Design: Systematic review and meta-analysis of prediction models
Population: Adults undergoing cardiac surgery with cardiopulmonary bypass
Primary outcome: Discrimination for AKI and AKI requiring dialysis
Effect: Thakar AUC 0.80 (95% CI 0.77-0.82) for dialysis-requiring AKI

A meta-analysis in Critical Care reviewed clinical prediction models for acute kidney injury after cardiac surgery with cardiopulmonary bypass. Published on 15 October 2026.

Pooled incidence of AKI was 26%, and AKI needing renal replacement therapy 2.2%. For dialysis-requiring AKI, the Thakar score performed best (pooled AUC 0.80), with sensitivity of 87% and specificity of 49% at 3 points or more, though it tended to underestimate risk. The Mehta and Wijeysundera models also discriminated well. Prediction of milder AKI was weaker.

Identifying high-risk patients before surgery allows nephrotoxin avoidance, closer monitoring and informed consent.

The older models were derived before current surgical and perfusion practice, and calibration was imperfect.

Use a validated score such as Thakar to identify patients at high risk of dialysis after cardiac surgery
Avoid nephrotoxins and contrast close to surgery in high-risk patients
Monitor urine output and creatinine closely after surgery in those with high scores
Discuss the risk of kidney injury when consenting high-risk patients

Why it matters

Nephrologists asked to see cardiac surgical patients can stratify risk before, not after, the injury.

The statistics, in plain English

An AUC of 0.80 is good discrimination. High sensitivity with low specificity means the score catches most patients who will need dialysis but also flags many who will not.