Establishing a Risk Prediction Model for Acute Kidney Injury: Methodological Considerations and Responses to Critical Appraisal
Acute kidney injury (AKI) is a severe complication of acute myocardial infarction (AMI) associated with elevated morbidity and mortality rates. Identifying risk factors for AKI in this population is critical for early intervention and improved patient outcomes. A retrospective study by Wang et al. involving 1,124 hospitalized AMI patients aimed to develop a risk prediction model for AKI. The study identified seven independent risk factors: age >60 years, hypertension, chronic kidney disease (CKD), Killip class ≥3, extensive anterior myocardial infarction, furosemide use, and non-use of angiotensin-converting enzyme inhibitors (ACEI) or angiotensin receptor blockers (ARB). While the model demonstrated good discriminative ability (area under the receiver operating characteristic [AUROC] curve: 0.907), methodological concerns were raised regarding CKD evaluation, missing covariates, model validation, and statistical interpretation. This article examines these issues alongside the authors’ responses to clarify the study’s implications and limitations.
CKD Evaluation and Diagnostic Criteria
The study identified CKD as an independent risk factor for AKI but faced criticism for omitting details about estimated glomerular filtration rate (eGFR) calculation and CKD staging. The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) criteria define CKD stages as follows:
- Normal renal function: eGFR ≥90 mL/min/1.73 m²
- CKD Stage 1: 75–89 mL/min/1.73 m²
- CKD Stage 2: 60–74 mL/min/1.73 m²
- CKD Stage 3A: 45–59 mL/min/1.73 m²
- CKD Stage 3B: 30–44 mL/min/1.73 m²
- CKD Stage 4: 15–29 mL/min/1.73 m²
Wang et al. clarified that eGFR was calculated using the Modification of Diet in Renal Disease (MDRD) equation:
[ text{eGFR}_{text{MDRD}} = 186 times text{serum creatinine}^{-1.154} times text{Age}^{-0.203} times 0.742 , (text{if female}) times 1.210 , (text{if African American}) ]
The study focused on patients with CKD stages 3–4 (eGFR 15–59 mL/min/1.73 m²). However, the exclusion of end-stage renal disease (eGFR <15 mL/min/1.73 m²) and the lack of granular CKD stage-specific analyses limited insights into how baseline renal dysfunction severity influenced AKI risk. Prior research highlights that lower baseline eGFR correlates with higher AKI incidence post-AMI, emphasizing the need for detailed CKD stratification in predictive models.
Omission of Emergent Percutaneous Coronary Intervention (PCI)
Emergent PCI is a cornerstone of AMI management but carries a risk of contrast-induced AKI. The original study excluded PCI data from multivariate regression, raising concerns about biased odds ratios for AKI risk factors. Wang et al. later disclosed that 65.1% of patients (734/1,124) underwent PCI, with 156 developing AKI. Statistical analysis confirmed PCI as a significant AKI risk factor (( P < 0.001 )).
Multivariate regression assumes all relevant confounders are included. Exposing PCI—a known AKI contributor—could distort associations between other variables (e.g., furosemide use) and outcomes. For instance, patients undergoing PCI may receive higher diuretic doses, creating confounding. The authors’ failure to adjust for PCI undermines the model’s validity, as unmeasured variables may inflate or obscure true risk factor effects.
Model Discrimination and Validation
The study assessed model discrimination using AUROC (0.907), indicating excellent discriminative ability. However, critics noted the absence of additional metrics like calibration plots or Brier scores. Discrimination evaluates how well the model separates AKI and non-AKI cases, while calibration assesses prediction accuracy across risk strata. The Hosmer-Lemeshow test (( chi^2 = 12.848, P = 0.117 )) suggested good fit, but reliance on a single metric leaves uncertainty about model robustness.
A critical omission was the lack of internal or external validation. Predictive models derived from retrospective data are prone to overfitting, especially with numerous predictors relative to outcome events. Internal validation (e.g., bootstrapping or split-sample validation) adjusts for over-optimism in performance estimates. External validation tests generalizability to different populations or settings. The authors acknowledged this gap, stating ongoing efforts to validate the model in future work.
Clinical and Methodological Implications
The study’s risk score—incorporating age, hypertension, CKD, Killip class, infarct location, diuretic use, and ACEI/ARB non-use—provides a pragmatic tool for AKI risk stratification. However, methodological shortcomings limit immediate clinical applicability:
- CKD Staging: The broad categorization of CKD (stages 3–4) overlooks incremental AKI risks at higher eGFR levels. Future models should incorporate granular CKD stages.
- PCI Adjustment: The model’s odds ratios may misrepresent true associations due to unadjusted PCI effects. Reanalysis with PCI as a covariate is warranted.
- Validation: Without validation, the model’s performance in new cohorts remains unproven.
Statistical and Reporting Standards
The study exemplifies common pitfalls in predictive modeling:
- Transparency: Incomplete reporting of CKD criteria and PCI data hindered reproducibility.
- Multivariable Analysis: Omission of key confounders risks biased estimates.
- Performance Metrics: Reliance on AUROC without calibration or validation metrics reduces interpretability.
Adherence to guidelines like TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) could enhance rigor. TRIPOD mandates detailed reporting of predictor selection, handling of missing data, and validation steps.
Conclusion
Wang et al.’s study underscores the high incidence of AKI in AMI patients (26.0%) and its link to in-hospital mortality. Their risk prediction model, while promising, requires methodological refinements to address omitted variables, improve CKD stratification, and validate generalizability. Clinicians should interpret the current model cautiously, recognizing its limitations in covariate adjustment and validation. Future research must prioritize transparent reporting, rigorous statistical adjustment, and external validation to translate predictive models into reliable clinical tools.
doi.org/10.1097/CM9.0000000000000505
Was this helpful?
0 / 0