Systems-level health of patients living with end-stage kidney disease using standard lab values

We present a systems-level analysis of end-stage kidney disease (ESKD) with a dynamical network analysis of 14 commonly measured blood-based biomarkers in patients undergoing regular haemodialysis. Utilizing a validated pipeline for declining homeostatic systems, our approach learns a dynamical model together with an invertible transformation that simplifies the behaviour of observed biomarkers into natural variables. Within the natural variables, we identified two distinct dynamical behaviours: (i) stochastic accumulation, the random accumulation of abnormal values, and (ii) mallostasis, a deterministic drift towards worse health. These behaviours are identified by persistent fluctuations indicating weak stability, or a gradual shift in homeostatic set point, respectively. Both lead to worsening natural variable values, making the natural variables salient survival predictors with preferred directions of increasing risk. When this worsening is transformed back into observable biomarkers, it generates a coherent spectrum of worsening medical signs characteristic of a medical syndrome. Specifically, we found that small modules of natural variables corresponded to two existing syndromes commonly afflicting ESKD patients: protein-energy wasting and sepsis. We also identified new prospective syndromes. Our findings suggest that natural variables are robust, systems-level biomarkers, capturing the complex, holistic changes in health associated with ESKD.
Comment: 50 pages (15 for main, 35 supplemental), 10 figures in main