Integrated Machine Learning Approach for the Early Prediction of Pressure Ulcers in Spinal Cord Injury Patients
Abstract
:1. Introduction
2. Subjects and Methods
2.1. Ethics and Study Design
2.2. Machine Learning Analysis
2.3. Statistics
3. Results
3.1. Flow of the Machine Learning Algorithm
3.2. Data Characteristics and Dataset Selection for Each Hospital
3.3. Predictive Performance of the Machine Learning Models
4. Discussion
5. Conclusions
Supplementary Materials
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Ahuja, C.S.; Wilson, J.R.; Nori, S.; Kotter, M.R.N.; Druschel, C.; Curt, A.; Fehlings, M.G. Traumatic spinal cord injury. Nat. Rev. Dis. Primers 2017, 3, 17018. [Google Scholar] [CrossRef]
- Chen, H.L.; Cai, J.Y.; Du, L.; Shen, H.W.; Yu, H.R.; Song, Y.P.; Zha, M.L. Incidence of Pressure Injury in Individuals With Spinal Cord Injury: A Systematic Review and Meta-analysis. J. Wound Ostomy Cont. Nurs. 2020, 47, 215–223. [Google Scholar] [CrossRef]
- Zakrasek, E.C.; Creasey, G.; Crew, J.D. Pressure ulcers in people with spinal cord injury in developing nations. Spinal Cord. 2015, 53, 7–13. [Google Scholar] [CrossRef]
- Gorecki, C.; Nixon, J.; Madill, A.; Firth, J.; Brown, J.M. What influences the impact of pressure ulcers on health-related quality of life? A qualitative patient-focused exploration of contributory factors. J. Tissue Viability 2012, 21, 3–12. [Google Scholar] [CrossRef] [PubMed]
- Sen, C.K. Human Wound and Its Burden: Updated 2020 Compendium of Estimates. Adv. Wound Care 2021, 10, 281–292. [Google Scholar] [CrossRef] [PubMed]
- Kruger, E.A.; Pires, M.; Ngann, Y.; Sterling, M.; Rubayi, S. Comprehensive management of pressure ulcers in spinal cord injury: Current concepts and future trends. J. Spinal Cord. Med. 2013, 36, 572–585. [Google Scholar] [CrossRef] [PubMed]
- Consortium for Spinal Cord Medicine Clinical Practice. Pressure ulcer prevention and treatment following spinal cord injury: A clinical practice guideline for health-care professionals. J. Spinal Cord. Med. 2001, 24 (Suppl. S1), S40–S101. [Google Scholar] [CrossRef] [PubMed]
- Shiferaw, W.S.; Akalu, T.Y.; Mulugeta, H.; Aynalem, Y.A. The global burden of pressure ulcers among patients with spinal cord injury: A systematic review and meta-analysis. BMC Musculoskelet. Disord. 2020, 21, 334. [Google Scholar] [CrossRef] [PubMed]
- Brienza, D.; Krishnan, S.; Karg, P.; Sowa, G.; Allegretti, A.L. Predictors of pressure ulcer incidence following traumatic spinal cord injury: A secondary analysis of a prospective longitudinal study. Spinal Cord. 2018, 56, 28–34. [Google Scholar] [CrossRef] [PubMed]
- Verschueren, J.H.; Post, M.W.; de Groot, S.; van der Woude, L.H.; van Asbeck, F.W.; Rol, M. Occurrence and predictors of pressure ulcers during primary in-patient spinal cord injury rehabilitation. Spinal Cord. 2011, 49, 106–112. [Google Scholar] [CrossRef] [PubMed]
- Gelis, A.; Dupeyron, A.; Legros, P.; Benaim, C.; Pelissier, J.; Fattal, C. Pressure ulcer risk factors in persons with SCI: Part I: Acute and rehabilitation stages. Spinal Cord. 2009, 47, 99–107. [Google Scholar] [CrossRef] [PubMed]
- Mortenson, W.B.; Miller, W.C.; Team, S.R. A review of scales for assessing the risk of developing a pressure ulcer in individuals with SCI. Spinal Cord. 2008, 46, 168–175. [Google Scholar] [CrossRef] [PubMed]
