Coronary disease & ACSCohort study
Prognostication of Acute Kidney Injury Following Percutaneous Coronary Intervention
- retrospective cohort study
- single-center
- tertiary medical center
- machine learning model (CatBoost)
- held-out test set (20%)
Authors' conclusion
In this single-center study, we developed a machine learning model using preprocedural variables that predicts post-PCI AKI. Although external validation is required, this model could support individualized risk stratification and preventive strategies.
Study context
- Population
- Patients undergoing percutaneous coronary intervention (PCI) at a tertiary medical center (2004–2022), routine electronic health record dataPCI procedures from 17,943 patients
- Participants
- 23,728 patients
- Exposure
- Machine learning model using preprocedural variables against acute kidney injury (AKI) defined by Kidney Disease: Improving Global Outcomes criteria
- Outcome
- AKI (absolute serum creatinine increase ≥0.3 mg/dL or relative increase ≥50% within 48 hours)
- Follow-up
- Within 48 hours
Main finding
As reported in the abstract
area under the receiver operating characteristic curve (held-out test set)
0.85 (95% CI: 0.82-0.87)
precision-recall curve (held-out test set)
0.38 (95% CI: 0.32-0.43)
Observational data: associations, not effects.
Further findings
AKI prevalence
- AKI prevalence
- 7.0%
Calibration
- integrated calibration index
- 0.017
- Calibration was excellent
Sensitivity at screening threshold (probability 0.041)
- sensitivity
- 0.83 (95% CI: 0.80-0.87)
Specificity at rule-in threshold (probability 0.199)
- specificity
- 0.92 (95% CI: 0.91-0.93)
Key predictors
- baseline creatinine, hemoglobin, uric acid, and white blood cell count
Original publication
Summarised from the published abstract; every number is checked against it