EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Machine learning models based on adipocyte fatty acid-binding protein help predict the chronic and lethal outcomes of patients with drug-induced liver injury

Haoshuang Fu, Shuying Song, Hong Zhao, Bingying Du, Yaoxing Chen, Yuelin Xiao, Xinya Zang, Rongtao Lai, Ruidong Mo, Yan Huang, Tianhui Zhou, Qing Xie iD

Postgraduate Medical Journal · 23.01.2026 · 102 (1204) · 158-168

DOI: 10.1093/postmj/qgaf142 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Postgraduate Medical Journal, online erschienen am 01.09.2025. Verfasst von Haoshuang Fu und 11 weiteren Autor:innen. Beteiligte Einrichtung: Department of Infectious Diseases, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, 197 Ruijin Er Road, Huangpu District, Shanghai …. Hauptthemen laut MEDLINE: Fatty Acid-Binding Proteins, Machine Learning, Chemical and Drug Induced Liver Injury. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: National Natural Science Foundation of China.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Purpose

Some patients with drug-induced liver injury (DILI) would progress into chronicity or lethal. Although adipocyte fatty acid-binding protein (AFABP) is essential in liver diseases, its role in DILI is unknown. We aimed to investigate their association and construct predictive models for chronic/lethal DILI using machine learning.

Methods

DILI patients (n = 331) were enrolled and categorized into recovery (n = 213), chronicity (n = 89), or death/liver transplantation (LT) group (n = 29) based on 6-month follow-up. ELISA and immunohistochemistry were used to determine serum and hepatic AFABP levels, respectively. Patients were randomly divided into training (70%) and validation (30%) cohorts. Machine learning models were constructed for chronic and death/LT outcomes based on serum AFABP. Furthermore, the performance of previous models and constructed models were evaluated for predicting death/LT outcome.

Results

The AFABP level was associated with the progression of DILI patients, whatever in serum or liver. The Extreme Gradient Boosting model presented the best predictive performance for chronic DILI, with the AUROC of 0.87 (95%CI = 0.82–0.91) in training cohort and AUROC of 0.90 (95%CI = 0.82–0.95) in validation cohort. The logistic regression model presented the best predictive performance for death/LT outcome, with the AUROC of 0.90 (95%CI = 0.85–0.94) in training cohort and AUROC of 0.92 (95%CI = 0.83–0.96) in validation cohort. Furthermore, it showed better predictive performance for death/LT outcome than previous models.

Conclusion

Serum AFABP level was associated with DILI progression, and machine learning models based on AFABP accurately predicted DILI outcomes, potentially assisting clinical management. Key messages What is already known on this topic: The chronic and lethal drug-induced liver injury (DILI) harms human health. Although adipocyte fatty acid-binding protein (AFABP) is essential in liver diseases, its role in DILI is unknown. We aimed to investigate their association and construct predictive models for chronic/lethal DILI using machine learning. What this study adds: The AFABP was associated with the progression of DILI patients, whatever in serum or liver. The Extreme Gradient Boosting model presented the best predictive performance for chronic DILI. The logistic regression model presented the best predictive performance for lethal DILI. Furthermore, it showed better predictive performance for lethal DILI than previous models. How this study might affect research, practice, or policy: We demonstrated that serum AFABP level was associated with the progression of DILI, and constructed accurate machine learning models to predict DILI outcomes based on serum AFABP, which could assist the clinical management of DILI patients.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Fatty Acid-Binding ProteinsMachine LearningChemical and Drug Induced Liver InjuryHumansMaleFemaleMiddle AgedAdultLiver TransplantationLiverDisease ProgressionBiomarkersPrognosis

Stichwörter der Autor:innen

adipocyte fatty acid-binding proteinchemical and drug induced liver injurychronic diseaseclinical characteristicclinical outcomeliver failuremachine learningnoninvasivepredictive modelprognosis

Substanzen

Fatty Acid-Binding ProteinsFABP4 protein, humanBiomarkers

Autor:innen und Institutionen

  • Department of Infectious Diseases, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, 197 Ruijin Er Road, Huangpu District, Shanghai 200025, China.
  • Gastroenterology, Qingdao Central Hospital, University of Health and Rehabilitation Sciences (Qingdao Central Medical Group), 127 Siliu South Road, Shibei District, Qingdao, Shandong 266042, China.

ORCID-Profile: Qing Xie

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Fu, H., Song, S., Zhao, H., Du, B., Chen, Y., Xiao, Y., Zang, X., Lai, R., Mo, R., Huang, Y., Zhou, T., & Xie, Q. (2026). Machine learning models based on adipocyte fatty acid-binding protein help predict the chronic and lethal outcomes of patients with drug-induced liver injury. Postgraduate Medical Journal, 102(1204), 158-168. https://doi.org/10.1093/postmj/qgaf142

Vancouver

Fu H, Song S, Zhao H, Du B, Chen Y, Xiao Y, et al. Machine learning models based on adipocyte fatty acid-binding protein help predict the chronic and lethal outcomes of patients with drug-induced liver injury. Postgraduate Medical Journal. 2026;102(1204):158-168. doi:10.1093/postmj/qgaf142

BibTeX
@article{Fu2026,
  author = {Fu, Haoshuang and Song, Shuying and Zhao, Hong and Du, Bingying and Chen, Yaoxing and Xiao, Yuelin and Zang, Xinya and Lai, Rongtao and Mo, Ruidong and Huang, Yan and Zhou, Tianhui and Xie, Qing},
  title = {Machine learning models based on adipocyte fatty acid-binding protein help predict the chronic and lethal outcomes of patients with drug-induced liver injury},
  journal = {Postgraduate Medical Journal},
  year = {2026},
  volume = {102},
  number = {1204},
  pages = {158--168},
  doi = {10.1093/postmj/qgaf142}
}
Kostenlos starten →