A similar prevalence tactic: Injecting a sliver of inflated results

The Regeneron report added in a group from a catheterization laboratory. This was disclosed, along with a partial disclosure of resulting distortions through “Ascertainment Bias,” which is a form of selection bias.

The Regeneron report added in a group from a catheterization laboratory. This was disclosed, along with a partial disclosure of resulting distortions through “Ascertainment Bias,” which is a form of selection bias. In brief, it is only natural that a population of patients already selected for clinical treatment will tend to have a higher prevalence of illnesses. And if that clinic is related to cardiovascular problems, then naturally it will have a higher concentration of those with the underlying causes of the problem, including FH. Sure enough, when the 6,747 from the cardiac catheterization laboratory were excluded, the prevalence of FH among the less[1] selected population became 1:256.  When isolated, the FH patients from the lab had a prevalence of 1:118.  This small group, only 13% of the 50,726, accounts for the distortion of the final result, from 1:256 to 1:222. 

One problem with the disclosure is that it is not complete. We know that the 6,747 lab patients accounted for 57 of the mutation hits in the overall study. But we don’t know how these are broken down. For example, how many are LDLR and how many are APOB? And how many of the APOB are p.Arg3558Cys?

LDLR APOB and PCSK9 breakdown

This is important because p.Arg3558Cys are included in the 50,726 in spite of their controversy: they are milder and some scientists have declared that despite laboratory findings (in vitro) they are not strong enough in living subjects (in vivo) to be considered to have hypercholesterolemia. (Click here.)

More LDLR than APOB with higher scores

Also, even the standard APOB mutations are weaker than LDLR, as can be seen in the screenshot on the right taken from the 2nd report.[2] The results in the table on the right are from a random selection of an unselected population and are then later subjected to the selection bias of a clinical scoring system. We don’t expect the LDLR-APOB proportions to match up with the above Regeneron selection biased result in the 6,747. We use them here only to illustrate the principle that the APOB are milder, and will probably result in fewer carriers in a clinical setting than the much stronger LDLR mutations. From this, it is not beyond reason to estimate a disproportionate loss of LDLR verses the APOB when the mutation-rich population of 6,747 is eliminated.[3] Or, another way of saying this is that there were probably fewer APOB among the 6,747 patients in the lab to begin with, and in fact, there would be very few, if not a complete absence of the very mild p.Arg3558Cys, in vivo.

And so when we subtract the inflated 6,747 from the 50,726 we also subtract more LDLR than APOB.23 Thus, it is probable that among the less selected 43,979 remaining, an even greater disproportion of APOB over LDLR would remain.[4]  It is widely known that in the general population APOB is rarer than LDLR. This can be flipped however if one concentrates solely on whites of European descent, then adds in a sliver of those with Founder Effect (the Amish of Lancaster, PA), and also skips over the controversy surrounding p.Arg3558Cys, including them.


[1] As noted by the authors, other distortions from selection bias still remain after the removal of the 6,747. However, the most notable instances are the inclusion of the Amish of Lancaster, PA and the exclusion of non-whites.

[2] Click here for a list of reports.

[3] This assumes the Amish were not overrepresented in the 6,747.

[4] Click here for another explanation of the mathematical consequences of reversing a selection bias.