Merck’s $800M bet on Sir Austin Bradford Hill

Why Merck’s CETP Inhibitor programme is such a huge gamble – and why DrugBaron fears the odds are stacked against them.

Few programmes in the pharmaceutical industry were born with such optimism as the search for CETP inhibitors.  Decades of epidemiology had already pointed to the importance of HDL-cholesterol (the so-called “good cholesterol”) as a predictor of future risk of heart disease, but no-one really knew how they might raise HDL with a drug.  That changed in 1990, with a New England Journal of Medicine paper describing a Japanese family with a mutation in the CETP gene that abolished CETP expression and dramatically raised HDL.

Despite challenges finding a CETP inhibitor, Pfizer eventually got a CETP inhibitor, torcetrapib, into the clinic and the results were astounding: not only was HDL-cholesterol doubled, but LDL-cholesterol (the “bad cholesterol” lowered by statins) was reduced.  Against this background, the dramatic failure of the resulting Phase III trial of torcetrapib in 2006 hit the cardiovascular field like an earthquake.   Not only was torcetrapib ineffective, it actually increased heart attacks and strokes.

After years of soul-searching, Merck, who were following behind with anacetrapib, decided to continue development.  Their assessment was simple: the rationale for HDL elevation was so good that their had to be another explanation for the failure of torcetrapib.  That something, perhaps, was the hypertensive effect of the Pfizer molecule that wasn’t a class-wide liability.   Since hypertension is associated with increased risk of cardiovascular events, surely their clean CETP Inhibitor would reveal the benefits of HDL elevation?

The Phase 3 trials are running now to answer the question.  DrugBaron, though, predicts more disappointment when the final results are published.

The opposing conclusions reached by DrugBaron and Merck hinge on the interpretation of the epidemiology data.  And in particular on the link between an observed association and an implied causation.  In effect, Merck are betting the entire cost of the CETP Inhibitor programme on Sir Austin Bradford Hill.

Merck shareholders without a background in epidemiology may be relieved to discover that Sir Austin Bradford Hill isn’t the name of a racehorse but a famous British epidemiologist who formulated an eponymous set of rules, the Bradford Hill Criteria, for interpreting epidemiology data.

“The odds on a decent racehorse may be better than the chances for CETP inhibitors once the full clinical dataset has been assembled”

Back in 1965, Bradford Hill published nine criteria for judging whether an association observed in epidemiology studies is likely to be causal.  The distinction between cause and association matters a great deal to the pharmaceutical industry because of the ever-increasing reliance on biomarkers in early stage drug development.  And nowhere is the dependency on biomarkers greater than in the cardiovascular field.

The reason is obvious enough: demonstrating an effect on the clinically relevant (‘hard’) end-points, such as heart attack and stroke, requires thousands of subjects treated for years.  No surprise then that the failed development torcetrapib cost Pfizer $800M.   And the only way to de-risk those huge, expensive hard end-point trials is to use biomarkers that, hopefully, predict efficacy,

In that respect, the cardiovascular field is well blessed – a wide range of lipid measurements have been described that are associated with increased risk of heart attack and stroke.  From the landmark studies in the Framingham cohort that first demonstrated a link between blood cholesterol concentration and future risk of heart disease, a stream of high quality studies have looked in ever increasing detail at the role of different cholesterol-containing particles (so-called ‘lipoprotein particles’) in the pathogenesis of heart disease.  The biggest advance came with the realization that not all cholesterol is bad for you – low density lipoprotein particles (LDL) with lots of triglyceride in them were bad, but high density lipoprotein (HDL), rich in cholesterol, were actually associated with protection.

Individuals with low HDL or high LDL (or worse still both) were at higher risk of heart attack.

From here, the story splits in two: the familiar tale describes the discovery of the statin class of drugs that lower LDL cholesterol, deliver real reductions in heart attack and stroke, and yielded Lipitor™ atorvastatin, the biggest selling drug in history.

The other half of the story leads to the CETP Inhibitors.  It took much longer to find a suitable target that modulated HDL – until the chance discovery of a Japanese family with very high HDL levels who, it turned out, had an inactivating mutation in the gene encoding the CETP enzyme.  It took almost another decade to find a viable CETP inhibitor, but when torcetrapib was taken into the clinic it showed dramatic elevation in HDL.

Lets put it in big letters: when using spot measures, even if elevated triglyceride is the true causative factor behind heart disease, then lowered HDL would still be the strongest epidemiological association.

Such was the strength of the association of lowered HDL with heart disease risk, many expected a procession through clinical development to a market that matched or even exceeded the statins.  After all, Bradford Hill’s criteria told us that strength of the association was one of the key determinants of whether the observed association was causal.   Indeed, HDL ticked all of Bradford Hill’s criteria.  So what could possibly go wrong?

Something did.  In 2006, the Phase III trial comparing a combination of torcetrapib and atorvastatin with atorvastatin alone was stopped early, because of an excess of heart attacks in the patients receiving the torcetrapib combination.  How could that be?

The generally accepted explanation hinges on the relatively mild hypertensive side-effect that was already known to accompany torcetrapib treatment.  After all, like elevated LDL, elevated blood pressure is known to be associated with an increased risk of heart attack and stroke.  By any rationale assessment, the massive increase in HDL should have outweighed the small rise in blood pressure.  Certainly that had been Pfizer’s judgment as they invested heavily in the Phase III programme knowing well the hypertensive side-effect.  But when the data came in, what other conclusion could they draw?

It was like a whodunit with only one suspect.  Somehow the hypertensive effect must have sunk the drug.

