Use of censored data methods to estimate blood pressure percentiles in US adults

Blood pressure (BP) levels are important modifiable risk factors for incident cardiovascular disease (CVD). However, many adults are on anti-hypertensive medication (MEDS) which lowers their BP but is a marker for increased CVD risk. To capture the association of BP and its treatment with risk, it is important to estimate BP levels in the absence of medication, i.e., Underlying BP. One proposed ad-hoc solution is to add a fixed increment e.g., 10/5 mm Hg to systolic (SBP)/diastolic (DBP) for people on MEDS to approximate their Underlying BP. However, what this increment should be, and whether it applies to all MEDS users is unclear. In this manuscript, we propose a novel time-to-event approach to estimate Underlying BP, treating BPs on MEDS as censored observations. The results indicate that the median Treatment Effect (i.e., Underlying BP - Observed BP) for SBP/DBP is 12.0/7.7 for men and 16.9/6.5 for women and is significantly modified by age, BMI and black race. The Underlying BP for MEDS users is then combined with the observed BP for non-MEDS users to derive age-sex-specific BP percentiles for Underlying BP based on a representative sample of normal weight U.S. adults aged 18-79 using NHANES data from 1999-2018.