The SPRINT trial suggested a more aggressive blood pressure-lowering target reduced the rate of cardiovascular events in hypertensive people, but a new secondary analysis of the trial indicates such an approach may actually be harmful for smokers.
Lead author Joseph Scarpa, MD, PhD, and colleagues used the machine learning method of random forest analysis to predict treatment associations among subgroups in the study, which compared treatment to a modest systolic blood pressure (SBP) target (less than 140 mm Hg) against a more aggressive one (less than 120 mm Hg). The researchers used half of the trial data to identify patient subgroups that may have been harmed with intensive blood pressure control, and then used the other half to validate those hypotheses.
A total of 466 SPRINT participants were current smokers with an SBP above 144 mm Hg at baseline. They were 60.7 years old on average and 61.4 percent were men.
Among the smokers who comprised the testing cohort, 10.9 percent of those in the more intensive therapy group experienced the primary composite outcome of acute coronary syndrome, stroke, heart failure or cardiovascular death. Only 4.8 percent of those undergoing standard treatment met one of these endpoints during the median 3.3 years of follow-up.
The hazard ratio for the intensive versus normal treatment group was 10.6 and the number needed to harm was 43.7 people to cause one event in this subgroup. Likewise, the risk of an acute kidney injury event was higher among smokers who were treated to the lower SBP target versus the higher target (10 percent vs. 3.2 percent).
“These findings are consistent with descriptive evidence from the Hypertension Optimal Treatment study, which also suggested that aggressive diastolic blood pressure lowering in smokers may increase risk for cardiovascular events,” Scarpa et al. wrote in JAMA Network Open.
“Because both smoking and hypertension have harmful effects on high-pressure vasculature, it has been suggested that patients with both risk factors may be more sensitive to a significant reduction in perfusion pressure. Lowering diastolic blood pressure in other cohorts with reduced arterial compliance and elasticity, such as elderly hypertensive populations, has also been noted to induce harm.”
The authors pointed out traditional subgroup analyses often fail to identify heterogeneous treatment effects (HTEs) “because the analyses are limited by multiple testing concerns, estimation bias, and prespecified univariate covariate testing,” but recent statistical advances may help tease out HTEs in large populations.
“This work highlights the role that machine learning can play in the analysis of clinical trials,” Benjamin A. Goldstein, PhD, with Duke University School of Medicine; and Joseph Rigdon, PhD, with Stanford University School of Medicine, wrote in a related editorial. “Finding effect heterogeneity is a notoriously difficult statistical problem.
“Nonetheless, it is of great research interest to fully interrogate clinical trials to understand how effects may look in different subpopulations. As the authors show, machine learning can aid in these assessments.”
Scarpa and co-authors noted their results should be considered hypothesis-generating and interpreted cautiously because of splitting the data set in half to form a validation cohort limited their statistical power, among other limitations.
“While this is inherently a post hoc analysis, given the degree of validation performed, the results also should not be easily dismissed,” Goldstein and Rigdon wrote. “Methods like random forest analyses should be embraced as they provide investigators with tools to find such effects (HTEs). While this will ultimately make the provisioning of therapy more challenging, it also has the potential to make it more effective.”