That at least was Merck’s take on the data.  When they discovered their CETP Inhibitor, anacetrapib, lacked the same hypertensive side-effects they moved – albeit cautiously – into Phase III.  Roche, with dalcetrapib and Lilly with evacetrapib, two more CETP inhibitors queing up behind Merck’s trailblazer, presumably feel the same way.  After all, what else could explain torcetrapib’s failure, if not the increase in blood pressure?

But there is another candidate – Sir Austin Bradford Hill.

His nine criteria have survived the test of time, and although Total Scientific recently suggested some updates, the core principles set out by Sir Austin are as valid today as they were in 1965.  Except he missed one.  And the missing criterion is particularly important in the context of lipid biomarkers.

The problem lies in the inter-correlation between all the markers of lipid metabolism in use today (LDL cholesterol, HDL cholesterol, total cholesterol, triglycerides and even more complex sub-fractions defined by their apolipoprotein content).  A change in one parameter is, across the population, associated with changes in the other parameters.  This inter-correlation makes it difficult to tell which changes are really causative and which are just bystanders.  HDL only wins the argument as the “best” candidate as a causative factor because of the strength of its association with heart disease.

However, this misses a key “in-built” advantage that HDL has over other lipid markers: its temporal stability.   If you measure HDL today, tomorrow and next week in the same individual the answer will be virtually the same.  But that cannot be said for triglycerides, for example.  Triglycerides go up and down after every meal, hour by hour, day by day.  As a result, a single measure of your blood triglyceride levels is a poor proxy for your average level of weeks or years.  By contrast, a single measure of HDL cholesterol is an excellent estimate of your lifetime value.

This affects how good each of these markers can possibly be as a biomarker.  And it depends on the temporal stability of the phenotype you are trying to predict as to which will be better.  Try and predict something like coronary heart disease, which develops over decades, and a spot measure of HDL has a much better chance of being associated than a spot measure of triglyceride (which may well have been elevated in the hours after your last meal when the blood sample was taken, or depressed because you missed breakfast on the way to the doctor’s surgery).  On the other hand, if you try and predict a phenotype that varies hour by hour, such as how hungry you are, then HDL has no hope of being associated – it never varies, unlike your appetite.

The “in-built” advantage of HDL as a biomarker for heart disease is its temporal stability.

So when it comes to predicting a slowly changing phenotype (like heart disease) from spot measures of lipid metabolism, its no surprise that all the top associations are with the stable parameters (HDL and Lp(a), which hardly vary at all, are strongest; LDL and total cholesterol are in the middle, while the temporally variable triglycerides, VLDL and chylomicrons languish at the bottom of the table).   But does that pole position really reflect the importance of HDL, or just its temporal stability characteristic?  After all, low HDL is correlated with a higher triglyceride level (on average).  So even if it was triglyceride that was the major causative factor in heart disease, HDL would still be the strongest correlate.

That point surely bears repeating: its absolutely true to say that, when using spot measures, even if elevated triglyceride is the true causative factor behind heart disease, then lowered HDL would be the strongest epidemiological association.

So you can tick all the Bradford Hill criteria, and still not be causative.  Where does that leave the Merck programme, which was progressed on the continuing belief that low HDL was a causative factor behind cardiovascular disease?

Well, if a proper assessment of the epidemiological data should really conclude that we don’t know if HDL is causatively associated with heart disease, and on top of that the Phase III torcetrapib data shows there was no dramatic benefit associated with HDL elevation (allowing the smaller detrimental effect due to raised blood pressure to exhibit itself), the foundations for an expensive Phase III hard end-point trial looks considerably shakier.

In November 2010, early results from the anacetrapib trial (called DEFINE) were announced at the American Heart Association (AHA) meeting in Chicago, IL: the effect on the lipid profile was as magnificent as expected.  HDL was doubled, LDL was lowered as much as with a moderate statin.  There was no increase in blood pressure.  Nor was there any meaningful reduction in events.

Expert cardiologist Steve Nissen at the Cleveland Clinic reasonably concluded that anacetrapib lacked the liabilities of torcetrapib, but that DEFINE had not included enough events to determine if there was a benefit.   Results from the two-year follow-up in DEFINE are set to be announced in 2012.  A much larger hard end-point study with dalcetrapib (the dal-OUTCOMES trial with 15,400  subjects) is also underway, together with another larger trial with anacetrapib (called REVEAL).

What will happen?  Even DrugBaron does not have a crystal ball.  But it is far from a done deal that these vast studies will show any significant benefit on the hard end-points.  It will also be difficult to interpret the result if the benefit is small: any benefit there is could be due to the additional LDL lowering rather than the HDL raising.

The core principles set out by Bradford Hill are as valid today as they were in 1965.  Except he missed one.

But the fact that the data already released from DEFINE did not show a significant benefit, like the torcetrapib data before it, even now strongly suggests that HDL is not the single major causative lipid factor behind heart disease, as has long been thought.

Yes, low HDL is a good marker for who is at risk from heart disease.  Nothing will change that.  But as a therapeutic target, its days may be numbered.

Merck (together with Roche and Lilly) are backing Sir Austin Bradford Hill, and his criteria, to have made the right call.  We don’t yet know the cost of the anacetrapib clinical development programme – Pfizer’s torcetrapib programme was quoted at $800M – so we don’t really know the size of the bet.  But it is large.  And if DrugBaron is correct, and Bradford Hill’s “missing criterion” turns out to be the villain of the piece, Merck’s shareholders may yet be wishing that Sir Austin Bradford Hill was the name of a runner in the Cheltenham Gold Cup.  The odds on a decent racehorse may be better than the chances for CETP inhibitors once the full clinical dataset has been assembled.