Table 2.
Comparison of clinical prediction models for early acute kidney injury diagnosis in adult intensive care units
| Characteristic | Characteristic subgroup | Malhotra et al.(25) | Flechet et al.*(26) | Deng et al.†(28) | Chiofolo et al.(29) | Zimmerman et al.‡(30) |
|---|---|---|---|---|---|---|
| Year of publication | 2017 | 2017 | 2017 | 2019 | 2019 | |
| Sample size | Development cohort | 573 | 2,123 | 1,084 | 4,572 | 19,160 |
| Internal validation cohort | 144 | 2,123 | 1,084 | 1,958 | 4,790 | |
| External validation cohort | 1,300 | 2,367 (independent split set of the original data) | Not applicable | Not applicable | Not applicable | |
| Study type | Prospective or retrospective | Prospective | Retrospective | Prospective | Retrospective | Retrospective |
| Single-center or multicenter | Multicenter | Multicenter | Multicenter | Single-center | Single-center | |
| Patient population | Adult patients admitted to ICU without known AKI at enrollment | Adult patients admitted to ICU without a history of ESRD and a baseline SCr ≥ 4mg/dL | Adult patients admitted to ICU | Adult patients admitted to ICU without ESRD and history of AKI | Adult patients admitted to ICU without CKD and AKI on admission | |
| AKI definition | Criteria | KDIGO SCr criteria | KDIGO SCr criteria | KDIGO SCr and/or UO criteria | AKIN SCr and/or UO criteria | KDIGO SCr criteria |
| Baseline SCr | SCr measurements 7 to 365 days before ICU admission | SCr measurements 1 week to 3 months before | SCr measurements 3 to > 365 days before ICU admission | SCr measurements 180 days before ICU admission | Day1 minimum SCr | |
| Imputed baseline SCr | SCr at ICU admission | MDRD back calculation | SCr at ICU admission, or follow-up up to 365 days | MDRD back calculation | Not applicable since no patients lacked baseline SCr | |
| Incidence of outcome (%) | Development cohort | 22 | 27.7 | 30.1 | 30 | 16.5 |
| Internal validation cohort | 24 | 27.7 | 30.1 | 30 | 16.5 | |
| External validation cohort | 45 | 29.2 (independent split set of the original data) | Not applicable | Not applicable | Not applicable | |
| Missing preadmission SCr values (%) | Development cohort | 24 | 22.8 | 36.0 | Not reported | Not reported |
| Internal validation cohort | 24 | 22.8 | 36.0 | Not reported | Not reported | |
| External validation cohort | 30 | 22.9 (independent split set of the original data) | Not applicable | Not applicable | Not applicable | |
| Timeframe | Collection of risk predictors | Within 48 hours after ICU admission | Within 1 day after ICU admission | Within 1 day after ICU admission | Up to 1 day before the time of prediction | Within 1 day after ICU admission |
| Prediction | Within 1 week after study enrollment | Within 1 week after ICU admission | Within 1 week after ICU enrollment | Every 15 minutes after ICU admission | Within 72 hours after ICU admission | |
| Number of risk factors for any AKI | 10 | 12 | 5 | 19 | 16 | |
| Techniques | Variable selection | Stepwise forward elimination | Backward elimination method | Stepwise method | Stepwise regression algorithm | Backward elimination method |
| Modeling algorithm | MLR | Random forest | ULR and MLR | Random forest | MLR | |
| Internal validation | 5-fold cross-validation | Bootstrapping | Bootstrapping | Random split | 5-fold cross-validation | |
| External validation | Independent prospective cohort | Independent split set of the original data | Not applicable | Not applicable | Not applicable | |
| Discrimination (AUROC) | Development cohort | Not reported | 0.86 (0.86 - 0.86) | 0.821 (0.792 - 0.850) | 0.949 (0.943 - 0.954) | Not reported |
| Internal validation cohort | 0.792 (0.697 - 0.887) | 0.86 (0.86 - 0.86) | 0.821 (0.792 - 0.850) | 0.882 (0.867 - 0.897) | 0.78 | |
| External validation cohort | 0.81 (0.78 - 0.83) | 0.81 (0.81 - 0.81) | Not applicable | Not applicable | Not applicable | |
| Calibration | Development cohort | Not reported | Calibration slope: 0.87 (0.87 - 0.88), | Not applicable | H-L test p-value: | Not applicable |
| Calibration in the large: -0.00, and calibration curve | 0.3, and calibration curve | |||||
| Internal validation cohort | Hosmer-Lemeshow test p-value: 0.293, and calibration curve | Calibration slope: 0.87 (0.87 - 0.88), | Not applicable | Hosmer-Lemeshow test p-value: 0.3, and calibration curve | Not applicable | |
| calibration in the large: -0.00 (-0.01 - 0.00), and calibration curve | ||||||
| External validation cohort | Calibration curve | Calibration slope: 0.78 (0.78 - 0.79), calibration in the large: -0.01 (-0.01 – -0.01), and calibration curve | Not applicable | Not applicable | Not applicable |
ICU - intensive care unit; AKI - acute kidney injury; ESRD - end-stage renal disease; SCr - serum creatinine; CKD - chronic kidney disease; KDIGO - Kidney Disease: Improving Global Outcomes; MDRD - Modification of Diet in Renal Disease; UO - urine output; AKIN - Acute Kidney Injury Network; MLR - multivariate logistic regression; ULR - univariate logistic regression; AUROC - area under the receiver operating characteristics.
*
Data from Flechet et al. are only reported for the prediction model for any AKI on the first day;
†
data from Deng et al. are only reported for the prediction model for any AKI;
‡
data from Zimmerman et al. are only reported for the multivariate logistic regression model derived with the backward elimination method.