Severe acute kidney injury predicting model based on transcontinental databases: a single-centre prospective study.

Area of Science:

• Nephrology
• Intensive Care Medicine
• Machine Learning in Healthcare

Background:

• Acute kidney injury (AKI) diagnosis models often lack external and prospective validation.
• Predicting severe AKI within 48 hours in intensive care unit (ICU) patients is crucial.

Purpose of the Study:

• To construct and validate machine-learning models for predicting severe AKI within 48 hours in ICU patients.
• To assess the models' performance using multicenter databases and prospective validation.

Main Methods:

• A retrospective and prospective cohort study involving 58,492 ICU patients from three databases (SHZJU-ICU, MIMIC, AmsterdamUMC).
• Machine-learning algorithms were used to predict severe AKI based on demographics, vital signs, and laboratory results.
• Prospective real-time validation was conducted at a single center over one year.

Main Results:

• The prediction model achieved an area under the receiver operating characteristic curve (AUROC) of 0.86 in internal validation (SHZJU-ICU, MIMIC) and external validation (AmsterdamUMC).
• Prospective validation showed a sensitivity of 0.72, specificity of 0.80, and AUROC of 0.84.
• Out of 2532 prospectively validated patients, 344 were diagnosed with severe AKI after 358 positive predictions.

Conclusions:

• The developed severe AKI prediction model shows promise for clinical application.
• The model's effectiveness is supported by multicenter data, prospective validation, and utilization of dynamic vital signs and laboratory results.