- Higgins, J.; Laramee, M.T.; Harrison, K.R.; Delparte, J.J.; Scovil, C.Y.; Flett, H.M.; Burns, A.S. The Spinal Cord Injury Pressure Ulcer Scale (SCIPUS): An assessment of validity using Rasch analysis. Spinal Cord. 2019, 57, 874–880. [Google Scholar] [CrossRef]
- Ash, D. An exploration of the occurrence of pressure ulcers in a British spinal injuries unit. J. Clin. Nurs. 2002, 11, 470–478. [Google Scholar] [CrossRef] [PubMed]
- Flett, H.M.; Delparte, J.J.; Scovil, C.Y.; Higgins, J.; Laramee, M.T.; Burns, A.S. Determining Pressure Injury Risk on Admission to Inpatient Spinal Cord Injury Rehabilitation: A Comparison of the FIM, Spinal Cord Injury Pressure Ulcer Scale, and Braden Scale. Arch. Phys. Med. Rehabil. 2019, 100, 1881–1887. [Google Scholar] [CrossRef] [PubMed]
- Chan, M.S.; Avsar, P.; McEvoy, N.L.; Patton, D.; O’Connor, T.; Nugent, L.; Moore, Z. The role of proinflammatory cytokines in the detection of early pressure ulcer development: A systematic review. J. Wound Care 2023, 32, 83–91. [Google Scholar] [CrossRef] [PubMed]
- Barrientos, S.; Brem, H.; Stojadinovic, O.; Tomic-Canic, M. Clinical application of growth factors and cytokines in wound healing. Wound Repair. Regen. 2014, 22, 569–578. [Google Scholar] [CrossRef] [PubMed]
- Tran, J.P.; McLaughlin, J.M.; Li, R.T.; Phillips, L.G. Prevention of Pressure Ulcers in the Acute Care Setting: New Innovations and Technologies. Plast. Reconstr. Surg. 2016, 138, 232S–240S. [Google Scholar] [CrossRef]
- Esteva, A.; Robicquet, A.; Ramsundar, B.; Kuleshov, V.; DePristo, M.; Chou, K.; Cui, C.; Corrado, G.; Thrun, S.; Dean, J. A guide to deep learning in healthcare. Nat. Med. 2019, 25, 24–29. [Google Scholar] [CrossRef]
- Taye, M.M. Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Future Directions. Computers 2023, 12, 91. [Google Scholar] [CrossRef]
- Kato, C.; Uemura, O.; Sato, Y.; Tsuji, T. Decision Tree Analysis Accurately Predicts Discharge Destination After Spinal Cord Injury Rehabilitation. Arch. Phys. Med. Rehabil. 2024, 105, 88–94. [Google Scholar] [CrossRef]
- Merali, Z.; Wang, J.Z.; Badhiwala, J.H.; Witiw, C.D.; Wilson, J.R.; Fehlings, M.G. A deep learning model for detection of cervical spinal cord compression in MRI scans. Sci. Rep. 2021, 11, 10473. [Google Scholar] [CrossRef] [PubMed]
- McCoy, D.B.; Dupont, S.M.; Gros, C.; Cohen-Adad, J.; Huie, R.J.; Ferguson, A.; Duong-Fernandez, X.; Thomas, L.H.; Singh, V.; Narvid, J.; et al. Convolutional Neural Network-Based Automated Segmentation of the Spinal Cord and Contusion Injury: Deep Learning Biomarker Correlates of Motor Impairment in Acute Spinal Cord Injury. AJNR Am. J. Neuroradiol. 2019, 40, 737–744. [Google Scholar] [CrossRef] [PubMed]
- Luther, S.L.; Thomason, S.S.; Sabharwal, S.; Finch, D.K.; McCart, J.; Toyinbo, P.; Bouayad, L.; Lapcevic, W.; Hahm, B.; Hauser, R.G.; et al. Machine learning to develop a predictive model of pressure injury in persons with spinal cord injury. Spinal Cord. 2023, 61, 513–520. [Google Scholar] [CrossRef] [PubMed]
- Ghaffari-Rafi, A.; Peterson, C.; Leon-Rojas, J.E.; Tadokoro, N.; Lange, S.F.; Kaushal, M.; Tetreault, L.; Fehlings, M.G.; Martin, A.R. The Role of Magnetic Resonance Imaging to Inform Clinical Decision-Making in Acute Spinal Cord Injury: A Systematic Review and Meta-Analysis. J. Clin. Med. 2021, 10, 4948. [Google Scholar] [CrossRef] [PubMed]
