Conflation of several diseases to make the new FH prevalence:

(For a larger explanation of the confusion between FH and FDB, please see my other report here.)

How can one increase a prevalence rate without having to find more people?


It is a linguistic strategy, not a scientific study. The claim of higher prevalence is false.

Again, the issue here is not whether or not those with other diseases should be given medication indicated for FH, or whether all should now be called “FH” or whether the former disease names should be discarded. That’s a question for the medical community and the FDA. What concerns us here is whether or not new people have been found. Is there a higher prevalence than previously thought? No, there is not. That is false and a misrepresentation of both the gravity of the situation and the addressable market of the drugs in question. Once the linguistics is understood and the relevant math laid out in rows and columns, Regeneron’s prevalence estimate for FH-as-LDLR is actually lower than the established rate. Other diseases, as before, have their own prevalence. If we decide to blend them linguistically under one name, we do the required math. That’s it.

Imagine that I count 200 people. 100 of them have LDLR mutations and are in a room with “FH” painted on the door, and the other 100 have APOB mutations and are in another room with “FDB” painted on its door. If I herd the 100 people from the “FDB” room to the “FH” room, and then whitewash over “FDB,” leaving that room empty, then do I have 100 more people than previously thought? No. I still have a total of 200. Do I have more “FH” than previously thought. Perhaps. Sort of. But it would be the expansion of the previous linguistic definition of “FH” into something new. It would not however be a “discovery” in the explorer’s or scientist’s sense of the term, certainly not the discovery of more diseased patients. Anyone who thinks otherwise would have serious difficulty barring a professor of linguistics from entering the debate. Inside the old room with “FH” still painted on the door we still have 100 with LDLR mutations and 100 with APOB mutations.

Why do this?  There is some confusion about what FH is and what it isn’t. And one will find FDB and FH both referred to as FH.  Strictly speaking, however, FH and FDB are two different inherited diseases.  (The emphasis is mine.)

FDB is a disorder which is clinically and biochemically indistinguishable from familial hypercholesterolemia (FH), a disease caused by LDL receptor gene mutation. This was demonstrated by the fact that approximately 3-5 % of FDB patients are incorrectly diagnosed as FH (Weisgraber et al. 1988). However, reviews dealing with the comparison between FH and FDB homozygotes and heterozygotes showed that hypercholesterolemia, which arises from the genetic condition, is generally milder and more variable in FDB (Miserez and Keller 1995). Furthermore, the development of atherosclerosis is delayed in comparison with FH patients (Brousseau et al. 1995, Tybjaerg-Hansen et al. 1998, Če.ka et al. 2000). ~ “Major Apolipoprotein B-100 Mutations in Lipoprotein Metabolism and Atherosclerosis,” M. VRABLÍK, R. ČE.KA, A. HOŘÍNEK

The Regeneron report, like the Danish reports, blends FH and FDB due to available linguistic ambiguity and then refers to the natural mathematical result as a larger FH prevalence count.

Controversial FDB mutation p.Arg3558Cys

The established prevalence of FH-as-LDLR mutation is 1:500. Of FDB-as-APOB mutation, it is 1:1,000.  If I combine FH and FDB, I mathematically derive a prevalence of 1:333. If I call these combined patients “FH,” have I really found more patients or have I simply put two formerly distinct diseases under a single umbrella-term, “FH,” which then requires that I follow through with the required math? What if I add in FH3, which refers to PCSK9?  I can also add in a controversial APOB mutation into FDB.  P.Arg3558Cys, AKA, R3531C, Arg3531Cys,[1] is found to interfere with the cholesterol process, in the lab. In the living, however, it has been said to be too weak to be included. Vrablok reveals the point of confusion: in vitro versus in vivo. (Emphasis mine.)

“These are mutations leading to amino acid substitution at positions 3500 (R3500Q and R3500W) and 3531 (R3531C) that have been shown to decrease the binding affinity of apoB-100 in vitro. However, only the former mutations have been unequivocally demonstrated to cause hyperlipidemia in vivo.” ~ VRABLÕK [1]

Here is Tybjaerg-Hansen and Nordestgaard (who will later co-author the Danish studies, with financial influence from Big Pharma):

“In contrast, the Arg3531Cys mutation, which is just as common, is not in itself associated with hypercholesterolemia or an increased risk of ischemic heart disease.” ~ TYBJÆRG-HANSEN, et al. [1]

Here is the title of a study which focused on this issue:

“R3531C mutation in the apolipoprotein B gene is not sufficient to cause hypercholesterolemia.” ~ Rabes JP, et al. [1]

In short, p.Arg3558Cys, AKA R3531C, was not counted in previous FDB prevalence estimates. (click here.)

Below, with Dr. Rader’s responsible breakdown of the specific prevalence rates, we can clearly see the established rates of the individual diseases. We’ll combine them and do the required math.

