Blood pressure and mortality: using offspring blood pressure as an instrument for own blood pressure in the HUNT study

Abstract

Given that observational associations may be inaccurate, we used offspring blood pressure (BP) to provide alternative estimates of the associations between own BP and mortality. Observational associations between BP and mortality, estimated as hazard ratios (HRs) from Cox regression, were compared to HRs obtained using offspring BP as an instrumental variable (IV) for own BP (N = 32,227 mother-offspring and 27,535 father-offspring pairs). Observationally, there were positive associations between own BP and mortality from all-causes, cardiovascular disease (CVD), coronary heart disease (CHD), stroke and diabetes. Point estimates of the associations between BP and mortality from all-causes, CVD and CHD were amplified in magnitude when using offspring BP as an IV. For example, the HR for all-cause mortality per standard deviation (SD) increase in own systolic BP (SBP) obtained in conventional observational analyses increased from 1.10 (95% CI: 1.09-1.12; P < 0.0001) to 1.31 (95% CI: 1.19-1.43; P < 0.0001). Additionally, SBP was positively associated with diabetes and cancer mortality (HRs: 2.00; 95% CI: 1.12-3.35; P = 0.02 and 1.20; 95% CI: 1.02-1.42; P = 0.03, respectively), and diastolic BP (DBP) with stroke mortality (HR: 1.30; 95% CI: 1.02-1.66; P = 0.03). Results support positive associations between BP and mortality from all-causes, CVD, and CHD, SBP on cancer mortality, and DBP on stroke mortality.

Figure 1. Diagram representing the underlying IV methodology employed in analyses using the example of assessing the association between SBP and CVD mortality.

A) In observational epidemiology, the association between the exposure (here, own SBP (SBP_own)) and an outcome of interest (here, own mortality from CVD (CVD_own)) may be distorted due to confounding, bias and reverse causation. B) Using offspring SBP (SBP_off) as an instrument for SBP_own (where SBP_off explains ~2.4% of the variance in SBP_own) reduces the possibility of such limitations. Firstly, unlike observational analyses, the portion of variance in SBP_own explained by SBP_off will not be directly affected by CVD_own; therefore, reducing bias due to reverse causation. Secondly, the effect of potential confounding variables is reduced because they are less associated with SBP_off than they are with SBP_own. Any remaining confounding due to the association between confounders and SBP_off can be further reduced by adjustment for shared and measured environmental confounders (dotted lines).

Figure 2. Fitted HRs (median SBP as a reference) from cubic spline models for selected mortality causes.

Shaded areas represent 95% CI and vertical lines represent the 1^st and 99^th percentiles of systolic blood pressure (SBP). Although similar, the upper X-axis applies to male SBP and the lower X-axis to female SBP.

Figure 3. Fitted HRs (median DBP as a reference) from cubic spline models for selected mortality causes.

Shaded areas represent 95% CI and vertical lines represent the 1st and 99th percentiles of diastolic blood pressure (DBP). Although similar, the upper X-axis applies to male DBP and the lower X-axis to female DBP.