- Chandra, J.; Sheerin, F.; Lopez de Heredia, L.; Meagher, T.; King, D.; Belci, M.; Hughes, R.J. MRI in acute and subacute post-traumatic spinal cord injury: Pictorial review. Spinal Cord. 2012, 50, 2–7. [Google Scholar] [CrossRef] [PubMed]
- Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V.; et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 2011, 12, 2825–2830. [Google Scholar]
- Choudhary, K.; DeCost, B.; Chen, C.; Jain, A.; Tavazza, F.; Cohn, R.; Park, C.W.; Choudhary, A.; Agrawal, A.; Billinge, S.J.L.; et al. Recent advances and applications of deep learning methods in materials science. Npj Comput. Mater. 2022, 8, 59. [Google Scholar] [CrossRef]
- Paszke, A.; Gross, S.; Massa, F.; Lerer, A.; Bradbury, J.; Chanan, G.; Killeen, T.; Lin, Z.; Gimelshein, N.; Antiga, L. Pytorch: An imperative style, high-performance deep learning library. Adv. Neural Inf. Process. Syst. 2019, 32, 8026–8037. [Google Scholar]
- Kapoor, D.; Xu, C. Spinal Cord Injury AIS Predictions Using Machine Learning. eNeuro 2023, 10. [Google Scholar] [CrossRef]
- Kato, C.; Uemura, O.; Sato, Y.; Tsuji, T. Functional Outcome Prediction After Spinal Cord Injury Using Ensemble Machine Learning. Arch. Phys. Med. Rehabil. 2023, 105, 95–100. [Google Scholar] [CrossRef] [PubMed]
- Okimatsu, S.; Maki, S.; Furuya, T.; Fujiyoshi, T.; Kitamura, M.; Inada, T.; Aramomi, M.; Yamauchi, T.; Miyamoto, T.; Inoue, T.; et al. Determining the short-term neurological prognosis for acute cervical spinal cord injury using machine learning. J. Clin. Neurosci. Off. J. Neurosurg. Soc. Australas. 2022, 96, 74–79. [Google Scholar] [CrossRef]
- Fallah, N.; Noonan, V.K.; Waheed, Z.; Rivers, C.S.; Plashkes, T.; Bedi, M.; Etminan, M.; Thorogood, N.P.; Ailon, T.; Chan, E. Development of a machine learning algorithm for predicting in-hospital and 1-year mortality after traumatic spinal cord injury. Spine J. 2022, 22, 329–336. [Google Scholar] [CrossRef] [PubMed]
- DeVries, Z.; Hoda, M.; Rivers, C.S.; Maher, A.; Wai, E.; Moravek, D.; Stratton, A.; Kingwell, S.; Fallah, N.; Paquet, J. Development of an unsupervised machine learning algorithm for the prognostication of walking ability in spinal cord injury patients. Spine J. 2020, 20, 213–224. [Google Scholar] [CrossRef]
- Inoue, T.; Ichikawa, D.; Ueno, T.; Cheong, M.; Inoue, T.; Whetstone, W.D.; Endo, T.; Nizuma, K.; Tominaga, T. XGBoost, a machine learning method, predicts neurological recovery in patients with cervical spinal cord injury. Neurotrauma Rep. 2020, 1, 8–16. [Google Scholar] [CrossRef] [PubMed]
- Haisma, J.A.; van der Woude, L.H.; Stam, H.J.; Bergen, M.P.; Sluis, T.A.; Post, M.W.; Bussmann, J.B. Complications following spinal cord injury: Occurrence and risk factors in a longitudinal study during and after inpatient rehabilitation. J. Rehabil. Med. 2007, 39, 393–398. [Google Scholar] [CrossRef] [PubMed]
- Scivoletto, G.; Fuoco, U.; Morganti, B.; Cosentino, E.; Molinari, M. Pressure sores and blood and serum dysmetabolism in spinal cord injury patients. Spinal Cord 2004, 42, 473–476. [Google Scholar] [CrossRef] [PubMed]
- Abdulazeem, H.; Whitelaw, S.; Schauberger, G.; Klug, S.J. A systematic review of clinical health conditions predicted by machine learning diagnostic and prognostic models trained or validated using real-world primary health care data. PLoS ONE 2023, 18, e0274276. [Google Scholar] [CrossRef]
