Identifying pathways to cardiovascular mortality by causal graphical models and mediation analysis among hypertensive...

Many biochemical markers are involved in cardiovascular (CV) prognosis in the hypertensive population, but most findings are derived from a single-exposure setting, and their interaction and potential pathways remain scarce. The aim of this study was to determine the direct cause-effect relationship and the mediating effect of CV mortality to suggest potential pathways.

This prospective study analysed a data from 3559 hypertensive individuals from the National Health and Nutrition Examination Survey (1999-2018), with their CV mortality ascertained through linkage to the National Death Index on December 31, 2019. Baseline sociodemographic characteristics, habits, medical history data and serum biochemical markers, including cardiometabolic markers, inflammatory markers, liver enzyme markers, blood-cell based inflammatory and immune markers and kidney and renal markers were recorded. The Mixed Graphical Model-Fast-Causal Inference-Maximum algorithm (MGM-FCI-MAX) was applied to build a causal graphical model (CGM) depicting direct and indirect causes of CV mortality, then pathways were further identified from CGM where mediation analyses were performed.

Of the total participants, 562 (15.79%, 302 men and 260 women) CV deaths occurred after a median follow-up of 154 months. Survival analysis revealed significant sex- and ethnicity-specific differences in CV mortality rates (log-rank P < 0.01 and P < 0.01, respectively). Based on the resulting CGM, we identified three direct causes, estimated glomerular filtration rate (eGFR), blood urea nitrogen (BUN) and monocytes, of CV mortality, representing direct pathways underlying kidney and renal function and blood-cell based inflammatory function, respectively. BUN significantly mediated 30.29% of the effect of the eGFR on CV mortality, whereas neither the liver enzyme markers nor insulin pathway with the eGFR as a mediator showed a significant tendency towards a mediated effect after adjusting for covariates. Sex and race were significantly (21.73% and 20.96%, respectively) mediated by monocytes and the eGFR for CV mortality.

By using prospective survey data and background clinical knowledge, CGM retrieved direct and indirect causes of CV prognosis and identified pathways and the associated mediated effects. These insights will be useful in designing clinical protocols and targeting improvements in hypertensive patient management.