Genetically Predicted On-statin LDL Response Is Associated With Higher Intracerebral Haemorrhage Risk

Ernst Mayerhofer; Rainer Malik; Livia Parodi; Stephen Burgess; Andreas Harloff; Martin Dichgans; Jonathan Rosand; Christopher D. Anderson; Marios K. Georgakis


Brain. 2022;145(8):2677-2686. 

In This Article


Baseline Characteristics

A total of 225 195 of the UKB participants had available genetic, primary care and outcome data, as well as no history of ICH at baseline and were thus eligible for inclusion in the analysis (Figure 1). Baseline characteristics and outcome data of participants included in the analyses are presented in Table 1. A total of 4 151 471 statin prescriptions for six statin agents were extracted from the primary care data. The majority of the prescriptions referred to simvastatin (60.8%) or atorvastatin (30.4%) and the number of statin prescriptions increased over time. We extracted at least one statin prescription for 75 973 of the participants (33.7%) with available primary care data. Of the statin users, 41% had prescriptions for two or more different drugs at different time points indicating a medication switch. The distributions of statin prescriptions over time, age, dose and presence of vascular risk factors are depicted in Figure 2 and the detailed distribution of different statins is shown in Supplementary Table 3. Of all prescriptions, 7.2% accounted for a low (<10 mg), 79.9% for a medium (≥10 and <40 mg) and 12.9% for a high (≥40 mg) atorvastatin equivalency dose with proportion of individuals prescribed a statin and statin dose increasing with age. Similarly, statin use and higher statin doses were more common among individuals with more vascular risk factors (active smoking, diabetes, hypertension, hypercholesterolaemia, prevalence of MI, stroke or PAD, or age >65 years) (Figure 2).

Figure 2.

Statin prescriptions in the UKB primary care data. (A) Number of statin prescriptions per year among 75 973 statin users. (B) Percentage of participant's receiving statins per participant's age at baseline. (C) Percentage of participants at baseline receiving statins per number of vascular risk factors (diabetes mellitus, hypercholesterolaemia, hypertension, active smoking, age >65 years). Statin intensity in B and C was classified as low (<10 mg), medium (≥10 mg and <40 mg) and high (≥40 mg) based on the atorvastatin equivalent dose according to the 2018 AHA guideline on cholesterol management.3 Estimated drug potencies were used to harmonize all drug doses to atorvastatin equivalent doses (see 'Materials and methods' section).

Genetic Score for On-statin LDL and LDL Trajectories in Primary Care Data

To validate the genetic scores for on-statin LDL lowering in the UKB, we extracted LDL measurements recorded in the primary care data. A total of 46 909 participants (62% of the total statin users) had at least one LDL value before and one after their first statin prescription. There were on average 8.1 ± 5.1 measurements spanning a total of 27.4 years (8.7 ± 4.2 years between the first and last measurements). The mean pre-statin LDL was 147.7 ± 38 mg/dl and the mean post-statin LDL 133.0 ± 42.7 mg/dl. The mean LDL decreased significantly over time [−3.45 mg/dl per year, 95% CI: (−3.47, −3.42)] among statin users. In a mixed linear model adjusting for age and sex, the genetic score and statin dose both had significant effects on absolute LDL levels [−2.3 mg/dl for each SD increase of genetic score, 95% CI: (−2.59, −2.00), Figure 3A, and −18.8 mg/dl for each SD of statin dose, 95% CI (−18.91, −18.66), Figure 3B and Supplementary Table 4]. Importantly, there was a significant interaction of the genetic score with time implying a more rapid on-statin LDL decrease among participants with a higher genetic score [−0.05 mg/dl per year for 1 SD of the genetic score, 95% CI (−0.07, −0.02)].

Figure 3.

Effect of the genetic score and statin dose on on-statin LDL levels (A and B) and lipid metabolites (C and D) among statin users. LDL trajectories derived from the primary care data of the UKB separated (A) by quartiles of the genetic score for LDL response after statin intake and (B) by mean statin dose over lifetime in 46 909 individuals with statin intake and at least one LDL measurement before the first statin prescription. (C) Estimate and 95% CIs of the effect of statin dose and the genetic score for statin response on total, LDL and high-density lipoprotein cholesterol (mmol/l) and on apolipoprotein B and A1 (g/l) derived from nuclear magnetic resonance among statin users. (D) Comparison of the effect size of the genetic score for statin response (1 SD increment) versus the statin dose (1 SD increment) on lipidomic metabolites among statin users. Each point represents the effect sizes for 1 of 228 lipidomic particles. Correlation coefficient r = 0.65. The results from C and D are derived from linear regression models adjusted for age, sex, PC1–10, race, kinship and genetic assay.

