Researchers develop predictive model for kidney failure in adults with type 2 diabetes

Researchers have developed a robust and well-tuned predictive model for renal decline. type 2 diabetes (T2D) and early to moderately advanced chronic kidney disease (CKD).

According to a prognostic study published in JAMA network open, the model was found to be reliable and accurate in predicting renal function decline up to 5 years after baseline testing. A public web-based application with results and predictive models was developed. I was. The authors hope to improve the prediction of individual eGFR trajectories and disease progression in CKD.

Between February 2010 and December 2019, investigators performed a prognostic analysis using baseline and follow-up data from three prospective multinational cohort studies (PROVALID, GCKD, and DIACORE) . This study included a total of 4,637 individuals from her 18 years to her 75 years with T2D and mild to moderate renal impairment, defined as his eGFR at baseline of at least 30 mL/min/1.73 m. of adult participants were included.²Data were analyzed between June 2021 and January 2023.

The researchers chose the following variables as predictors of eGFR rate: These variables are readily available from routine clinical practice: age, sex, body mass index, smoking status, hemoglobin A._1c levels, hemoglobin levels, serum cholesterol levels, mean arterial pressure, urinary albumin-to-creatinine ratio, and intake of glucose-lowering, blood pressure-lowering, or lipid-lowering drugs.

This study included 4,637 Caucasian adults (57.8% male) with T2D and CKD, with a mean (SD) age of 63.5 (9.1) years and a follow-up of 5.0 (0.6) years. The model development cohort included 3,323 participants from the PROVALID and GCKD studies, and the external validation cohort included 1,314 participants from the DIACORE study.

The researchers found that updating the random coefficient estimates of the prediction model with baseline eGFR values ​​resulted in better predictive performance, especially seen in the calibration curve (calibration slope at 5 years: 1.09; 95% CI, 1.04–1.15). The predictive model showed good discrimination in the validation cohort, with the lowest C statistic 5 years after baseline (0.79; 95% CI, 0.77-0.80).The predictive accuracy of the model was also R² Range from 0.70 (95% CI, 0.63-0.76) at 1 year to 0.58 (95% CI, 0.53-0.63) at 5 years.

“We addressed some common methodological limitations of predictive modeling studies, such as the lack of external validation, the generalizability of their application to unseen cohorts, and the fact that visits in primary clinical practice routinely ensure the inclusion of predictors (e.g., genetic information and serum biomarkers) that are not publicly available.”

They gave an example of one study that incorporated genetic covariates along with conventional covariates to investigate associations between known genetic variants and eGFR trajectories. However, the utilization of genome-wide genotyping is not feasible in routine clinical settings and thus not suitable for universal application.

“Furthermore, the utility of molecular biomarkers in addition to traditional clinical predictors to improve prediction is still under investigation,” they added. “Therefore, we restricted the model to data obtained in early clinical practice in patients with CKD and type 2 diabetes.”

According to the authors, some strengths of this study include the use of baseline eGFR values ​​as part of the outcome vector rather than a covariate, the use of a prespecified set of predictors, and the follow-up time. including as a primary consequence the interaction with And rigorous internal and external validation of the model. They also pointed to web-based applications as a strength. This is to provide a user-friendly predictive tool that can be used to identify high-risk patients for recruitment in clinical studies.

Some limitations of this study include that all three large cohort studies were conducted in European countries, the creatinine assay was not standardized across cohorts, and data points were sparse at later time points. and the fact that the drugs used were not current with current treatments for CKD.

Although not explicitly listed as a limitation by the authors, it is important to note that the study consisted entirely of Caucasian participants.

“Despite its complexity, the predictive model was robust, well-calibrated, and suitable for implementation in web-based applications. We have uncovered the potential of a public online tool that can be used to predict individual eGFR trajectories and disease progression from baseline to 5 years,” the authors conclude.

reference

Gregorich M, Kammer M, Heinzel A, et al. Development and validation of predictive models for future estimated glomerular filtration rate in patients with type 2 diabetes and chronic kidney disease. JAMA net open2023;6(4):e231870.doi:10.1001/jamanetworkopen.2023.1870