Added predictive value of Lipoprotein(a) on top of the SMART2 risk model in patients with atherosclerotic cardiovascular disease
T H Reitsma, E B Haug, M N Gynnild, T P Van de Hoef, M Teraa, E J van Dijk, B O Asvold, A T R Bayyoudtamimi, Y Khalil, J Ebbers, S H J Hageman, F L J Visseren, J A N DorresteijnAbstract
Background
Lipoprotein(a) (Lp(a)) is a genetically determined lipoprotein and an independent causal risk factor for atherosclerotic cardiovascular disease (ASCVD). Among patients with established ASCVD, elevated Lp(a) contributes to residual cardiovascular (CV) risk. However, Lp(a) is not incorporated into contemporary risk models used to estimate recurrent CV risk, such as the SMART2 model, potentially underestimating risk in individuals with (very) high Lp(a) levels.
Purpose
To assess the added predictive value of Lp(a) on top of the existing SMART2 model.
Methods
Individuals with established ASCVD from the ongoing prospective UCC-SMART cohort were included. The original SMART2 model and an extended model including Lp(a) as a continuous linear variable were fitted using Fine–Gray models to account for competing risks. The outcome was aligned with the original predicted outcome of the SMART2 model, which estimates the 10-year risk of recurrent CV events, and was defined as a composite of non-fatal myocardial infarction, stroke or CV death. Model performance was assessed by calibration and change in C-statistic (ΔC-statistic). The absolute difference in predicted 10-year risks was plotted against Lp(a) levels to evaluate clinical relevance.
Results
Among 4,559 individuals, 944 (21%) experienced a recurrent CV event during a median follow-up of 10.7 years (interquartile range 2.5 – 14.8). The subdistribution hazard ratio per 500 mg/L increase in Lp(a) was 1.02 (95% confidence interval [CI] 0.94 – 1.10). The extended SMART2 model including Lp(a) did not improve discrimination (ΔC-statistic −0.0003; 95% CI −0.00103 to 0.00050) and showed no visible alteration in calibration (Figure 1). Absolute changes in predicted risk increased progressively with higher Lp(a) concentrations (Figure 2). Clinically relevant down-classification was uncommon, as absolute reductions in predicted risk were generally small (<0.5%). In contrast, up-classification was more pronounced after integration of Lp(a) on top of SMART2, reflecting the right-skewed distribution of Lp(a). Overall, 44 (1%) individuals showed an absolute increase of ≥1% in predicted recurrent CV risk, mainly among those with very high Lp(a) levels (above 1000 mg/L) and higher baseline SMART2 risk (Figure 2).
Conclusion
Addition of Lp(a) to the SMART2 risk model did not improve prognostic performance at group level. However, in individuals with Lp(a) levels above 1000 mg/L, addition of Lp(a) may lead to a potentially clinically relevant increase in predicted risk.