The prevalence of FH is explained conflation

We can now say, “Our addressable market is twice what we previously thought it was.” How clever.  This is a culturally broader definition of “FH,” not an addition of newly discovered patients. The total of FH + FDB + FH3+ p.Arg3558Cys does not change and neither do any of these constituents when teased back out. There is nothing new here. Before the necessary math, FH-as-LDLR was 1:500 and after the math, FH-as-LDLR is still 1:500. Here’s the recent study funded and staffed by Regeneron, published in Science magazine.

Ironically, in the very act of conflating Non-LDLR with LDLR, and calling the result “FH” the prevalence for the previous LDLR-based definition of FH is actually supported: 1:500. (50,726 ÷ 98 = prevalence of 1 in 518.) In the act of declaring what is actually a mere linguistic victory, the old, established prevalence of FH-as-LDLR is silently supported.[2]

Classic FH prevalence

Regeneron’s “new” prevalence, with conflation, is between 1:222 and 1:256. The “old” prevalence, after factoring in that conflation, is 1:232. The new prevalence is not really different from the old prevalence.

The difference in Regeneron’s “higher” prevalence is not found in an addition of new people, but in the subtraction of a necessary explanation of this linguistic maneuver.

This is not a prevalence study. It is a linguistic shell game.  But the FH numbers deteriorate even further. Once we remove the cohort inflated by ascertainment bias (a form of selection bias), we see that what Regeneron declares as a higher prevalence, is actually lower, due to two factors.

First, let’s look at what was disclosed. A 6,747-cohort inflated by selection bias was inserted into the results. Its prevalence was almost twice that of the total results’ prevalence. Thus, after the removal of the bias, the prevalence comes down to 1:256.

ascertainment bias
The old and new FH prevalence
FH Prevalence results

Per million, basic math required by the linguistic conflation, using the already established rates yields 4,310 and this is clearly higher than the 3,906 in Regeneron’s results.  Even if we give Regeneron’s and the established prevalence a 10% margin of error, the resulting parity alone shows that the claim of higher prevalence is false.  Regeneron’s result would be precisely what was previously thought, not higher.

Second, we’ll look at what was not disclosed: the breakdown of APOB-to-LDLR after the 6,747 are backed out of the results. It is highly likely that the LDLR will be removed at a much higher rate than the APOB, [3] further distorting the already inverted ratio of APOB-to-LDLR.

Let’s run through the reasoning. In this case, we begin with a biased total and we are calculating the LDLR count after the removal of the bias. To illustrate the force at work here, consider the following example. If I give a customer a bag of pure popcorn and another bag, half-popped, without a detailed explanation, then I can tell the customer that the prevalence of un-popped kernels of both bags combined is 1:4. If I then take back the bag with pure popcorn, the prevalence of un-popped kernels in the remaining bag is 1:2. Without an explanation of the breakdown in the original bags, the customer will naturally believe that the prevalence of un-popped kernels at 1 in 4 still applies to his snack. It does not. The force of the selection bias, in reverse, increases the proportion of un-popped kernels in the customer’s possession, and the customer is left, so to speak, holding the bag.

This same force is exerted on the Regeneron prevalence estimate when one of the acknowledged selection biases is reversed back out. The health consequences of the mutations in question are more variable than big pharma wants us to believe.The clinical scoring systems themselves, due to this variability, result in an inherent selection bias.As we can see below, the passing scores are in actuality a minority of total mutations and of those that do pass, there is a predominance of LDLR over APOB. This is because, as is widely known, APOB is weaker than LDLR.

Danish Study, mutations causative of familial hypercholesterolemia

The 6,747-cohort inflated by selection bias was inserted into the Regeneron results. Although there is a different circumstance behind this selection bias, because both instances6 of the bias filter out weaker cases both will result in collecting the more severe mutations, and that would be mostly the LDLR. So the proportion of LDLR-to-APOB between the two instances [4] of bias will differ, but precedent being our guide, the principle will hold: the stronger LDLR will predominate, and the weaker APOB will be far less represented.  Consequently, in the net 43,979, there will be a disproportionate loss of LDLR from the results as this LDLR-inflated group is taken away. Prevalence for FH-as-LDLR will be even lower, and the APOB/LDLR ratio will be even more at odds with other studies. With this biased cohort in place, FH-as-LDLR prevalence was 1:518 – in line with established estimates. After removing this biased cohort, it follows by deduction, given the preceding points, that Regeneron’s study does not actually have a higher FH-LDLR prevalence than the established 1:500, and not even equal, it is actually lower than previously thought.[5]

[1] The established FDB prevalence is also silently proven. Click here.

[2] Click here for these and other references to the exclusion of p.Arg3558Cys, AKA, Arg3531Cys and R3531C. Click here for p.Arg3558Cys equivalents.

[3] This assumes the Amish were not overrepresented in this group.

[4]Both instances” and “two instances of bias” = the 6,747-cohort in the Regeneron study and the bias inherent in DLCN scoring as demonstrated in Benn’s 2nd report.

[5] See also this page.