Mendelian Randomization is not appropriate for LDL-C. It fails the restrictions of the method.
In Ference's figure 5a, they cherry-picked the papers so that they would have an association.
"According to Ference et al. [3], the most compelling clinical evidence for causality is provided by ‘the presence of more than 30 randomized cholesterol-lowering trials that consistently demonstrate that reducing LDL-C reduces the risk of CVD events proportional to the absolute reduction in LDL-C’. As previously noted, this is not true exposure–response. Furthermore, in their Figure 5(a), that illustrates the association, the authors have only included data from 12 of the 30 trials they refer to. If all of the trials in Table 1 are included, as we have done in Figure 3, there is no association between LDL-C lowering and coronary event rate."
Ravnskov, U., Lorgeril, M. de, Diamond, D. M., Hama, R., Hamazaki, T., Hammarskjöld, B., Hynes, N., Kendrick, M., Langsjoen, P. H., Mascitelli, L., McCully, K. S., Okuyama, H., Rosch, P. J., Schersten, T., Sultan, S., & Sundberg, R. (2018). LDL-C Does Not Cause Cardiovascular Disease: A Comprehensive Review of the Current Literature. Expert Review of Clinical Pharmacology, 11(10), 959–970.
doi.org/10.1080/17512433.201…
Here you go, this represents hundreds of thousands of study subjects in rigorous well designed trials. Ference et al Eur Heart J 2017. Everyone can plot their own "average" risk on this graph. 1 mmol/L = 38.67, so 2.5 mmol/L is about 100 mg/dL. Bottom line; risk for CVD at population level (not individual level per se) is proportional to "walking around" LDL-C, but risk is higher if you had a prior event. Of course the issue is one cant predict risk with certainty at individual level. If we could then we would not need to treat so many people. Maybe some day we will be better able to define risk at individual level to "personalize" therapies, but I suspect not in next 1-2 decades.