Predicting Future eGFR in People With Type 2 Diabetes and CKD

Approximately 40% of individuals with type 2 diabetes develop chronic kidney disease (CKD). Key to prevention of CKD in patients with type 2 diabetes is early awareness and identification of those at risk of rapid progression. CKD is characterized by progressive loss of kidney function, assessed by sequential estimated glomerular filtration rate (eGFR), a measurement that can vary between individuals.

According to Mariella Gregorich, MS, and colleagues, a clinically useful prediction model of future eGFR measurements based on routinely collected laboratory data could aid clinicians in implementing interventions designed to slow the course of kidney function decline. The researchers conducted a prognostic study to develop and externally validate a model to predict future trajectories in eGFR in adults with type 2 diabetes and CKD. Results were reported online in JAMA Network Open.

The study utilized baseline and follow-up data collected between February 2010 and December 2019 from three European multinational cohorts: PROVALID (Prospective Cohort Study in Patients With Type 2 Diabetes Mellitus for Validation of Biomarkers); GCKD (German Chronic Kidney Disease); and DIACORE (Diabetes Cohorte). The total cohort included 4637 participants 18 to 75 years of age with type 2 diabetes and mildly to moderately impaired kidney function (baseline eGFR ≥30 mL/min/1.73 m²). Data analysis occurred between June 30, 2021, and January 21, 2023.

The primary outcomes and measures were 13 variables available from routine clinical care visits: age; sex; body mass index (BMI); smoking status (never or ever); hemoglobin A1c (mmol/mol and percentage); hemoglobin and serum cholesterol levels; mean arterial pressure; urinary albumin-creatinine ratio; and intake of glucose-lowering, blood pressure-lowering, or lipid-lowering medications (yes/no). Baseline was defined as the individual’s first study visit.

The primary outcome of interest was repeated measurements of eGFR recorded at baseline and at follow-up visits. The eGFR values were calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation from 2021 that includes the person’s age, sex, and serum creatinine level. A linear mixed-effects model for repeated eGFR measurements at study entry up to the last recorded follow-up visit (up to 5 years after baseline) was fit and externally validated.

Of the 4637 adults with type 2 diabetes and CKD in the overall study cohort, mean age at baseline was 63.5 years, 2680 (57.8%) were men and 1957 (42.2%) were women, and all were White. The development cohort included 3323 individuals from the PROVALID and GCKD studies (mean age, 63.2 years, 1864 men [56.1%], and 1459 [43.9%] women). The external validation cohort included 1314 individuals from the DIACORE study (mean age, 64.5 years, 816 [62.1%] men and 498 [37.9%] women; mean follow-up, 5.0 years).

The median rate of decline in eGFR, estimated using individual-specific linear regression analysis, was similar across cohorts (PROVALID: median, –1.45 mL/min/1.73 m² per year; GCKD: median, –1.43 mL/min/1.73 m² per year; DIACORE: median, –1.28 mL/min/1.73 m² per year).

Prior to updating the random effects coefficients using baseline eGFR values, the overall conditional R² was 0.90, and the marginal R² was 0.20. Age was the most important predictor, with a decrease in marginal R² of 0.10 (95% CI, 0.08-0.10). The time-specific predicted R² values ranged from 0.74 (95% CI, 0.59-0.84) at year 1 to 0.47 (95% CI, 0.25-0.68) at year 5. The C statistic ranged from 0.84 (95% CI, 0.78-0.88) at year 1 to 0.75 (95% CI, 0.67-0.82) at year 5, with lower values seen after the first follow-up year.

With the exception of BMI, smoking status, mean arterial pressure, serum cholesterol, glucose-lowering medication, and lipid-lowering medication, there was an association between all other study variables and significant decreases in eGFR; age had the greatest reduction (estimate, –0.30; 95% CI, –0.32 to –0.28). However, there was an association between the interaction of the six variables and significant decreases in eGFR. Compared with the main effect size estimates, the magnitude of the standardized interaction effects was low.

In the external validation cohort, measurements of eGFR were available from year 2 to year 5 after baseline. Overall, updating the random effects with baseline eGFR yielded excellent agreement between the predicted and the observed eGFR values in the validation cohort, particularly in the early follow-up years. The externally validated R² ranged from 0.70 (95% CI, 0.63-0.76) at year 1 to 0.58 (95% CI, 0.53-0.63) at year 5. The C statistic ranged from 0.83 (95% CI, 0.81-0.85) at year 1 to 0.79 (95% CI, 0.77-0.80) at year 5. The calibration slope was highest at follow-up year 4 and lowest at follow-up year 5, suggesting stable predictive capabilities of the model in individuals the model has not been trained on. The assessment of time-specific calibration slopes revealed an almost perfect calibration at up to 4 years after baseline and only minimal shrinkage at 5 years after baseline.

The prediction model for an individual’s eGFR at future follow-up time points, visualization of model results, and risk assessment for rapid progression was implemented as an online risk calculator.

The researchers cited some limitations to the study findings, including the three large-scale cohort studies being conducted in Europe, necessitating the use of the CKD-EPI equation to estimate GFR; the lack of standardization of creatinine assays across cohorts; the limited available data at later time points; and the lack of data on routine use of medications approved for the treatment of CKD  shortly before or after 2010 and 2011.

“This prognostic study used a linear mixed-effects model to predict eGFR trajectories among adults with type 2 diabetes and CKD; this model naturally circumvented the inherent issues related to eGFR slope estimation and fully incorporated the observed data into model estimation,” the researchers said. “Despite its complexity, the prediction model was robust, well calibrated, and suitable for implementation in a web-based application, revealing the potential of a publicly available online tool that can be used by patients, caregivers, and primary health care professionals to predict individual eGFR trajectories and disease progression up to 5 years after baseline.”

Takeaway Points

1. Researchers utilized data from three multinational cohort studies to develop and externally validate a model to predict future trajectories in estimated glomerular filtration rate (eGFR) in adults with type 2 diabetes and chronic kidney disease.
2. The model was robust and well calibrated and capable of predicting decline in kidney function up to 5 years after baseline.
3. The prediction model is publicly available and suitable for implementation in an accompanying web-based application.

Source: JAMA Network Open