Introduction
The SPRINT study demonstrated that intensive systolic blood pressure (SBP) treatment consistently and significantly reduces the primary outcome by 25%. However, a recent secondary analysis by the SPRINT Group evaluated whether this protective effect persists post-trial. By applying a linear mixed-effects model to repeated SBP measurements, they found the intervention lost its effect, citing a loss of SBP control during the post-trial period. Our group previously showed that this protection vanishes at different follow-up stages depending on the subgroup analyzed. This study explores why results diverge when using the Cox proportional hazards model versus Joint Model Analysis (cJM) within the trial. We propose that Simpson’s paradox explains these divergent findings.
Materials and methods
We analyzed 9,068 participants with 128,139 repeated SBP measurements, randomized to either intensive treatment (SBP <120 mmHg) or standard treatment (SBP 130–139 mmHg). We employed a cJM to integrate longitudinal SBP data with the Cox model. Additionally, we evaluated the causal association of Serious Adverse Events (SAEs), blood pressure variability (BPV), and cumulative SBP with a composite outcome of myocardial infarction, acute coronary syndromes, heart failure, stroke, and cardiovascular mortality. We performed all analyses using R Studio v.12.0 and STATA v.19
Results
The intensive treatment’s protective effect lost significance during the intra-trial follow-up across all groups: the total population (3.4 years), women (1.1 years), and patients with cardiovascular disease (1.3 years), chronic kidney disease (1.3 years), Black race (1.8 years), or age under 75 (2.1 years). Participants experiencing SAEs lost the protective effect earlier (3.4 years) than those without SAEs (4.2 years). Intensive treatment significantly increased systolic BPV. Furthermore, SAEs in the intensive group tripled the risk of the primary outcome compared to the standard group.
Conclusion

In the context of survival analysis, we postulate that the SPRINT trial suffers from a Longitudinal Simpson’s Paradox, where temporal aggregation and the omission of endogenous time-varying mediators (SAEs, BPV) generate a marginal Hazard Ratio (HR) that is highly misleading regarding the true causal effect at the individual level.