), Cambridge, MA: The MIT Press. 504 p.
[16] Fleuren, L. M., Klausch, T. L. T., Zwager, C. L., Schoonmade, L. J., Guo, T., Roggeveen, L. F., Swart, E. L., Girbes, A. R. J., Thoral, P., Ercole, A., Hoogendoorn, M. & Elbers, P. W. G. (2020) Machine learning for the prediction of sepsis: a systematic review and meta-analysis of diagnostic test accuracy. Intensive Car Med 46, 383–400. Doi.org/10.1007/s001.34-019-05872-y.
[17] Kruschke, J. K. (2015) Doing Bayesian Data Analysis. A Tutorial with R, JAGS, and Stan. Elsevier, Amsterdam.
Cite This Article
  • APA Style

    Hermann Prossinger, Hubert Hetz, Alexandra Acimovic, Reinhard Berger, Karim Mostafa, et al. (2022). A Neural Network Classifies Traumatic Brain Injury Outcomes: Glasgow Coma Triples Are Needed. Clinical Medicine Research, 11(6), 150-158. https://doi.org/10.11648/j.cmr.20221106.11

    Copy | Download

    ACS Style

    Hermann Prossinger; Hubert Hetz; Alexandra Acimovic; Reinhard Berger; Karim Mostafa, et al. A Neural Network Classifies Traumatic Brain Injury Outcomes: Glasgow Coma Triples Are Needed. Clin. Med. Res. 2022, 11(6), 150-158. doi: 10.11648/j.cmr.20221106.11

    Copy | Download

    AMA Style

    Hermann Prossinger, Hubert Hetz, Alexandra Acimovic, Reinhard Berger, Karim Mostafa, et al. A Neural Network Classifies Traumatic Brain Injury Outcomes: Glasgow Coma Triples Are Needed. Clin Med Res. 2022;11(6):150-158. doi: 10.11648/j.cmr.20221106.11

    Copy | Download

  • @article{10.11648/j.cmr.20221106.11,
      author = {Hermann Prossinger and Hubert Hetz and Alexandra Acimovic and Reinhard Berger and Karim Mostafa and Alexander Grieb and Heinz Steltzer},
      title = {A Neural Network Classifies Traumatic Brain Injury Outcomes: Glasgow Coma Triples Are Needed},
      journal = {Clinical Medicine Research},
      volume = {11},
      number = {6},
      pages = {150-158},
      doi = {10.11648/j.cmr.20221106.11},
      url = {https://doi.org/10.11648/j.cmr.20221106.11},
      eprint = {https://article.sciencepublishinggroup.com/pdf/10.11648.j.cmr.20221106.11},
      abstract = {The Glasgow Coma Score (GCS) is statistically dubious because its calculation assumes that (a) the diagnostic scores used to assess degree of consciousness are numerical and (b) there is an implied metric. The assessed diagnostic scores are, however, categorical and there exists no metric; hence, summing is neither permitted nor medically informative. Novel methods: In this paper, we statistically analyze the Glasgow Coma Triples (GCTs) of 162 patients (114 males; 48 females; aged 3–93 years) by using unsupervised machine-learning techniques: first, one-hot encoding; second, a dimension reduction autoencoder; and finally KDE (Kernel Density Estimation). Results: We find that this sequence can classify how the resulting segmentation (triage) results in (a) the dead patients clustering separately from the survivors, and (b) the survivors clustering into five groups with different hospital discharge outcomes: from those with GCT={1,1,1} to those with GCT={4,6,5}, albeit with varying trajectories. Conclusions: The use of machine learning techniques can uncover the medical progressions of TBI patients that are impossible to discover using conventional GCS analysis. We also find a triage for outcomes, including five clusters for surviving patients. Further research is needed to verify what medically determines these varying trajectories and their ranges in probabilities; using GCS cannot contribute to these extended investigations, however.},
     year = {2022}
    }
    

    Copy | Download

  • TY  - JOUR
    T1  - A Neural Network Classifies Traumatic Brain Injury Outcomes: Glasgow Coma Triples Are Needed
    AU  - Hermann Prossinger
    AU  - Hubert Hetz
    AU  - Alexandra Acimovic
    AU  - Reinhard Berger
    AU  - Karim Mostafa
    AU  - Alexander Grieb
    AU  - Heinz Steltzer
    Y1  - 2022/11/11
    PY  - 2022
    N1  - https://doi.org/10.11648/j.cmr.20221106.11
    DO  - 10.11648/j.cmr.20221106.11
    T2  - Clinical Medicine Research
    JF  - Clinical Medicine Research
    JO  - Clinical Medicine Research
    SP  - 150
    EP  - 158
    PB  - Science Publishing Group
    SN  - 2326-9057
    UR  - https://doi.org/10.11648/j.cmr.20221106.11
    AB  - The Glasgow Coma Score (GCS) is statistically dubious because its calculation assumes that (a) the diagnostic scores used to assess degree of consciousness are numerical and (b) there is an implied metric. The assessed diagnostic scores are, however, categorical and there exists no metric; hence, summing is neither permitted nor medically informative. Novel methods: In this paper, we statistically analyze the Glasgow Coma Triples (GCTs) of 162 patients (114 males; 48 females; aged 3–93 years) by using unsupervised machine-learning techniques: first, one-hot encoding; second, a dimension reduction autoencoder; and finally KDE (Kernel Density Estimation). Results: We find that this sequence can classify how the resulting segmentation (triage) results in (a) the dead patients clustering separately from the survivors, and (b) the survivors clustering into five groups with different hospital discharge outcomes: from those with GCT={1,1,1} to those with GCT={4,6,5}, albeit with varying trajectories. Conclusions: The use of machine learning techniques can uncover the medical progressions of TBI patients that are impossible to discover using conventional GCS analysis. We also find a triage for outcomes, including five clusters for surviving patients. Further research is needed to verify what medically determines these varying trajectories and their ranges in probabilities; using GCS cannot contribute to these extended investigations, however.
    VL  - 11
    IS  - 6
    ER  - 

    Copy | Download

Author Information
  • Department of Evolutionary Biology, University of Vienna, Vienna, Austria

  • Department of Anesthesiology and Intensive Medical Care, Trauma Center Vienna, Location Meidling, Vienna, Austria

  • Department of Anesthesiology and Intensive Medical Care, Trauma Center Vienna, Location Meidling, Vienna, Austria

  • Department of Anesthesiology and Intensive Medical Care, Trauma Center Vienna, Location Meidling, Vienna, Austria

  • Faculty of Medicine, Sigmund Freud Private University, Vienna, Austria

  • Faculty of Medicine, Sigmund Freud Private University, Vienna, Austria

  • Department of Anesthesiology and Intensive Medical Care, Trauma Center Vienna, Location Meidling, Vienna, Austria

  • Sections