Postoperative mid-to-long-term adverse event prediction model for patients receiving non-cardiac surgery: An extension of the Simple Postoperative AKI RisK (SPARK) model.

Abstract

Background: Postoperative acute kidney injury (PO-AKI) is a critical complication of adverse kidney outcomes, both short and long-term. We aimed to expand our pre-existing PO-AKI prediction model to predict mid-to long-term adverse kidney outcomes.

Methods: We included patients who underwent major non-cardiac surgeries from the original SPARK cohort, two external validation cohorts, and a temporal validation cohort. Mid-to-long-term adverse kidney outcomes were defined as end-stage kidney disease progression or death within 1 or 3 years after surgery. We verified and tuned the original Simple Postoperative AKI RisK (SPARK) model to predict mid-to-long-term adverse kidney events.

Results: We included 33 636 patients in development, 71 232 patients in external validation, and 33 944 patients in temporal validation cohorts, respectively. One- and 3-year adverse kidney events occurred in 5.5% and 13.2% in the development cohort, respectively. The original SPARK score demonstrated an acceptable discriminative power for 1-year and 3-year adverse outcome risks with C indices mostly >0.7. However, the power was relatively poor when restricted to high-risk patients or those who actually developed PO-AKI. When we re-calculated the regression coefficients from a Cox model, the discriminative performances were better, especially for those with high-risk characteristics (e.g. 1-year outcome, C-index 0.72 in developmental and 0.73‒0.77 in validation datasets). Furthermore, when the model integrated the PO-AKI stage and history of malignancy with the SPARK variables, the performance was significantly enhanced (1-year, C-index 0.86 in development and 0.86‒0.88 in validation results). With the above findings, we constructed an online postoperative risk prediction system (https://snuhnephrology.github.io/postop/).

Conclusions: The addition of two clinical factors and recalibration of SPARK variables significantly improved mid-to-long-term postoperative risk prediction for mortality or dialysis after non-cardiac surgery. Our calculator helps clinicians easily predict a mid-to-long-term risk and PO-AKI occurrence by entering a few variables.