Distinguishing Diabetic Nephropathy from Hypertensive Renal Injury Using Computational Modelling of Renal Blood Flow

Chronic kidney disease (CKD) remains a major global health challenge. Diabetes mellitus (DM) and hypertension (HTN) are the two leading pathogenic drivers of CKD and frequently coexist in the same patient. Diabetic kidney disease (DKD) and hypertensive kidney disease (HKD) often exhibit overlapping clinical features in the early stages of CKD, making early and accurate diagnosis particularly challenging in patients with both DM and HTN. As loss of kidney function is irreversible, timely intervention is essential to reduce long-term healthcare costs and mortality risk.

This study presents a 0D-1D whole-circulation model incorporating a detailed renal vascular network to investigate distinct haemodynamic mechanisms across systemic and renal circulations. Virtual populations capture sex- and age-specific physiological variability and are validated against in vivo data reported in the literature. DKD and HKD models are parameterised and calibrated independently according to their respective pathophysiological characteristics and are further validated using in vivo data from published studies. Haemodynamic waveforms in virtual patients are analysed to extract biomarkers, which are subsequently used to train a logistic regression classifier for differentiating between DKD and HKD.

The results identify renal blood flow rate during diastole as the optimal biomarker for distinguishing DKD from HKD, with the main renal artery as the optimal measurement site and showing an area under the receiver operating characteristic curve (AUC = 0.84). Classification performance does not improve when more than three biomarkers are considered. The optimal biomarker set comprises pulsatility index, renal blood pressure during diastole, and renal blood pressure acceleration (AUC = 0.97), whereas the optimal measurable set consists of mean, systolic, and diastolic renal blood flow rates (AUC = 0.93).

In summary, this study demonstrates the potential of mechanistic modelling and waveform analysis to enable non-invasive differential diagnosis of CKD subtypes and to advance precision medicine. Building on this potential, the approach provides a foundation for developing virtual patients that underpin clinical decision-support tools to guide early intervention strategies.