Authors' Reply: Proteomic CKD Prediction in Type 2 Diabetes: Underexplored Limitations and Translational Gaps

We thank Chen et al.1 for their recognition and interest in our article published in JASN on a newly developed five-plasma protein–based algorithm, the eGFR+five-protein panel predicting CKD (eGFR+FPPC) model, for predicting incident CKD in type 2 diabetes.2 We agree with their view that, given its high sensitivity and negative predictive value but relatively low positive predictive value, the eGFR+FPPC algorithm is more useful for screening, such that high-risk individuals can be identified and reminded for adherence to regular eGFR and albuminuria testing, given the increasing availability of preventive pharmacotherapy such as sodium-glucose cotransporter 2 inhibitors.3 Chen et al. also raised concerns regarding the potential redundancy of certain proteins within the algorithm, including α-1-microglobulin/bikunin precursor and kidney injury molecule-1, due to their potentially overlapping biologic functions. In fact, we have specifically used the least absolute shrinkage and selection operator–based Cox-regression analysis to develop the eGFR+FPPC model. Such an approach can effectively remove redundant or irrelevant variables, thereby minimizing overfitting and collinearity. Future development of a multiplex protein panel would provide a more practical, cost-effective, and standardized method for the measurement of the five proteins in our model simultaneously. As noted in the limitations section of our article,2 we acknowledged that the widespread use in current clinical practice of the sodium-glucose cotransporter 2 inhibitor and glucagon-like peptide-1 receptor agonists, which are both known to reduce CKD incidence among patients with type 2 diabetes,4 was not adequately accounted for in our study. This limitation was primarily due to the recruitment period of both the Hong Kong West Diabetes Registry and the UK Biobank Pharma Proteomics Project, during which the adoption of these therapies was less prevalent. We concur with Chen et al. that the model may potentially overestimate current risk, and recalibration with contemporary data will be necessary to further optimize its performance before it can be clinically implemented. Future external validation in therapy-exposed cohorts would facilitate the adjustment of coefficients to reflect the current risk profiles and modern treatment practices, ensuring a more accurate prediction. To enhance the generalizability of our model, we included only proteins demonstrating consistent direction of effects and significant associations with incident CKD in participants with type 2 diabetes from both Hong Kong West Diabetes Registry, comprising individuals of Chinese ancestry, and UK Biobank Pharma Proteomics Project, predominantly of Caucasian descent. Further validations in larger and ethnically diverse cohorts will be important to confirm the model performance and calibration of the protein biomarkers for better prediction.