Neuren - where to from here ?, page-54

Hashan has done an excellent job of discussing Neuren’s potential. However, an otherwise excellent report is marred by several significant errors and unjustified assumptions that render many of his numbers invalid.

A strong statement perhaps, but I’ll back up this up with evidence and some simple maths anyone can double-check.

These issues make a huge difference to the numbers. Some cause large overestimates, some large underestimates. I’ve done my best to correct for these errors and also explore a fuller range of valuation options that Hashan has not covered. Read on to see where this takes us.

Hashan starts his analysis by estimating a value for Daybue (Retts only). Daybue for Retts is pretty far advanced now so there's plenty of numbers to plug into his calculations. So this part of the valuation is arguably solid. He comes up with AU$2.7b-AU$3.6b, or $21-$29 a share. Of this valuation AU$1.2b is allocated to the ROW market (meaning he’s estimated ROW is worth 50%-80% of the US market, a point I’ll return to later).

This leads to the remarkable conclusion (not news for many of us) that Daybue Retts accounts for the entirety of Neuren’s current market valuation. As Jon has emphasised multiple times, NNZ-2591 is the key value driver of Neuren looking forward. Yet it appears to account for zero value currently. This clearly makes no sense from any perspective. On this alone we can conclude Neuren represents an investment with almost zero downside (can’t discount NNZ-2591 any lower than zero)

Appropriately, most of Hashan’s analysis then focuses on NNZ-2591. Using estimates of patient numbers, drug pricing, market penetration he’s estimated peak sales of AU$22b for the current four Phase II indications.

He then applies a Probability of Success (POS) (i.e. probability drug gets through to approval) for each disease:

Phelan-McDermid: Phase III 30%

Angelman: Phase II 15%

Pitt-Hopkins: Phase II 15%

Prader-Willi: Phase II 10%

This gives an average sales weighted POS of 19% (81% discount) across all diseases.

Then using a “typical price-to-sales ratio for pharma of approximate ~3x” he says “this results in a risked valuation for NNZ-2591 of ~US$12.3b”. He then discusses updating the numbers using Skylarys 4.8x price-to-sales ratio.

Hashan comes up with some big numbers. However, there’s some serious issues with pretty much all of them.

Firstly, comparing the patient numbers in his Table 4 vs those Neuren has provided for the US market Hashan has almost double the total patients - 128k vs Neuren’s 66k.

Hashan Neuren

Phelan-McDermid 40k 24k

Angelman 27k 19k

Pitt-Hopkins 29k 6k

Prader-Willi 32k 17k

Where did he get these wildly different numbers from?

If you calculate them out it appears Hashan has taken the estimated prevalence rates which he listed in Table 1 (and which also don't match Neuren's) and applied this to the total US population. And he's taken the upper bound prevalence estimates for all indications. For example, he appears to show Pitt-Hopkins as 1/11,000-1/41,000 (Neuren shows 1/34,000-1/41,000). Divide the US population 330 million by 11,000 and you get ~30k (very close to Hashan's ~29k).

In doing a conservative analysis why would you take the upper bound prevalence rate? And where did he get these rates from? Neuren at least is using the mid-point. I love this numbers, but I think the credibility of the analysis is improved significantly if we stick with Neuren's own numbers. Doing this halves the peak sales number.

Next we have an obvious error. The numbers in Table 4 are in AUD (multiply them out using USD/AUD of 0.65 to confirm this). However, when Hashan multiplies his risk adjusted peak revenue by 3 he suddenly swaps out AUD for USD. $4.1b x 3 = $12.3b. But its AU$12.3b, not US$12.3b as he's written. A typo perhaps? Maybe, but it's not a small error - it gives a valuation 50% higher than it should be.

After coming up with $12.3b using a 3x price/sales Hashan goes on to suggest a more appropriate ratio might be the 4.8x implied by the recent Skyclarys sale. Hashan then adjusts his numbers to a higher acquisition value based on this

Actually no, he then skips on finishing this train of logic and instead suddenly drops in a mysterious 90-95% discount (“to account for clinical and regulatory risks”), to finish with a LOWER acquisition value of “US$5.3b – US$10.6b”.

Hashan's original model in Table 4 showed an average discount of 81% (avg POS of 19%). But now he’s changed this (without any reasoning) to a 90-95% discount. This is a POS of 5%-10%, just 25%-50% of his first model's values. Perhaps this is to help compensate for the overvaluation introduced by his earlier error? Either way, it makes no sense as I'll cover in detail below.

So what about the US$5.3b-US$10.6b? Firstly, as already indicated, these should be AUD. Secondly, when he refers to this as the “acquisition value” he must be talking about NNZ-2591 only because (1–0.95)*22*4.8 = 5.3b and (1–0.9)*22*4.8 = 10.6b. So he's left out his just completed Daybue valuation. However, the numbers are so thoroughly wrong by this point it probably doesn't matter.

Putting aside why Hashan has made so many errors and unqualified assumptions the more interesting question is what happens if we correct for these and plug in credible (as in evidence based) values for the POS?

A first step is a detailed look at the POS. Choosing this correctly has an enormous effect on the valuation. Poorly justified POS/discount factors are one of the biggest problems I have with all analysts reports I’ve read to date on Neuren. Analysts consistently use drug phase success discount factors that seem to be pulled from nowhere.