- Ho, S.Y.; Phua, K.; Wong, L.; Bin Goh, W.W. Extensions of the External Validation for Checking Learned Model Interpretability and Generalizability. Patterns 2020, 1, 100129. [Google Scholar] [CrossRef]
Parameters | DKUH (n = 238) | CNUH (n = 385) | ||||
---|---|---|---|---|---|---|
Non-PU (n = 199) | PU (n = 39) | p Value | Non-PU (n = 362) | PU (n = 23) | p Value | |
Baseline characteristics | ||||||
Sex (male) | 158 (79.4%) | 31 (79.5%) | 0.99 | 247 (68.2%) | 19 (82.6%) | 0.148 |
Age | 54.22 ± 14.26 | 48.95 ± 17.02 | 0.042 * | 58.51 ± 15.62 | 57.00 ± 17.91 | 0.657 |
Height | 166.85 ± 7.71 | 167.85 ± 7.51 | 0.867 | 165.00 ± 9.06 | 166.55 ± 9.54 | 0.495 |
Weight | 65.18 ± 10.59 | 65.17 ± 11.66 | 0.996 | 64.55 ± 12.38 | 63.41 ± 9.89 | 0.680 |
Alcohol consumption | 99 (50.0%) | 26 (66.7%) | 0.057 | 103 (28.6%) | 6 (26.1%) | 0.795 |
Smoking status | 73 (36.9%) | 16 (41.0%) | 0.624 | 97 (27.0%) | 5 (21.7%) | 0.579 |
Diabetes mellitus | 37 (18.7%) | 8 (20.5%) | 0.79 | 67 (18.5%) | 4 (17.4%) | 0.893 |
Hypertension | 60 (30.3%) | 13 (33.3%) | 0.708 | 130 (35.9%) | 8 (34.8%) | 0.913 |
Neurologic disease | 34 (17.1%) | 4 (10.5%) | 0.313 | 16 (4.4%) | 0 (0.0%) | 0.303 |
Cardiovascular disease | 2 (1.0%) | 3 (7.7%) | 0.008 * | 38 (10.5%) | 0 (0.0%) | 0.102 |
Pulmonary disease | 9 (4.5%) | 4 (10.3%) | 0.15 | 25 (6.9%) | 2 (8.7%) | 0.745 |
Clinical parameters | ||||||
Hospital days | 53.49 ± 31.87 | 96.72 ± 67.15 | 0.000 * | 68.15 ± 42.92 | 118.17 ± 81.66 | 0.008 * |
Braden scale | 15.48 ± 3.62 | 13.69 ± 2.45 | 0.000 * | Null | Null | Null |
Traumatic injury | 158 (79.8%) | 33 (84.6%) | 0.487 | 199 (55.0%) | 16 (69.6%) | 0.172 |
Mechanism of injury | ||||||
Traffic accident | 67 (37.6%) | 14 (40.0%) | 0.581 | 59 (30.1%) | 5 (31.3%) | 0.183 |
Falls | 58 (32.6%) | 12 (34.3%) | 48 (24.5%) | 8 (50.0%) | ||
Hit by falling objects | 12 (6.7%) | 4 (11.4%) | 7 (3.6%) | 0 (0.0%) | ||
Sports | 0 (0%) | 0 (0%) | 2 (1.0%) | 0 (0.0%) | ||
Others | 41 (23.0%) | 5 (14.3%) | 80 (40.8%) | 3 (18.8%) | ||
Combined injury | 42 (21.1%) | 13 (33.3%) | 0.098 | 56 (15.5%) | 6 (26.1%) | 0.179 |
Number of operations | 1.13 ± 0.741 | 1.51 ± 1.048 | 0.035 * | 0.82 ± 1.01 | 0.61 ± 0.84 | 0.333 |
Total time of operations (min) | 229.58 ± 146.85 | 256.77 ± 164.36 | 0.301 | 269.18 ± 176.42 | 245.67 ± 85.33 | 0.692 |
GCS total | 14.69 ± 0.92 | 14.06 ± 2.76 | 0.212 | 14.00 ± 2.68 | 12.00 ± 1.41 | 0.337 |
GCS Eye | 3.83 ± 0.44 | 3.63 ± 0.942 | 0.236 | 3.75 ± 0.87 | 4.00 ± 0.00 | 0.700 |
GCS Motor | 5.93 ± 0.30 | 5.66 ± 0.90 | 0.098 | 5.83 ± 0.58 | 6.00 ± 0.00 | 0.700 |
GCS Verbal | 4.92 ± 0.64 | 4.78 ± 1.52 | 0.605 | 4.50 ± 1.17 | 2.00 ± 1.41 | 0.018 * |
ISNCSCI | ||||||
ASIA impairment scale (AIS) | ||||||
A | 12 (6.3%) | 14 (36.8%) | 0.000 * | 23 (7.6%) | 7 (31.8%) | 0.000 * |
B | 9 (4.8%) | 5 (13.2%) | 8 (2.6%) | 4 (18.2%) | ||
C | 27 (14.3%) | 12 (31.6%) | 50 (16.5%) | 8 (36.4%) | ||
D | 138 (73.0%) | 7 (18.4%) | 221 (72.9%) | 3 (13.6%) | ||
E | 3 (1.6%) | 0 (0%) | 1 (0.3%) | 0 (0.0%) | ||
NLI (neurologic level of injury) | 9.60 ± 8.14 | 10.31 ± 7.14 | 0.612 | 10.52 ± 8.57 | 8.87 ± 6.23 | 0.241 |
Motor | ||||||
Motor level | 10.54 ± 8.66 | 10.46 ± 7.07 | 0.95 | 11.90 ± 8.72 | 8.96 ± 6.17 | 0.040 * |
UER | 20.06 ± 6.68 | 18.15 ± 8.51 | 0.193 | 20.07 ± 5.21 | 14.91 ± 9.00 | 0.012 * |
