Development of a machine learning model to predict acute kidney injury in patients receiving colistin : A case study at Kamphaeng Phet Hospital

Abstract

Colistin is an essential last-resort antibiotic for multidrug-resistant Gram-negative bacterial infections, but its use is constrained by a high risk of nephrotoxicity. Early identification of patients at risk of colistin-associated acute kidney injury (AKI) is crucial to improve outcomes and optimize clinical management. This study aimed to develop and evaluate machine learning (ML) models for predicting AKI in patients receiving intravenous colistin. A retrospective analytical study was conducted using electronic medical records from Kamphaeng Phet Hospital, Thailand, between 2013 and 2025. Adult patients who received intravenous colistin were included. AKI was defined according to KDIGO criteria. Thirty-two baseline variables were extracted. Three ML algorithms—Logistic Regression (LR), Random Forest (RF), and XGBoost (XGB)—were trained and tested using an 80/20 split, with Synthetic Minority Oversampling Technique (SMOTE) applied to address class imbalance. Model performance was assessed by recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). Of 3,176 patients, 1,132 (35.6%) developed AKI. Patients with AKI were older (70.4 vs. 62.2 years, p<0.001), had lower baseline eGFR, and more frequent exposure to concomitant nephrotoxic drugs. The LR model with SMOTE achieved the best predictive performance (recall 0.727, F1-score 0.657, AUC-ROC 0.802). The most influential predictors were concomitant use of ≥ 2 nephrotoxic drugs, chronic liver disease, loop diuretic use, prior AKI, and carbapenem exposure. The LR model with SMOTE demonstrated robust performance in predicting colistin -associated AKI. This model may serve as a practical screening tool to identify high-risk patients, guide nephroprotective strategies, and support clinical decision-making in antimicrobial stewardship.

References

ศูนย์เฝ้าระวังเชื้อดื้อยาต้านจุลชีพแห่งชาติ สถาบันวิจัย วิทยาศาสตร์สาธารณสุข กรมวิทยาศาสตร์การแพทย์ กระทรวงสาธารณสุข, รายงานประจำปี NARST ศูนย์เฝ้า ระวังเชื้อดื้อยาต้านจุลชีพแห่งชาติ ประจำปีงบประมาณ 2567. [Online]. Available: https://narst.dmsc.moph.go.th/static/NARST%20Annual%20Report%202567.pdf

Jirasakpisarn, "Prevalence of Antibiotic-Resistant Gram-Negative Bacteria and Susceptibility Patterns in Provincial Hospital, Thailand: A Decade Review," J Med Assoc Thai, vol. 108, no. 3, pp. 232-40, 2025, doi: 10.35755/jmedassocthai.2025.3.232-240-02250

World Health Organization, Health and Economic Impacts of Antimicrobial Resistance in the Western Pacific Region, 2020–2030, WHO Regional Office for the Western Pacific, 13 June 2023, ISBN: 978- 9290620112. [Online]. Available: https://www.who.int/southeastasia/publications/i/item/9789290620112

Macesic, A. C. Uhlemann, and A. Y. Peleg, "Multidrug-resistant Gram-negative bacterial infections," (in eng), Lancet, vol. 405, no. 10474, pp. 257-272, Jan 18 2025, doi: 10.1016/s0140-6736(24)02081-6

Li, E. Ebrahimi, M. Sholeh, R. Dousti, and E. Kouhsari, "A systematic review and meta-analysis: rising prevalence of colistin resistance in ICU-acquired Gram-negative bacteria," APMIS, vol. 133, no. 1, p. e13508, Jan 2025, doi: 10.1111/apm.13508

Wongsampan, A. Jittsue, N. Santiyanon, S. Yeepu, K. Ariyawithaya, and N. Kitjakarn, "Efficacy and Safety of the Treatment of Acinetobacter baumannii Infection with Colistin Injection," TJPP, vol. 10, no. 2, pp. 375-81, 2019.

Namwong, "Evaluation of Predictive Scoring System to Forecast the Acute Kidney Injury Associated with Intravenous Colistin in Neurosurgical Patients," Isan J Pharm Sci, vol. 21, no. 1, pp. 31-50, 2025, doi: 10.14456/ijps.2025.3

Seanglaw and T. Morasert, "Development of a Prediction Model for Acute Kidney Injury after Colistin Treatment for Multidrug-resistant Acinetobacter baumanii Ventilator-Associated Pneumonia: A Pilot Study," Journal of Health Science and Medical Research, vol. 41, no. 1, pp. 1-12, 04/19 2023, doi: 10.31584/jhsmr.2022891

Moghnieh et al., "The Prevalence and Risk Factors of Acute Kidney Injury during Colistin Therapy: A Retrospective Cohort Study from Lebanon," Antibiotics (Basel), vol. 12, no. 7, Jul 13 2023, doi: 10.3390/antibiotics12071183

Arrayasillapatorn, P. Promsen, K. Kritmetapak, S. Anunnatsiri, W. Chotmongkol, and S. Anutrakulchai, "Colistin-Induced Acute Kidney Injury and the Effect on Survival in Patients with Multidrug-Resistant Gram-Negative Infections: Significance of Drug Doses Adjusted to Ideal Body Weight," Int J Nephrol, vol. 2021, p. 7795096, 2021, doi: 10.1155/2021 /7795096.

B. Bufkin, Z. A. Karim, and J. Silva, "Review of the limitations of current biomarkers in acute kidney injury clinical practices," (in eng), SAGE Open Med, vol. 12, p. 20503121241228446, 2024, doi: 10.1177/20503121241228446

W. Chiu et al., "Machine learning algorithms to predict colistin-induced nephrotoxicity from electronic health records in patients with multidrug-resistant Gram-negative infection," Int J Antimicrob Agents, vol. 64, no. 1, p. 107175, Jul 2024, doi: 10.1016/jijantimicag.2024.107175

Kidney Disease: Improving Global Outcomes (KDIGO) Acute Kidney Injury Work Group, “KDIGO Clinical Practice Guideline for Acute Kidney Injury,” Kidney Int. Suppl., vol. 2, no. 1, pp. 1–138, 2012. [Online]. Available: https://kdigo.org/wp-content/uploads/2016/10/KDIGO-2012-AKI-Guideline-English.pdf

K. Shields et al., "Defining the incidence and risk factors of colistin-induced acute kidney injury by KDIGO criteria," PLoS One, vol. 12, no. 3, p. e0173286, 2017, doi: 10.1371/journal.pone.0173286

Deng et al., "Does Monitoring Total and Free Polymyxin B1 Plasma Concentrations Predict Polymyxin B-Induced Nephrotoxicity? A Retrospective Study in Critically Ill Patients," (in eng), Infect Dis Ther, vol. 11, no. 4, pp. 1591-1608, Aug 2022, doi: 10.1007/s40121-022-00655-3