Because we found the genetic score to be also associated with absolute off-statin LDL levels among non-users [−3.0 mg/dl for each SD increase of genetic score, 95% CI: (−3.3, −2.8)], we tested the effect of each SNP and found four of them to be significantly associated with off-statin LDL levels (Supplementary Table 5). Thus, in a sensitivity analysis, we constructed an alternative genetic score with the remaining 31 SNPs that was no longer associated with off-statin LDL levels among non-users (P > 0.05), but was associated with significant on-statin LDL lowering among statin users [−0.03 mg/dl per year per 1 SD, 95% CI (−0.05, −0.01)].

Influence of the Genetic Scores on Lipid Particle Metabolites

Next, to explore whether a higher genetic score for on-statin LDL lowering mimics an exposure to higher statin intake at a metabolomic level, we investigated its effects on cholesterol measurements, as well as lipid particle metabolites, as assessed by standardized methodologies at baseline among statin users. The comparisons of the effect sizes of statin intake on nuclear magnetic resonance-assessed lipid particle metabolites are presented in Figure 3C and D, respectively. Of the 228 lipid particle metabolites analysed, the genetic score was significantly (Bonferroni-adjusted P < 0.05) associated with 161 and a higher statin dose with 97 (Supplementary Table 6). There was a correlation between the effect sizes of the genetic score and statin dose (r = 0.52, P < 0.001).

Genetically Predicted On-statin LDL Lowering and Risk of Incident ICH

Following the validation of the genetic score as a proxy of on-statin LDL lowering, we next tested associations with the risk of incident ICH among statin users (Figure 4). There were 679 incident ICH over an observation period of 2 514 994 person-years, yielding an incidence of 27 per 100 000 person-years. Over a mean follow-up of 11 years, 383 statin users developed ICH. In Cox proportional hazard models, higher genetic scores for on-statin LDL lowering were associated with a higher risk of incident ICH among statin users [hazard ratio (HR) 1.16, 95% CI (1.05, 1.28) for 1 SD difference]. Sensitivity analyses confirmed robustness of the findings among unrelated individuals [kinship coefficient <0.0884, n = 69 327; HR per SD increment 1.18, 95% CI (1.06, 1.31)] as well as with the alternative genetic score not influencing off-statin LDL levels [HR per SD increment 1.12, 95% CI (1.02, 1.24)]. Other variables associated with ICH risk were older age [HR: 1.07 per year, 95% CI: (1.05, 1.09)], lower body mass index [HR: 0.80 per 1 SD, 95% CI: (0.71, 0.90)], higher systolic blood pressure [HR: 1.18 per 1 SD, 95% CI: (1.07, 1.31)] and use of antiplatelet [HR: 1.30 95% CI: (1.06, 1.60)] or anticoagulant [HR: 3.47, 95% CI: (2.30, 5.24)] medications. We found no significant association between mean statin dose and ICH risk [HR: 1.07 per 10 mg atorvastatin equivalent dose, 95% CI: (0.99, 1.15)].

Figure 4.

Effect of on-statin genetically predicted LDL response on study outcomes. (A) Hazard ratios of statin dose and the genetic scores for statin response among statin users and non-users on ICH, MI and PAD. The results are derived from Cox proportional hazard models adjusted for age, sex, BMI, smoking status, history of diabetes, systolic blood pressure, mean statin dose, LDL levels, PC1–10, race, kinship and genotyping assay; use of anticoagulation and antiplatelet drugs at baseline in models for ICH. (B) Kaplan–Meier curves for survival ICH in statin users.

As positive controls, we tested the effects of genetically predicted on-statin LDL lowering on MI and PAD. After adjustment for cardiovascular risk factors, LDL levels and statin dose, we found significant associations of a higher genetic score with a lower risk of both incident MI [HR per SD increment 0.98 95% CI (0.96, 0.99)] and PAD [HR per SD increment 0.93 95% CI (0.87, 0.99)]. To reduce the risk for collider bias, we calculated models weighted for the inverse probability of being prescribed a statin, which yielded effect estimates in the same direction for both MI [HR per SD increment 0.96 95% CI (0.93, 0.99)] and PAD [HR per SD increment 0.84 95% CI (0.70, 0.99)]. Finally, to minimize the possibility that the observed effects are the result of pleiotropy on traits other than on-statin LDL, we also tested the same associations among statin non-users and found no significant effects on ICH, MI or PAD (all P > 0.05).