Machine Learning Model Outperforms Benchmarks in Predicting Renal Function

Machine learning models can be useful tools in predicting renal function after a partial or radical nephrectomy, which may help with clinical decision making, according to researchers from New York University (NYU) and Massachusetts General Brigham (MGB).

Jesse Persily, MD, a urology resident at NYU Langone Health in New York, New York, and colleagues developed and externally validated renal function after nephrectomy with machine learning (RFAN-ML) that outperformed previous benchmark datasets. Dr. Persily presented the study findings at the SUO 25^th Annual Meeting.

“Partial nephrectomy has been advocated as the preferred approach, when feasible, to preserve renal function and improve long-term cardiovascular and renal outcomes without compromising oncologic outcomes. However, partial nephrectomy is associated with increased perioperative morbidity,” the researchers said.

Electronic health records were used to identify 1,518 patients undergoing a partial or radical nephrectomy at MGB in Boston, Massachusetts. Input features included age at the time of nephrectomy, type of nephrectomy (partial vs radical), preoperative glomerular filtration rate (GFR), and body mass index.

Based on the hospital site, data were split into training and test samples. Then, the researchers used the selected input features to train and compare several supervised machine learning regression models. This method allowed researchers to estimate the new baseline estimated GFR, which was measured as the average of all GFR values between three and 12 months post-operatively.

Machine learning models were compared with previous benchmarks. The primary performance metric was root mean squared error (RMSE), followed by R squared and mean absolute errors (MAE) as secondary metrics. Ridge regression was the best machine learning model predicting a new baseline GFR, according to the researchers.

Study results demonstrated an RMSE of 13.5 (95% confidence interval [CI] 12.5-14.5) with RFAN-ML in the MGB patient sample (n=416). Furthermore, R squared was .732 (95% CI .677-.779) and MAE was 10.5 (95% CI 9.7-11.3).

In the validation sample from NYU patients (n=891), RFAN-ML showed an RMSE of 16.5 (95% CI 15.6-17.3), which significantly outperformed both benchmarks (benchmark 1 19.4, [95% CI 18.4- 20.3]; benchmark 2 19.1 [95% CI 18-20.2]).

“Estimating renal function after partial or radical nephrectomy can facilitate personalizing the treatment of renal masses,” the researchers said.

Reference

Persily J, Chang S, Chen C, et al. Personalizing surgery for renal masses by estimating renal function after nephrectomy with machine learning (rfan-ml). Poster #143. Presented at the 25th Annual Meeting of the Society of Urologic Oncology; December 4-6, 2024; Dallas, Texas.