Intensive antihypertensive therapy was associated with worse outcomes among smokers in an unplanned secondary analysis of SPRINT (Systolic Blood Pressure Intervention Trial) data, researchers said.
Current smokers with baseline systolic blood pressures greater than 144 mm Hg treated to a target of <120 mm Hg had significantly higher rates of myocardial infarction, stroke, and other cardiovascular events than those assigned to more conventional treatment, the analysis showed.
Intensive treatment was also associated with a higher incidence of acute kidney injury in smokers within the SPRINT cohort, wrote Aaron Baum, PhD, of Icahn School of Medicine at Mount Sinai, New York City, and colleagues in JAMA Network Open.
In a interview with MedPage Today, Baum said the study — using supervised learning algorithm random forest-based analysis — was designed to test the hypothesis that not all subgroups benefit from intensive blood pressure control, as the overall SPRINT trial findings suggested.
He noted that even in large intervention trials like SPRINT, traditional biostatistics techniques often fail to identify heterogeneous treatment effects in subgroups because they often examine just one characteristic at a time.
“Random forest analysis is particularly good at looking at combinations of characteristics,” he said.
The original SPRINT results suggested that lowering systolic pressure to <120 mm Hg was more beneficial than modest control (<140 mm Hg) in adults without diabetes for reducing CVD event risk. Findings and other aspects of the trial, including the interpretability of its automated blood pressure measurement, have proven controversial.
Post hoc analyses have also indicated that intensive blood pressure control was associated with kidney injury in SPRINT and in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial.
“It is common that treatments have variable benefit and off-target effects in a population, but it is often difficult to identify subpopulations that are optimally targeted by an intervention. Traditional subgroup analyses typically fail to identify these heterogeneous treatment effects (HTEs) because the analyses are limited by multiple testing concerns, estimation bias, and prespecified univariate covariate testing, but recent statistical advances in machine learning permit detection of HTEs in large populations with many covariates,” Baum and colleagues wrote.
The researchers performed an “exploratory, hypothesis-generating, ad hoc, secondary analysis” of data from the SPRINT trial. Out of the original SPRINT cohort, Baum and colleagues found 466 who were smokers with high blood pressure at enrollment. Half were assigned to a training dataset with the rest reserved for testing. About 60% of the smokers were male; mean age was about 61.
The exercise yielded a number-needed-to-harm of 43.7 for intensive treatment to cause one major cardiovascular event, and a hazard ratio of 10.6 (95% CI 1.3-86.1) relative to standard therapy.
“Further research is needed to directly evaluate intensive blood pressure control in hypertensive smokers to better characterize the potential tradeoffs in this population,” they wrote. “If the treatment effect heterogeneity identified in the present study holds in the general population, it would forecast an additional 88,700 cases per year of acute kidney injury and 56,100 cases per year of hypotension should the SPRINT intensive blood pressure regimen be adopted.”
The practice of splitting the data used in the analysis into separate training and testing sets was cited by the researchers as a potential study limitation that increased the chances of spurious results.
Baum told MedPage Today that while the split data approach has significant potential for identifying heterogeneous effects from randomized trials, it remains to be seen which algorithm or algorithms will prove most useful.
In an editorial published with the study, Benjamin A. Goldstein, PhD, of Duke University in Durham, North Carolina, and Joseph Rigdon, PhD, of Stanford University in Palo Alto, California, wrote that the study by Baum and colleagues “highlights the role that machine learning can play in the analysis of clinical trials.”
“One of the promises of personalized medicine is that treatment will be tailored to one’s particular set of clinical characteristics. Randomized clinical trials — as the current criterion standard for evaluation of treatment effects — have an important role to play in the realization of that vision. While forest plots are a reasonable first attempt to detect treatment heterogeneity, one should not enter into such an analysis hoping to conclude that the average treatment effect is sufficient,” they wrote.
They added that a paradigm shift is needed “where we embrace the underlying treatment heterogeneity and hope to discover subgroups who may or may not benefit from the therapy.”
“Methods like random forest analyses should be embraced as they provide investigators with tools to find such effects. While this will ultimately make the provisioning of therapy more challenging, it also has the potential to make it more effective,” Goldstein and Rigdon concluded.
Funding for this research was provided by the National Institute of Mental Health, the National Science Foundation, and the National Institute on Minority Health and Health Disparities.
The researchers reported no relevant relationships with industry related to this study.