Smokers with baseline SBP >144 mmHg may experience harm with intensive BP lowering
Assessment of Risk of Harm Associated With Intensive Blood Pressure Management Among Patients With Hypertension Who Smoke A Secondary Analysis of the Systolic Blood Pressure Intervention Trial
Introduction and methods
A lower systolic blood pressure (SBP) target of <120 mmHg in individuals without diabetes resulted in increased benefit compared to a more moderate target of <140 mmHg, demonstrated in the Systolic Blood Pressure Intervention Trial (SPRINT) . These data, in combination with evidence from meta-analyses [2,3], suggest that intensive BP lowering is largely beneficial and has been suggested as an optimal treatment goal. However, there have been discussions about the limitations of SPRINT including the performance of BP measurements and the generalizability to other hypertensive populations .
Of note, SPRINT, as well as the Action to Control Cardiovascular Risk in Diabetes Trial, also reported significant adverse events, including renal insufficiency and acute kidney injury episodes, associated with intensive BP lowering [5,6], which may limit long-term adherence to intensive BP control.
SPRINT enrolled individuals with an SBP between 130-180 mmHg and increased CV risk, without diabetes or stroke between Nov 2010 and March 2013. Participants were assigned to SBP target of <120 mmHg (intervention) or <140 mmHg (control). Intensive BP lowering was associated with reduction in primary outcome (composite outcome of MI, other acute coronary syndromes, stroke, HF or death from CV causes) with HR 0.7; 95%CI: 0.5-0.9; P<0.05.
It is important to identify subgroups of patients who poorly respond to or even are at risk of harm from BP lowering treatment, so clinicians can provide personalized, safe and cost-effective treatment. Traditional subgroup analyses do not always identify heterogeneous treatment effects (HTEs), but recent advances in machine learning can better detect HTEs in large populations with covariates .
In this exploratory, hypothesis-generating, ad hoc, secondary analysis of SPRINT, a machine learning method (random forest analysis) was used to investigate heterogeneity in CV effects of intensive BP control. Primary outcome was the SPRINT primary outcome (composite outcome of MI, other acute coronary syndromes, stroke, HF or death from CV causes). Median follow-up was 3.3 years (IQR: 2.7-3.8).
- Of 9361 participants, 466 (5%) were current smokers with SBP>144 mmHg at baseline.
- The random forest algorithm identified 1 subgroup of smokers with SBP>144 mmHg at baseline with higher number of primary outcome in the treatment group compared to the control group (10.9 vs 4.8%, HR 10.6, 95%CI:1.3-86.1, P=0.03). Number needed to harm was 43.7 to cause 1 event.
- There was no significant effect for separate groups of either current smokers or those with SBP>144 mmHg at baseline.
- Using the full 3-way interaction model resulted in an HR on the triple interaction term ‘treated, baseline current smoker, baseline SBP >144mmHg’ of 2.0 (95%CI:1.1-3.7, P<0.05).
- The subgroup of smokers with baseline SBP >144 mmHg was more likely to experience acute kidney injury in the intensive BP group (10.0%) vs control group (3.2%) (HR:9.4, 95%CI:1.2-77.3, P=0.04).
- Interestingly, reduction in mean arterial pressure across treatment and control participants was greater in the subgroup of smokers with baseline SBP >144 mmHg than the rest of the cohort (-8.7; 95%CI:-15.0 to -2.3, P=0.01) driven by a sustained and pronounced reduction in DBP (~2-fold) than SBP.
This exploratory analysis of SPRINT data using a machine learning method, revealed a subgroup of current smokers and baseline SBP >144 mmHg that may experience harm with intensive BP lowering. More specifically, this group had a higher rate of primary outcome and higher incidence of acute kidney injury with intensive BP lowering.
This analysis demonstrated that a machine learning method can reveal unexpected HTEs within an existing RCT to identify potential hypotheses, that need further testing. In addition, these findings suggest although benefit is observed in randomized clinical trials in the overall cohort, treatment may have unwanted effects in population subgroups. The identification of subgroups is important for clinicians to avoid unnecessary harm in high-risk groups.