Sierra JMA, Quintana BKP, Hernández GJA, Enríquez SLB, Pérez RMD, Arzate QC

Language: Spanish
References: 11
Page: 171-178
PDF size: 208.78 Kb.

ABSTRACT

Objective: To validate an artificial intelligence model that can predict the mortality prognosis of hospitalized patients with sepsis.
Materials and Methods: An ambispective observational cohort study, which included electronic records of adult patients from the Central Hospital of the State of Chihuahua, Mexico, from July 2018 to March 2020 and January 2021 to January 2022. Three models were analyzed: neural networks, support vector machine and random forests. For model validation, the sample was divided into 80% for training and 20% for testing. For the last group (20%), a 10-fold cross-validation was implemented to calculate sensitivity, specificity, positive predictive value, and negative predictive value.
Results: A total of 353 files were analyzed, of which only 218 were chosen. The best model was the neural networks; however, its area under the curve (AUC) score barely reached 0.80, the random forests algorithm (AUC 0.667) and the support vec- tor machine algorithm (AUC 0.641) were below this value. Of the 3 models, only the cross-validation with the neural networks was done, of 20% of the test data, 10 validations were implemented. The AUC scores obtained in each fold ranged from 0.771 to 0.830.
Conclusions: The model is good, even working with few data. It is intended to collect a larger sample to retrain and validate the model with more data and improve learning and performance and finally be applicable to patients.

REFERENCES

1. Gyawali B, Ramakrishna K, Dhamoon AS. Sepsis: The evolutionin definition, pathophysiology, and management.SAGE Open Medicine 2019; 205031211983504. https://doi.org/10.1177/2050312119835043.

2. Gorordo-Delsol LA. Sepsis: el enemigo oculto entre líneas.Rev Med Inst Mex Seg Soc 2017; 5 5(4): 423.

3. Qingqing M, Jay M, Hoffman JL, Calvert J, Barton C, ShimabukuroD, Shieh L, et al. Multicentre validation of asepsis prediction algorithm using only vital sign data in theemergency department, general ward and ICU. BMJ Open2018; 1: e017833. doi: 10.1136/bmjopen-2017-017833.

4. Hou N, Li M, He L, Xie B, Wang L, Zhang R, et al. Predicting30-days mortality for MIMIC-III patients with sepsis-3: a machinelearning approach using XGboost. J Transl Med. 2020;18: 462. https://doi.org/10.1186/s12967-020-02620-5.

5. Wernly B, Mamandipoor B, Baldia P, Jung C, Osmani V.Machine learning predicts mortality in septic patients usingonly routinely available ABG variables: a multi-centre evaluation.Int J Med Inform 2021; 145: 104312. doi: 10.1016/j.ijmedinf.2020.104312.

6. Kong G, Lin K, Hu Y. Using machine learning methodsto predict in-hospital mortality of sepsis patients in theICU. BMC Med Inform Decis Mak 2020; 20 (1): 251. doi:10.1186/s12911-020-01271-2.

7. Kok C, Jahmunah V, Oh SL, Zhou X, Gururajan R, Tao X,et al. Automated prediction of sepsis using temporalconvolutional network. Comput Biol Med 2020; 127(103957): 103957. http://dx.doi.org/10.1016/j.compbiomed.2020.103957.

8. De Alencar Saraiva JL, Becker OM, Silva E, KadirkamanatanV, et al. Sensitivity analysis–based sepsis prognosis usingartificial intelligence. Res Biomed Eng 2020; 36: 449-461.https://doi.org/10.1007/s42600-020-00083-7.

9. Jaimes F, Farbiarz J, Alvarez D, Martínez C. Comparisonbetween logistic regression and neural networks to predictdeath in patients with suspected sepsis in the emergencyroom. Crit Care 2005; 9 (2): 150-6. doi: 10.1186/cc3054.

10. Hasegawa D, Yamakawa K, Nishida K, Okada N, Murao S,Nishida O. Comparative analysis of three machine-learningtechniques and conventional techniques for predictingsepsis-induced coagulopathy progression. J Clin Med 2020;9 (7): 2113. https://doi.org/10.3390/jcm9072113.

11. Rodríguez A, Mendoza D, Ascuntar J, Jaimes F. Supervisedclassification techniques for prediction of mortality in adultpatients with sepsis. Am J Emerg Med 2021; 45: 392-7. doi:10.1016/j.ajem.2020.09.013.