SNUBH establishes algorithm to predict acute injury after kidney cancer surgery

A Seoul National University Bundang Hospital (SNUBH) research team has developed an algorithm that can predict acute kidney injury in patients who have received kidney cancer surgery.

A SNUBH research team, led by Professor Kim Se-joong, has developed an algorithm to predict acute kidney injury after kidney cancer surgery.

Acute kidney injury is a disease in which renal function weakens due to sudden damage to the kidney cells. If not treated early, it can progress irreversibly and lead to dangerous situations such as dialysis and death, the hospital said.

“Notably, careful monitoring is necessary to ensure that acute renal damage does not occur after surgery in patients with kidney cancer,” the hospital said.

Accordingly, the research team, led by Professors Kim Se-joong of the Department of Renal Internal Medicine and Lee Sang-cheol of the Department of Urology, developed an acute kidney injury prediction system after kidney cancer surgery using a machine learning algorithm. Professors Han Seung-seok and Kwak Cheol at Seoul National University Hospital also participated in the study.

In developing the algorithm, the research team created and verified an acute kidney injury prediction model using machine learning techniques using data from 4,104 patients who underwent unilateral nephrectomy for kidney cancer at SNUBH and SNUH from 2003 to 2017. The data used in the surgery included surgery type and time, gender, underlying disease, and tumor size.

Based on such data, the team built a predictive model using four machine learning techniques -- support vector machine, random forest, extreme gradient boosting, and light gradient boosting machine.

As a result, the team confirmed that sudden injury after surgery occurred in a total of 1,167 patients out of 4,104, accounting for an incidence of 28.4 percent.

Based on the area under the receiver operating characteristic (AUROC), an index for evaluating algorithm performance, the machine learning models showed higher performance than the simple postoperative AKI risk (SPARK) index used in the past. Among them, the light GBM model showed the most predictability with an AUROC of 0.81.

“Using multi-center clinical data rather than a single institution, we managed to develop an algorithm that can apply to multiple institutions through machine learning techniques,” Professor Kim said. When applied to clinical trials in the future, the team expects that the algorithm will improve the predictive value of acute kidney injury after kidney cancer surgery.”

Scientific Reports has published the result of the study.