Abstract: FR-PO036
Performance of an Artificial Intelligence-Generated Risk Score for AKI Prediction
Session Information
- AKI: Epidemiology, Risk Factors, and Prevention - 2
October 25, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 101 AKI: Epidemiology, Risk Factors, and Prevention
Authors
- Claure-Del Granado, Rolando, Division of Nephrology, Hospital Obrero No 2 - Caja Nacional de Salud, Cochabamba, Bolivia, Plurinational State of
- Moya-Mamani, Juan C., Universidad Mayor de San Simon, Cochabamba, Cochabamba, Bolivia, Plurinational State of
- Malhotra, Rakesh, University of California San Diego School of Medicine, La Jolla, California, United States
- Dasgupta, Subhasis, University of California San Diego, La Jolla, California, United States
Background
Acute kidney injury (AKI) is a frequent complication in hospitalized patients and is associated with worse short—and long-term outcomes. Recent advancements in Generative AI have facilitated the development of a model that enhances risk management and decision-making processes across different acute kidney disease (AKD) cohorts. Through collaboration with OpenAI and prompt engineering, we have created a promising algorithm for assessing the risk of a group of AKD patients.
Methods
We included 303 consecutive hospitalized patients who were at moderate to high risk of AKI using the AKI Risk Assessment algorithm (Figure 1). The LLM model score was calculated at admission. Renal function was followed up daily for seven days. AKI was defined and classified by KDIGO sCr criteria. We analyzed the predictive value of this score for the subsequent development of AKI, the need for kidney replacement therapy (KRT), and mortality.
Results
The incidence of AKI was 28% (n=84) most of cases (84.5%) were mild (KDIGO stage 1) , with 53.5% of cases secondary to the use of nephrotoxins. The risk score performance was assessed with the area under the curve receiver operating characteristics (AUC ROC). The AUC ROC for AKI was 0.705 (95% CI 0.638 – 0.771; p = 0.001).
The performance of the risk prediction score for AKI generated by ChatGPT showed a sensitivity of 85.71%, specificity of 36.07%, with a positive predictive value of 96.2%, a negative predictive value of 11.73%, and an odds ratio of 3.39 (95% CI of 1.7317 to 6.6197; p = 0.0004).
Conclusion
This parsimonious model, encompassing readily available clinical features, displayed acceptable efficacy in predicting AKI. Furthermore, the parsimonious model’s integration into clinical practice could support patient risk stratification and inform treatment decisions.
Figure 1. OpenAI-Developed Acute Kidney Injury Risk Score