Predicting salt and diuretic sensitivity in a virtual population using topological data analysis

Abstract

Salt sensitivity, with or without concomitant hypertension, is associated with increased mortality. In the clinical setting, determining salt‐sensitivity in patients is laborious and an expensive challenge with low patient compliance. A large subset of salt‐sensitive patients with hypertension are known to be resistant to diuretic therapy. Personalized medicine distinguishes specific subpopulations to better manage individual patients, but this concept has not been adequately applied to the treatment of hypertension. We hypothesized that topological data analysis (TDA) of physiological responses to salt loading in a computer‐generated cohort can predict patient salt‐sensitivity and responses to thiazide therapy. Using HumMod, an integrative mathematical model of human physiology, we studied the chronic blood pressure response to salt loading (500 mMol/day for 2 weeks) in a heterogeneous population of 4000 virtual patients created by varying cardiovascular and renal system parameters. A binned clustering algorithm distinguished salt‐sensitive ( >6 mmHg change in blood pressure after salt loading) and salt‐insensitive patient populations. One cluster of patients in the salt‐sensitive population, characterized by lower plasma sodium and glomerular filtration rate at baseline, had significantly greater fall in blood pressure in response to thiazide administration (−5.6±0.2 vs. −4.5±0.2 mmHg) which was associated with a blunted activation of the renin‐angiotensin system. We also determined that following an acute water loading simulation, urine volumes and salt output correlate with physiological sensitivity to high salt intake and diuretic therapy. This data suggests that TDA of physiological behavior in human populations could be used as a predictive tool and improve personalized medical treatment.Support or Funding InformationSupported by the National Science Foundation under Grant No. NSF EPS‐0903787 and the NIH P01 HL 51971.