UEL | 19.13 ± 6.97 | 18.59 ± 8.15 | 0.667 | 20.07 ± 5.21 | 15.30 ± 8.77 | 0.017 * |
UEMS | 39.07 ± 12.75 | 36.74 ± 16.54 | 0.41 | 40.15 ± 9.93 | 30.22 ± 17.65 | 0.014 * |
LER | 18.22 ± 8.71 | 6.21 ± 8.08 | 0.000 * | 16.18 ± 7.42 | 5.96 ± 7.97 | 0.000 * |
LEL | 17.84 ± 8.63 | 6.32 ± 8.59 | 0.000 * | 16.13 ± 7.49 | 5.39 ± 7.67 | 0.000 * |
LEMS | 35.72 ± 16.71 | 12.50 ± 16.48 | 0.000 * | 32.31 ± 14.40 | 11.35 ± 15.54 | 0.000 * |
Motor score, total | 74.76 ± 22.42 | 48.89 ± 20.97 | 0.000 * | 72.45 ± 18.69 | 41.57 ± 28.48 | 0.000 * |
Sensory | ||||||
Sensory level | 14.19 ± 10.44 | 12.69 ± 7.36 | 0.285 | 12.07 ± 9.30 | 11.83 ± 7.14 | 0.875 |
LTR | 45.76 ± 10.40 | 39.62 ± 12.57 | 0.001 * | 39.17 ± 11.06 | 34.04 ± 11.40 | 0.032 * |
LTL | 45.62 ± 10.61 | 39.15 ± 12.49 | 0.001 * | 39.28 ± 11.11 | 33.70 ± 10.87 | 0.020 * |
LT, total | 91.38 ± 20.56 | 78.77 ± 25.00 | 0.001 * | 78.44 ± 21.92 | 67.74 ± 22.15 | 0.024 * |
PPR | 45.72 ± 10.28 | 40.08 ± 12.29 | 0.003 * | 38.38 ± 11.54 | 33.83 ± 12.64 | 0.069 |
PPL | 45.90 ± 10.59 | 39.82 ± 12.17 | 0.002 * | 38.80 ± 11.42 | 33.61 ± 11.94 | 0.036 * |
PP, total | 91.62 ± 20.35 | 79.90 ± 24.42 | 0.002 * | 77.17 ± 22.51 | 67.43 ± 24.44 | 0.046 * |
Sensory score, total | 183.00 ± 40.56 | 158.67 ± 49.18 | 0.001 * | 155.61 ± 43.84 | 135.17 ± 45.96 | 0.031 * |
K-MBI | ||||||
Self-care | 2.90 ± 1.90 | 2.33 ± 1.95 | 0.088 * | 3.12 ± 1.72 | 2.33 ± 1.97 | 0.308 |
Bathing | 1.79 ± 1.61 | 0.64 ± 0.84 | 0.000 * | 2.15 ± 1.64 | 1.67 ± 1.51 | 0.502 |
Feeding | 5.87 ± 3.91 | 5.10 ± 4.27 | 0.268 | 6.12 ± 3.60 | 6.00 ± 4.73 | 0.941 |
Toileting | 4.05 ± 3.63 | 1.31 ± 1.45 | 0.000 * | 4.44 ± 3.41 | 1.83 ± 1.84 | 0.016 * |
Stair climbing | 1.62 ± 3.04 | 0.05 ± 0.32 | 0.000 * | 1.56 ± 2.87 | 0.00 ± 0.00 | 0.001 * |
Dressing | 4.23 ± 3.29 | 2.36 ± 2.08 | 0.000 * | 5.12 ± 3.05 | 1.83 ± 1.84 | 0.014 * |
Bowel management | 6.33 ± 4.13 | 2.51 ± 3.53 | 0.000 * | 6.83 ± 4.07 | 2.33 ± 3.88 | 0.015 * |
Bladder management | 5.04 ± 4.70 | 1.08 ± 2.93 | 0.000 * | 5.76 ± 4.12 | 1.17 ± 2.04 | 0.001 * |
Ambulation | 4.91 ± 5.40 | 0.69 ± 1.15 | 0.000 * | 5.07 ± 4.60 | 1.50 ± 1.64 | 0.002 * |
Transfer | 6.79 ± 5.42 | 2.08 ± 2.26 | 0.000 * | 7.27 ± 5.08 | 3.67 ± 3.62 | 0.102 |
Total | 43.53 ± 30.17 | 18.15 ± 14.92 | 0.000 * | 46.19 ± 27.48 | 22.33 ± 17.93 | 0.044 * |
FIM | ||||||
Eating | 4.13 ± 2.28 | 3.87 ± 2.54 | 0.556 | 4.50 ± 2.12 | 4.00 ± 4.24 | 0.773 |
Grooming | 3.94 ± 2.21 | 3.33 ± 2.13 | 0.114 | 3.56 ± 1.92 | 2.50 ± 2.12 | 0.472 |
Bathing | 2.63 ± 1.63 | 1.64 ± 0.78 | 0.000 * | 2.94 ± 1.59 | 1.50 ± 0.71 | 0.228 |
Dressing upper body | 3.49 ± 1.94 | 2.67 ± 1.71 | 0.015 * | 3.72 ± 1.97 | 4.00 ± 4.24 | 0.865 |
Dressing lower body | 3.07 ± 1.91 | 1.59 ± 0.94 | 0.000 * | 3.39 ± 1.88 | 2.00 ± 1.41 | 0.330 |
Toileting | 3.03 ± 2.00 | 1.64 ± 0.84 | 0.000 * | 3.50 ± 1.95 | 1.50 ± 0.71 | 0.175 |
Self-care, total | 20.29 ± 10.84 | 14.74 ± 7.72 | 0.000 * | 21.61 ± 10.86 | 15.50 ± 13.44 | 0.466 |
Bladder control | 3.88 ± 2.76 | 1.59 ± 1.70 | 0.000 * | 4.56 ± 2.41 | 1.00 ± 0.00 | 0.000 * |
Bowel control | 4.46 ± 2.50 | 2.31 ± 1.94 | 0.000 * | 4.78 ± 2.37 | 1.50 ± 0.71 | 0.009 * |
Sphincter control, total | 8.34 ± 4.88 | 3.90 ± 3.39 | 0.000 * | 9.33 ± 4.72 | 2.50 ± 0.71 | 0.000 * |
Transfer to bed/chair/wheelchair | 3.33 ± 2.04 | 1.74 ± 0.85 | 0.000 * | 3.06 ± 1.73 | 2.00 ± 1.41 | 0.420 |
