Determinants of individualized attenuation of carotid intima-media thickness progression following pictorial visualization of subclinical atherosclerosis

Background Visualizing subclinical atherosclerosis has emerged as a strategy to improve cardiovascular risk perception and preventive adherence. While the benefit from pictorial risk communication previously has been shown to vary in subgroups, these analyses have been restricted to group-level comparisons and have considered one effect modifier at a time. We aimed to identify determinants of the individualized response to pictorial visualization of subclinical atherosclerosis by simultaneously evaluating multiple potential effect modifiers. Methods This secondary analysis of the VIPVIZA (VisualiZation of asymptomatic Atherosclerotic disease) randomized controlled trial included 3,135 participants with carotid ultrasound data at baseline and 3-year follow-up. Participants were randomized to an intervention group (pictorial representation of carotid ultrasound plus a nurse phone call to confirm understanding) or a control group (not informed). The primary outcome was 3-year change in carotid intima–media thickness (cIMT). Intervention effect heterogeneity, defined as difference in treatment effect magnitude between individuals, was assessed using a causal forest machine learning framework, enabling simultaneous estimation of individualized effects conditional on all baseline covariates. Furthermore, variable importance scores were calculated to identify the baseline characteristics that most strongly drove the heterogeneity in beneficial response. Results The intervention by pictorial risk communication significantly attenuated cIMT progression compared with the control group. Calibration analysis demonstrated good agreement between predicted and observed treatment effects (slope for mean prediction 1.05, p = 0.002; slope for differential prediction 1.48, p = 0.026), supporting the presence of meaningful treatment effect heterogeneity. Baseline cIMT was the strongest determinant of benefit, with greater baseline values associated with larger reductions in progression. Metabolic markers, specifically waist circumference, Body mass index (BMI), triglycerides and fasting glucose, were major drivers of heterogeneity, whereas age and sex were relatively minor determinants of the intervention response. Conclusions This causal forest analysis demonstrates that the effect of pictorial risk communication on structural atherosclerosis is highly individualized. Asymptomatic individuals with greater carotid wall thickness and certain metabolic characteristics, such as higher BMI, adverse lipid levels, increased waist circumference, and elevated fasting glucose, experienced greater benefit from pictorial risk communication in preventing progression of subclinical atherosclerosis. These findings highlight the importance of individualized cardiovascular risk communication strategies and demonstrate the use of machine learning approaches to identify individuals most likely to benefit from preventive interventions.Figure 1Figure 2