Prognostic value of high-sensitivity C-reactive protein versus low-density lipoprotein cholesterol for recurrent atherosclerotic cardiovascular disease events using US real-world data

Introduction Individuals with atherosclerotic cardiovascular disease (ASCVD) are at risk of recurrent cardiovascular (CV) events. High-sensitivity C-reactive protein (hsCRP) is an inflammatory marker that may have predictive value for prevention of recurrent CV events beyond traditional biomarkers, such as low-density lipoprotein cholesterol (LDL-C). Purpose To evaluate the prognostic value of hsCRP relative to LDL-C for recurrent CV events in individuals with established ASCVD, using machine learning (ML)-derived covariates and survival analysis based on US real-world data. Methods This was a retrospective observational study using the US TriNetX electronic health records and claims data. Included were individuals ≥18 years of age with diagnosis codes for ASCVD between 2010 and 2023 and an eligible hsCRP measurement (index date) at least 30 days after ASCVD diagnosis. hsCRP measurements >10 mg/L or measurements taken in relation to suspected infection or autoimmune disease were excluded. Outcomes of interest were all-cause mortality, ASCVD event recurrence (new ASCVD diagnosis or procedure code in conjunction with a hospitalisation event), the composite of ASCVD event recurrence or all-cause mortality, and 3-point major adverse CV events (non-fatal stroke, non-fatal myocardial infarction or all-cause mortality); outcomes were assessed within a 1-year and a 3-year follow-up period. Cox regression and Fine–Gray competing risk analysis were used to evaluate outcomes in binary (hsCRP ≥2 mg/L vs <2 mg/L; LDL-C ≥55 mg/dL vs <55 mg/dL) and quartile (Q) linear trend (Q1 to Q4 for hsCRP and LDL-C) analyses, using either standard covariates of established CV risk factors or covariates identified through an ML process using SHAP-based feature selection algorithms. In addition, analyses were performed in an internal validation cohort (a subset with available LDL-C measurements within 30 days of index date). Results Of 250,591 individuals with ASCVD in the database, 37,639 individuals fulfilled all eligibility criteria. In binary and quartile trend analyses, elevated hsCRP was associated with a significantly increased (p<0.05) risk of all outcomes in year 1 and year 3 (Table 1). No consistent association between elevated LDL-C and risk of outcomes was apparent (Table 2). The results were not affected by covariate selection methods (standard or ML). In the validation cohort (n=22,193), similar trends were observed (Tables 1 and 2). Conclusions In this US real-world study of individuals with established ASCVD, hsCRP consistently predicted recurrent CV events and mortality. Measuring CV inflammation (hsCRP ≥2 mg/L) could be a potentially valuable means of CV risk prediction in individuals with ASCVD. A key strength of this study is its use of real-world data, alongside a data-driven ML approach to covariate selection, both of which enhance the clinical relevance and robustness of the findings.