Meeting News

Machine learning model may better predict kidney function decline in type 2 diabetes

WASHINGTON — Research presented at ASN Kidney Week suggested a machine learning model that combines blood biomarkers and electronic health record data performed better in predicting rapid kidney function decline in patients with type 2 diabetes than other commonly used models.

Writing in a poster abstract that “individuals with type 2 diabetes are at increased risk of rapid kidney function decline,” Girish N. Nadkarni, MD, of Icahn School of Medicine at Mount Sinai, and colleagues applied a machine learning model to 871 patients (baseline eGFR, 74 mL/min/1.73m2; median urine albumin-to-creatinine ratio, 13 mg/g). The model integrated plasma levels of tumor necrosis factor 1 and 2, kidney injury molecule-1 and clinical features, with the goal of predicting rapid kidney function decline (defined as eGFR decline of at least 5 mL/min/1.73m2 per year).

Patients were followed for a median of 4.7 years. During this time, 19% of the study population experienced rapid kidney function decline.

In both training and test sets, researchers observed that the model using both biomarkers and clinical features (AUC: 0.82 and 0.80, respectively) outperformed a standard clinical model (AUC: 0.64), a biomarker model (AUC: 0.76) and a machine learning model using clinical features alone (AUC: 0.74). In addition, patients were stratified by rapid kidney function decline score as high, intermediate or low probability, with the model demonstrating a positive predictive value of 53% in the high-probability group and a negative predictive value of 97% in the low-probability group.

“In patients with type 2 diabetes, a [machine learning] model combining plasma biomarkers and longitudinal EHR data significantly improved prediction of [rapid kidney function decline] RFKD over standard clinical or biomarker-only models,” the researchers concluded, adding that “further validation of such approaches is needed.” – by Melissa J. Webb

Reference:

Nadkarni GN, et al. Abstract TH-PO917. Presented at: ASN Kidney Week; Nov. 7-10, 2019; Washington, D.C.

Disclosure: Nadkarni reports ties to Renalytix AI and Pensieve Health.

WASHINGTON — Research presented at ASN Kidney Week suggested a machine learning model that combines blood biomarkers and electronic health record data performed better in predicting rapid kidney function decline in patients with type 2 diabetes than other commonly used models.

Writing in a poster abstract that “individuals with type 2 diabetes are at increased risk of rapid kidney function decline,” Girish N. Nadkarni, MD, of Icahn School of Medicine at Mount Sinai, and colleagues applied a machine learning model to 871 patients (baseline eGFR, 74 mL/min/1.73m2; median urine albumin-to-creatinine ratio, 13 mg/g). The model integrated plasma levels of tumor necrosis factor 1 and 2, kidney injury molecule-1 and clinical features, with the goal of predicting rapid kidney function decline (defined as eGFR decline of at least 5 mL/min/1.73m2 per year).

Patients were followed for a median of 4.7 years. During this time, 19% of the study population experienced rapid kidney function decline.

In both training and test sets, researchers observed that the model using both biomarkers and clinical features (AUC: 0.82 and 0.80, respectively) outperformed a standard clinical model (AUC: 0.64), a biomarker model (AUC: 0.76) and a machine learning model using clinical features alone (AUC: 0.74). In addition, patients were stratified by rapid kidney function decline score as high, intermediate or low probability, with the model demonstrating a positive predictive value of 53% in the high-probability group and a negative predictive value of 97% in the low-probability group.

“In patients with type 2 diabetes, a [machine learning] model combining plasma biomarkers and longitudinal EHR data significantly improved prediction of [rapid kidney function decline] RFKD over standard clinical or biomarker-only models,” the researchers concluded, adding that “further validation of such approaches is needed.” – by Melissa J. Webb

Reference:

Nadkarni GN, et al. Abstract TH-PO917. Presented at: ASN Kidney Week; Nov. 7-10, 2019; Washington, D.C.

Disclosure: Nadkarni reports ties to Renalytix AI and Pensieve Health.

    See more from American Society of Nephrology Annual Meeting