Transfer to toilet | 3.09 ± 2.04 | 1.51 ± 0.68 | 0.000 * | 3.00 ± 1.78 | 1.50 ± 0.71 | 0.263 |
Transfer to tub/shower | 2.93 ± 1.96 | 1.49 ± 0.64 | 0.000 * | 2.83 ± 1.76 | 1.50 ± 0.71 | 0.311 |
Locomotion with walk/wheelchair | 3.02 ± 1.93 | 1.44 ± 0.64 | 0.000 * | 2.89 ± 1.64 | 2.00 ± 1.41 | 0.474 |
Locomotion to stairs | 1.88 ± 1.67 | 1.05 ± 0.22 | 0.000 * | 2.00 ± 1.75 | 1.00 ± 0.00 | 0.440 |
Transfer/Locomotion, total | 14.25 ± 9.15 | 7.23 ± 2.72 | 0.000 * | 13.78 ± 8.16 | 8.00 ± 4.24 | 0.345 |
Comprehension | 6.86 ± 0.61 | 6.72 ± 0.916 | 0.344 | 6.17 ± 1.76 | 4.50 ± 3.54 | 0.255 |
Expression | 6.84 ± 0.66 | 6.69 ± 0.83 | 0.304 | 6.22 ± 1.73 | 4.50 ± 3.54 | 0.235 |
Social interaction | 6.81 ± 0.78 | 6.67 ± 1.01 | 0.324 | 6.28 ± 1.64 | 4.50 ± 3.54 | 0.201 |
Problem solving | 6.78 ± 0.81 | 6.67 ± 1.11 | 0.46 | 6.17 ± 1.76 | 4.50 ± 3.54 | 0.255 |
Memory | 6.79 ± 0.74 | 6.72 ± 0.97 | 0.58 | 6.22 ± 1.73 | 4.50 ± 3.54 | 0.235 |
Cognition, total | 33.98 ± 3.97 | 32.74 ± 6.36 | 0.248 | 31.06 ± 8.59 | 22.50 ± 17.68 | 0.235 |
FIM, total | 76.86 ± 23.31 | 58.62 ± 13.85 | 0.000 * | 75.78 ± 26.96 | 48.50 ± 36.06 | 0.201 |
Laboratory parameters | ||||||
White blood cells (×103/µL) | 8.87 ± 2.07 | 9.25 ± 2.13 | 0.301 | 6.95 ± 1.71 | 7.40 ± 2.53 | 0.234 |
Red blood cells (×106/µL) | 4.08 ± 0.39 | 3.83 ± 0.46 | 0.000 * | 4.10 ± 0.47 | 3.96 ± 0.51 | 0.150 |
Hemoglobin (g/dL) | 12.62 ± 1.26 | 12.06 ± 1.49 | 0.014 * | 12.57 ± 1.38 | 11.82 ± 1.33 | 0.012 * |
Hematocrit (%) | 37.42 ± 3.49 | 35.35 ± 4.20 | 0.001 * | 37.40 ± 3.85 | 35.44 ± 3.90 | 0.018 * |
Mean corpuscular volume (fl) | 91.77 ± 4.42 | 92.32 ± 4.22 | 0.471 | 91.41 ± 3.88 | 89.89 ± 4.81 | 0.074 |
Mean corpuscular hemoglobin (pg) | 30.95 ± 1.66 | 31.50 ± 1.70 | 0.060 | 30.68 ± 1.52 | 29.91 ± 1.61 | 0.019 * |
Mean corpuscular hemoglobin concentration (g/dL) | 33.72 ± 0.75 | 34.13 ± 0.88 | 0.003 * | 33.57 ± 0.69 | 33.29 ± 0.76 | 0.060 |
Platelets (×103/µL) | 240.29 ± 56.55 | 211.94 ± 67.70 | 0.006 * | 256.03 ± 63.97 | 284.55 ± 82.67 | 0.043 * |
Neutrophils, diff. count (%) | 60.50 ± 8.85 | 64.46 ± 8.60 | 0.016 * | 61.70 ± 7.13 | 64.88 ± 9.49 | 0.060 |
Lymphocytes, diff. count (%) | 28.24 ± 7.71 | 24.38 ± 7.31 | 0.004 * | 27.26 ± 6.53 | 23.59 ± 8.75 | 0.011 |
Monocytes, diff. count (%) | 7.48 ± 1.88 | 7.51 ± 1.80 | 0.938 | 7.37 ± 1.52 | 7.65 ± 1.69 | 0.394 |
Eosinophils, diff. count (%) | 3.33 ± 1.84 | 3.40 ± 2.29 | 0.842 | 3.07 ± 1.61 | 2.87 ± 1.51 | 0.578 |
Basophils, diff. count (%) | 0.45 ± 0.26 | 0.45 ± 0.31 | 0.985 | 0.51 ± 0.19 | 0.53 ± 0.44 | 0.792 |
Neutrophils, diff. count (×103/µL) | 4.46 ± 1.58 | 5.23 ± 2.14 | 0.010 * | 4.47 ± 1.42 | 5.09 ± 2.41 | 0.238 |
Lymphocytes, diff. count (×103/µL) | 1.84 ± 0.53 | 1.76 ± 0.61 | 0.420 | 1.74 ± 0.49 | 5.09 ± 2.41 | 0.028 * |
Monocytes, diff. count (×103/µL) | 0.52 ± 0.17 | 0.56 ± 0.16 | 0.149 | 0.50 ± 0.14 | 0.55 ± 0.19 | 0.222 |
Eosinophils, diff. count (×103/µL) | 0.21 ± 0.11 | 0.24 ± 0.18 | 0.169 | 0.19 ± 0.11 | 0.19 ± 0.10 | 0.694 |
Basophils, diff. count (×103/µL) | 0.03 ± 0.02 | 0.03 ± 0.02 | 0.470 | 0.03 ± 0.01 | 0.04 ± 0.03 | 0.602 |
Creatinine (mg/dL) | 0.71 ± 0.20 | 0.71 ± 0.27 | 0.902 | 2.25 ± 5.50 | 0.85 ± 1.03 | 0.000 * |
Blood urea nitrogen (mg/dL) | 16.11 ± 3.87 | 17.44 ± 5.61 | 0.072 | 15.23 ± 10.68 | 13.57 ± 4.23 | 0.461 |
Dataset | DKUH | CNUH | ||||
---|---|---|---|---|---|---|
Non-PU | PU | Total | Non-PU | PU | Total | |
