Training and evaluation of predictive models for selecting intravenous antibiotics in sepsis patients: A retrospective data case study at Khu Muang Hospital

Abstract

Sepsis is a critical condition that requires timely and effective antibiotic therapy to improve patient outcomes. With the complexity of patient presentations and the variety of available intravenous antibiotics, selecting the most appropriate treatment can be challenging. This study focuses on training and evaluating predictive models utilizing machine learning techniques to assist healthcare providers in selecting appropriate antibiotics for hospitalized sepsis patients. The objective of this research is to create a robust predictive model that leverages machine learning algorithms to identify the most suitable intravenous antibiotics based on individual patient characteristics and clinical presentations. This study was conducted at Khu Muang Hospital in Buriram, Thailand, involving a retrospective analysis of data from 190 adult patients who met the inclusion criteria and were diagnosed with sepsis between October 2022 and September 2024. The data were collected from medical records, including demographic and clinical characteristics such as age, sex, comorbidities, vital signs, laboratory test results and treatment outcomes. Several machine learning models, including Random Forest, XGBoost, Logistic Regression and Decision Tree were trained and evaluated for their predictive performance regarding antibiotic selection. The predictive models demonstrated promising performance metrics, with Random Forest, XGBoost, and Logistic Regression achieving identical accuracy and F1-scores of 0.95 and 0.92, respectively. Decision Tree showed the lowest performance accuracy and F1-scores of 0.89 and 0.86 respectively. After hyperparameter tuning, the Decision Tree model showed significant improvement by increasing its accuracy and F1-scores from 0.84 and 0.86 to 0.89. Overall, the models effectively identified key features influencing antibiotic efficacy, with age, White Blood Count and Neutrophil count being prioritized in the decision-making process. The study underscores the potential of machine learning models to enhance clinical decision-making in antibiotic selection for sepsis patients. While the models exhibited high predictive accuracy, limitations related to the dataset's size and diversity were identified. Future research should focus on obtaining larger, more representative datasets to further improve model robustness and applicability. The findings highlight the need for continued developments in predictive modeling to optimize treatment strategies for sepsis management.

References

C. Kim, Y. H. Choi, J. Y. Choi, H. J. Choi, R. W.Park, and S. J. Rhie, “Translation of machine learning-based prediction algorithms to personalised empiric antibiotic selection: A population-based cohort study, ”Int. J. Antimicrob. Agents, vol. 62, no. 5, p. 106966, 2023.

I. Poran et al., “Predicting in-hospital antibiotic use in the medical department: Derivation and validation study,” Antibiotics (Basel), vol. 11, no. 6, 2022.

C. Mistry et al., “Development and validation of a multivariable prediction model for infection-related complications in patients with common infections in UK primary care and the extent of risk-based prescribing of antibiotics,” BMC Med., vol. 18, no. 1, p. 118, 2020.

J. G. Wong, A.-H. Aung, W. Lian, D. C. Lye, C.-K. Ooi, and A. Chow, “Risk prediction models to guide antibiotic prescribing: a study on adult patients with uncomplicated upper respiratory tract infections in an emergency department,” Antimicrob. Resist. Infect. Control, vol. 9, no. 1, 2020.

J. Qin et al., “Antibiotic combinations prediction based on machine learning to multicentre clinical data and drug interaction correlation,” Int. J. Antimicrob. Agents, vol. 63, no. 5, p. 107122, 2024.

P. Theocharopoulos, S. Bersimis, S. V. Georgakopoulos, A. Karaminas, S. K. Tasoulis, and V. P. Plagianakos, “Developing predictive precision medicine models by exploiting real-world data using machine learning methods,” Journal of Applied Statistics, 2024. doi: 10.1080/02664763.2024.2315451.

B. Göksu, B. Berikol, and G. Berikol, “Predictive models in precision medicine,” pp. 177–188, 2020. doi: 10.1016/B978-0-12-817133-2.00007-0.

C.-H. Fang, V. Ravindra, S. Akhter, M. Adibuzzaman, P. Griffin, S. Subramaniam, and A. Grama, “Identifying and analyzing sepsis states: A retrospective study on patients with sepsis in ICUs,” PLOS Digital Health, vol. 1, no. 11, p. e0000130, 2022. doi: 10.1371/ journal.pdig.0000130.

M. Lugon, “Prognostic and treatment effect modeling in medical research,” 2023. doi: 10.33540/1789.

D. Lamba, W. H. Hsu, and M. Alsadhan, “Predictive analytics and machine learning for medical informatics: A survey of tasks and techniques,” pp. 1–35, 2021. doi: 10.1016/B978-0-12-821777-1.00023-9.