Preoperative Risk Stratification of Acute Kidney Injury After Transcatheter Aortic Valve Implantation Using a Multimarker Model

Figures

• 

  Figure 1. Patient flow diagram. A total of 435 patients underwent TAVI at our institution between December 2016 and June 2020. For the present study, we analyzed only the 186 patients in whom the author (Dr. Yumi Obata) was directly responsible for anesthetic management. Patients were excluded for the following reasons: emergency cases (n=12), conversion to open surgery (n=4), missing creatinine data (n=105), lack of consent (n=123), or explicit refusal (n=5). A preliminary analysis involving 173 patients was previously published; data from the remaining 13 patients were analyzed later to complete the full 186-case dataset. TAVI: Transcatheter aortic valve implantation.

• 

  Figure 2. Receiver operating characteristic (ROC) curves for all 19 acute kidney injury predictive models included in the internal validation. The discriminative performance of 19 predictive models was evaluated to assess their ability to predict AKI following transcatheter aortic valve implantation (TAVI). Model 1 includes the preoperative estimated glomerular filtration rate (eGFR) alone. Model 2 is a conventional clinical model consisting of eGFR, age, body mass index, and left ventricular ejection fraction. Model 3 is a full model incorporating all variables, including eGFR, urinary biomarkers (liver-type fatty acid–binding protein [L-FABP] and clusterin), serum albumin, EF, C-reactive protein, and hemoglobin. Models 4-7 are simplified models incorporating eGFR with conventional clinical variables. Models 8-19 evaluate combinations of eGFR with urinary L-FABP and/or clusterin and a single clinical variable. Internal validation was performed using the bootstrap method. Model compositions are summarized in Table 4. .

• 

  Figure 3. Calibration plots for all 19 acute kidney injury predictive models included in the internal validation. Calibration plots were used to evaluate the agreement between the predicted and observed probabilities of acute kidney injury (AKI) following transcatheter aortic valve implantation (TAVI) for 19 predictive models. Model 1 includes the preoperative estimated glomerular filtration rate (eGFR) alone. Model 2 is a conventional clinical model consisting of eGFR, age, body mass index, and left ventricular ejection fraction. Model 3 is a full model incorporating all variables, including eGFR, urinary biomarkers (liver-type fatty acid–binding protein (L-FABP) and clusterin), serum albumin, ejection fraction (EF), C-reactive protein, and hemoglobin. Models 4-7 are simplified models combining eGFR with conventional clinical variables. Models 8-19 evaluate combinations of eGFR with urinary L-FABP and/or clusterin and a single clinical variable. Internal validation was performed using the bootstrap method. Model compositions are summarized in Table 4.

• 

  Figure 4. Decision curve analysis for all 19 AKI predictive models included in the internal validation. Decision curve analysis (DCA) was employed to evaluate the clinical utility (net benefit) of 19 predictive models for acute kidney injury (AKI) following transcatheter aortic valve implantation across a range of threshold probabilities. Model 1 includes the preoperative estimated glomerular filtration rate (eGFR) alone. Model 2 is a conventional clinical model consisting of eGFR, age, body mass index, and left ventricular ejection fraction. Model 3 is a full model incorporating all variables, including eGFR, urinary biomarkers (liver-type fatty acid–binding protein (L-FABP) and clusterin), serum albumin, ejection fraction (EF), C-reactive protein, and hemoglobin. Models 4-7 are simplified models combining eGFR with conventional clinical variables. Models 8-19 evaluate combinations of eGFR with urinary L-FABP and/or clusterin and a single clinical variable. Internal validation was performed using the bootstrap method. Model compositions are summarized in Table 4.

• 

  Figure 5. Receiver operating characteristic curves (ROC) for the internal validation of acute kidney injury (AKI) predictive models ROC curves for four AKI predictive models: Model 1 (estimated glomerular filtration rate (eGFR) alone, dashed line, control), Model 2 (eGFR+age+body mass index+ejection fraction, long dashed line, traditional model), Model 10 (eGFR+urinary liver-type fatty acid–binding protein (L-FABP)+clusterin+serum albumin, dash-dot line), and Model 11 (eGFR+L-FABP+clusterin+C-reactive protein, solid line). L-FABP and clusterin were log-transformed. Curves represent internal validation using 200 bootstrap resamples.

• 

  Figure 6. Calibration plots for the internal validation of acute kidney injury (AKI) predictive models Calibration plots for four AKI predictive models: Model 1 (estimated glomerular filtration rate [eGFR] alone, dashed line, control), Model 2 (eGFR+age+body mass index+ejection fraction, long dashed line, traditional model), Model 10 (eGFR+urinary liver-type fatty acid–binding protein (L-FABP)+clusterin+serum albumin, dash-dot line, main Model 1), and Model 11 (eGFR+L-FABP+clusterin+C-reactive protein, solid line, main Model 2). The thin diagonal solid line denotes perfect calibration (predicted=observed). L-FABP and clusterin were log-transformed. Plots represent internal validation using 200 bootstrap resamples.

• 

  Figure 7. Decision curve analysis for the internal validation of acute kidney injury (AKI) predictive models Decision curves for four AKI predictive models: Model 1 (estimated glomerular filtration rate (eGFR) alone, dashed line, control), Model 2 (eGFR+age+BMI+ejection fraction, long dashed line, traditional model), Model 10 (eGFR+urinary L-FABP+clusterin+serum albumin, dash-dot line, main Model 1), and Model 11 (eGFR+L-FABP+clusterin+C-reactive protein, solid line, main Model 2). Curves for the “treat-all” and “treat-none” strategies are provided as references. L-FABP and clusterin were log-transformed. Curves represent internal validation using 200 bootstrap resamples.