Lab | 328 | 159 (253) | 487 | 434 | 73 (92) | 507 |
ISNCSCI | 221 | 46 (55) | 267 | 362 | 23 (24) | 385 |
K-MBI | 259 | 46 (59) | 307 | 62 | 6 (6) | 68 |
FIM | 250 | 46 (59) | 298 | 31 | 2 (2) | 33 |
ISNCSCI + K-MBI | 208 | 46 (48) | 248 | 41 | 3 (3) | 44 |
ISNCSCI + K-MBI + FIM | 200 | 46 (46) | 239 | 16 | 1 (1) | 17 |
Lab + ISNCSCI | 216 | 46 (55) | 262 | 362 | 23 (24) | 385 |
Lab + ISNCSCI + K-MBI | 207 | 46 (48) | 247 | 41 | 3 (3) | 44 |
Lab + ISNCSCI + K-MBI + FIM | 199 | 39 (46) | 238 | 16 | 1 (1) | 17 |
Model | Measure | Dataset | |||
---|---|---|---|---|---|
Lab | Lab + ISNCSCI | Lab + ISNCSCI + K-MBI | Lab + ISNCSCI + K-MBI + FIM | ||
GNN-GCN | Sensitivity | 0.442 ± 0.143 | 0.367 ± 0.190 | 0.508 ± 0.107 | 0.494 ± 0.163 |
Specificity | 0.883 ± 0.034 | 0.886 ± 0.068 | 0.913 ± 0.043 | 0.960 ± 0.036 | |
Accuracy | 0.808 ± 0.052 | 0.788 ± 0.041 | 0.837 ± 0.040 | 0.873 ± 0.040 | |
AUC | 0.656 ± 0.077 | 0.626 ± 0.078 | 0.710 ± 0.058 | 0.727 ± 0.082 | |
F1-score | 0.662 ± 0.084 | 0.622 ± 0.076 | 0.720 ± 0.064 | 0.754 ± 0.085 | |
DNN | Sensitivity | 0.420 ± 0.192 | 0.132 ± 0.101 | 0.472 ± 0.135 | 0.600 ± 0.129 |
Specificity | 0.903 ± 0.023 | 0.966 ± 0.031 | 0.920 ± 0.031 | 0.963 ± 0.035 | |
Accuracy | 0.834 ± 0.040 | 0.810 ± 0.030 | 0.836 ± 0.034 | 0.895 ± 0.040 | |
AUC | 0.647 ± 0.090 | 0.549 ± 0.053 | 0.696 ± 0.069 | 0.781 ± 0.069 | |
F1-score | 0.662 ± 0.105 | 0.545 ± 0.076 | 0.707 ± 0.067 | 0.808 ± 0.071 | |
SVM_linear | Sensitivity | 0.106 ± 0.169 | 0.245 ± 0.160 | 0.560 ± 0.132 | 0.840 ± 0.110 |
Specificity | 0.898 ± 0.008 | 0.945 ± 0.035 | 0.893 ± 0.034 | 0.968 ± 0.026 | |
Accuracy | 0.818 ± 0.015 | 0.813 ± 0.035 | 0.830 ± 0.039 | 0.944 ± 0.031 | |
AUC | 0.532 ± 0.056 | 0.595 ± 0.077 | 0.727 ± 0.071 | 0.904 ± 0.058 | |
F1-score | 0.502 ± 0.087 | 0.599 ± 0.104 | 0.723 ± 0.067 | 0.907 ± 0.052 | |
SVM_RBF | Sensitivity | 0.298 ± 0.144 | 0.278 ± 0.130 | 0.525 ± 0.139 | 0.492 ± 0.165 |
Specificity | 0.882 ± 0.024 | 0.935 ± 0.040 | 0.893 ± 0.040 | 0.930 ± 0.039 | |
Accuracy | 0.798 ± 0.037 | 0.811 ± 0.037 | 0.824 ± 0.044 | 0.847 ± 0.043 | |
AUC | 0.585 ± 0.062 | 0.606 ± 0.066 | 0.709 ± 0.075 | 0.711 ± 0.084 | |
F1-score | 0.590 ± 0.077 | 0.619 ± 0.079 | 0.709 ± 0.072 | 0.723 ± 0.085 | |
KNN | Sensitivity | 0.208 ± 0.195 | 0.432 ± 0.148 | 0.282 ± 0.128 | 0.246 ± 0.127 |
Specificity | 0.894 ± 0.018 | 0.898 ± 0.059 | 0.893 ± 0.047 | 0.811 ± 0.043 | |
Accuracy | 0.813 ± 0.031 | 0.810 ± 0.049 | 0.779 ± 0.044 | 0.811 ± 0.043 | |
AUC | 0.559 ± 0.074 | 0.665 ± 0.074 | 0.588 ± 0.067 | 0.594 ± 0.067 | |
F1-score | 0.551 ± 0.103 | 0.670 ± 0.074 | 0.593 ± 0.074 | 0.605 ± 0.086 | |
Random Forest | Sensitivity | 0.122 ± 0.131 | 0.262 ± 0.144 | 0.208 ± 0.133 | 0.073 ± 0.074 |
Specificity | 0.899 ± 0.008 | 0.933 ± 0.043 | 0.953 ± 0.042 | 0.990 ± 0.019 | |
Accuracy | 0.820 ± 0.015 | 0.807 ± 0.030 | 0.813 ± 0.030 | 0.818 ± 0.021 | |
AUC | 0.532 ± 0.040 | 0.597 ± 0.062 | 0.581 ± 0.058 | 0.532 ± 0.038 | |
F1-score | 0.510 ± 0.068 | 0.602 ± 0.073 | 0.584 ± 0.074 | 0.511 ± 0.065 | |
Logistic Regression | Sensitivity | 0.380 ± 0.196 | 0.352 ± 0.153 | 0.438 ± 0.144 | 0.683 ± 0.163 |
Specificity | 0.907 ± 0.017 | 0.942 ± 0.040 | 0.918 ± 0.036 | 0.964 ± 0.029 | |
Accuracy | 0.838 ± 0.031 | 0.831 ± 0.041 | 0.828 ± 0.036 | 0.911 ± 0.033 | |
AUC | 0.630 ± 0.084 | 0.647 ± 0.077 | 0.678 ± 0.072 | 0.823 ± 0.078 | |
F1-score | 0.643 ± 0.104 | 0.665 ± 0.083 | 0.689 ± 0.069 | 0.841 ± 0.065 |
Model | Measure | Dataset | ||
---|---|---|---|---|
Lab | ISNCSCI | Lab + ISNCSCI | ||
GNN-GCN | Sensitivity | 0.180 ± 0.195 | 0.426 ± 0.218 | 0.362 ± 0.200 |
Specificity | 0.923 ± 0.036 | 0.947 ± 0.035 | 0.945 ± 0.037 | |
Accuracy | 0.877 ± 0.034 | 0.914 ± 0.034 | 0.908 ± 0.033 | |
AUC | 0.551 ± 0.096 | 0.686 ± 0.107 | 0.653 ± 0.097 | |
F1-score | 0.538 ± 0.078 | 0.666 ± 0.092 | 0.635 ± 0.087 | |
DNN | Sensitivity | 0.108 ± 0.134 | 0.398 ± 0.214 | 0.388 ± 0.194 |
Specificity | 0.944 ± 0.031 | 0.947 ± 0.036 | 0.949 ± 0.033 | |
Accuracy | 0.892 ± 0.031 | 0.913 ± 0.035 | 0.914 ± 0.028 | |
AUC | 0.526 ± 0.068 | 0.672 ± 0.105 | 0.669 ± 0.092 | |
F1-score | 0.523 ± 0.066 | 0.654 ± 0.099 | 0.654 ± 0.077 | |
SVM_linear | Sensitivity | 0.124 ± 0.132 | 0.416 ± 0.204 | 0.418 ± 0.181 |
Specificity | 0.961 ± 0.035 | 0.950 ± 0.027 | 0.925 ± 0.038 | |
Accuracy | 0.909 ± 0.035 | 0.889 ± 0.032 | 0.894 ± 0.036 | |
AUC | 0.543 ± 0.069 | 0.643 ± 0.072 | 0.672 ± 0.089 | |
F1-score | 0.547 ± 0.084 | 0.614 ± 0.065 | 0.636 ± 0.081 | |
SVM_RBF | Sensitivity | 0.140 ± 0.173 | 0.362 ± 0.153 | 0.420 ± 0.214 |
Specificity | 0.944 ± 0.028 | 0.924 ± 0.035 | 0.957 ± 0.028 | |
Accuracy | 0.894 ± 0.028 | 0.917 ± 0.025 | 0.924 ± 0.025 | |
AUC | 0.542 ± 0.086 | 0.683 ± 0.098 | 0.689 ± 0.103 | |
F1-score | 0.537 ± 0.083 | 0.661 ± 0.086 | 0.676 ± 0.085 | |
KNN | Sensitivity | 0.340 ± 0.175 | 0.562 ± 0.236 | 0.538 ± 0.223 |
Specificity | 0.875 ± 0.040 | 0.913 ± 0.030 | 0.910 ± 0.029 | |
Accuracy | 0.842 ± 0.037 | 0.891 ± 0.027 | 0.887 ± 0.028 | |
AUC | 0.607 ± 0.085 | 0.737 ± 0.114 | 0.724 ± 0.109 | |
F1-score | 0.559 ± 0.053 | 0.661 ± 0.068 | 0.651 ± 0.069 | |
Random Forest | Sensitivity | 0.066 ± 0.114 | 0.386 ± 0.191 | 0.378 ± 0.194 |
Specificity | 0.996 ± 0.008 | 0.943 ± 0.031 | 0.945 ± 0.033 | |
Accuracy | 0.938 ± 0.010 | 0.909 ± 0.028 | 0.910 ± 0.033 | |
AUC | 0.531 ± 0.056 | 0.665 ± 0.093 | 0.662 ± 0.101 | |
F1-score | 0.533 ± 0.083 | 0.646 ± 0.080 | 0.649 ± 0.093 | |
Logistic Regression | Sensitivity | 0.148 ± 0.160 | 0.408 ± 0.184 | 0.442 ± 0.191 |
Specificity | 0.941 ± 0.035 | 0.926 ± 0.033 | 0.928 ± 0.033 | |
Accuracy | 0.892 ± 0.032 | 0.895 ± 0.032 | 0.898 ± 0.032 | |
AUC | 0.545 ± 0.078 | 0.667 ± 0.093 | 0.685 ± 0.094 | |
F1-score | 0.538 ± 0.070 | 0.636 ± 0.083 | 0.648 ± 0.082 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Kim, Y.; Lim, M.; Kim, S.Y.; Kim, T.U.; Lee, S.J.; Bok, S.-K.; Park, S.; Han, Y.; Jung, H.-Y.; Hyun, J.K. Integrated Machine Learning Approach for the Early Prediction of Pressure Ulcers in Spinal Cord Injury Patients. J. Clin. Med. 2024, 13, 990. https://doi.org/10.3390/jcm13040990
Kim Y, Lim M, Kim SY, Kim TU, Lee SJ, Bok S-K, Park S, Han Y, Jung H-Y, Hyun JK. Integrated Machine Learning Approach for the Early Prediction of Pressure Ulcers in Spinal Cord Injury Patients. Journal of Clinical Medicine. 2024; 13(4):990. https://doi.org/10.3390/jcm13040990
Chicago/Turabian StyleKim, Yuna, Myungeun Lim, Seo Young Kim, Tae Uk Kim, Seong Jae Lee, Soo-Kyung Bok, Soojun Park, Youngwoong Han, Ho-Youl Jung, and Jung Keun Hyun. 2024. "Integrated Machine Learning Approach for the Early Prediction of Pressure Ulcers in Spinal Cord Injury Patients" Journal of Clinical Medicine 13, no. 4: 990. https://doi.org/10.3390/jcm13040990