How do we correctly estimate the POS? I'm not an industry expert nor a qualified analyst, so I'll just try some common sense. And common sense suggests to me a sensible unbiased approach would be to use average success rates based on historical trial success rates. And the good thing her is there's enormous publicly available datasets covering hundreds of thousands of drug trials. With such huge datasets calculating statistically valid historical drug success rates is possible. And in fact there are plenty of high quality research papers that have done this.

For example, a 2016 paper looking at data from 2000 to 2015 covering 9,985 transitions for 7,455 clinical drug development programs across 1,103 companies https://www.bio.org/sites/default/files/legacy/bioorg/docs/Clinical%20Development%20Success%20Rates%202006-2015%20-%20BIO,%20Biomedtracker,%20Amplion%202016.pdf calculated:

POS All indications - Phase II to Approval: 15%

POS All indications - Phase III to Approval: 49.6%

Because these datasets are so large its possible to filter down to specific disease areas and still have a large enough N (number of studies) to maintain statistically valid numbers. Looking at Neurology drugs only this study found:

POS Neurology - Phase II to Approval: 14%

POS Neurology - Phase III to Approval: 48%

Not much different from the average. More relevant perhaps was a drill-down into just Orphan diseases (N=~1300), which found a doubling in the Phase II to Approval POS.

POS Orphan - Phase II to Approval: 33%

POS Orphan - Phase III to Approval: 66%

Possibly this makes sense as the Orphan drug framework is designed to tilt the playing field and make it easier to bring Orphan drugs to market, with lower thresholds for study size, effect sizes, etc

There are further studies published by Hay et al (2014) and Thomas et al (2016) who variously give Phase II to Approval of 15% - 16% and Phase III to Approval of 50%.

One of the most recent and highest quality paper appears to be this https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6409418/. Published in 2019 by several MIT researchers it has a very high 1,395 citations (number of citations is often used as a measure of research quality/significance). It analysed data from a truly gigantic 185,994 trials.

They found much higher success rates than earlier studies in Phase II.

POS All indications - Phase II to Approval: 35%

POS All indications - Phase III to Approval: 59%

Reasons they provide for the difference is that some of the most promising drugs skip Phase II altogether, something not captured by other research, and there are also combined Phase 1/2 and Phase 2/3 trials that were not classified by other studies.

Drilling down into CNS only (i.e. excluding oncology, infections diseases, etc) the results were noticeably lower. Pretty similar to the earlier studies above.

POS CNS - Phase II to Approval: 19.5%

POS CNS - Phase III to Approval: 51%

This is one of the few studies that drills down into Orphan drugs studies in significant detail. Unlike the earlier study finding of an improved POS for Orphan drugs they found the opposite. They don't discuss why this might be that I could find.

POS Orphan - Phase II to Approval: 12.7%

POS Orphan - Phase III to Approval: 47%

Drilling down even further into just CNS Orphan diseases (what we're most interested in) the results were even worse.

POS CNS - Orphan Phase II to Approval: 8.3%

POS CNS - Orphan Phase III to Approval: 32%

This implies a Ph II success rate of just 26% (8.3%/32%), a pretty low number which I'll discuss again later.

It should be noted that neurodevelopment disorders (what we're interested in) is just one segment of CNS disorders (which also include neurodegenerative disorders, demyelinating diseases, cerebrovascular disease, etc).

Perhaps its not unexpected that CNS Orphan diseases show such a low success rate. We already know that CNS orphan drug development has been particularly unsuccessful area historically. I think I remember a prior hotcopper post about a string of repeated CNS drug failures at one of the big Pharma (hottod re Pfizer?).

We also know that Daybue is the only approved neurodevelopment orphan drug to date. Which implies all the others failed…

That was alot of detail. But I wanted to show that, unlike Hashan and other analysts appear to be doing, I'm not pulling numbers out of thin air.

So if we just base our analysis on the historical success rate of CNS Orphan drugs as calculated by our the highly cited research source, then 8.3% and 32% are the numbers to put into the model (being completely objective about whether we think NNZ-2591 is special or not).

Contrasting this with the numbers Hashan used in his model

Phase Table 4 90-95% disc Wong et al (2019) MIT

Phelan-McDermid Phase III 30% 5%-10% 32%

Angelman Phase II 15% 5%-10% 8.3%

Pitt-Hopkins Phase II 15% 5%-10% 8.3%

Prader-Willi Phase II 10% 5%-10% 8.3%

Putting all this into a corrected model (with Neuren's patient numbers, and sticking with AUD) and then multiplying by the price/peak-sales of 3x for the conservative end, and 4.8x for the less conservative we get a range of between AU$5.8b to AU$9.2b. Adding back in the lower bound Daybue valuation Hashan provided of $2.7b to the 3x value gives AU$8.5b for a lower end valuation. Adding the upper bound Daybue value of AU$3.6b to our 4.8x value gives AU$12.8b as the upper end valuation. This gives a range of $67 - $102 per share.

This is then the estimated CURRENT acquisition value of Neuren. i.e. the value as of today.

Not as good as US$10b perhaps, but don't worry, because this valuation still ignores ROW.

ROW was important enough to include in the Daybue valuation. In fact it Hashan valued it at 50% - 80% of the US market valuation, even though Daybue is yet to be approved in any ROW market.

So is 50% to 80% of the US market value a reasonable valuation for ROW NNZ-2591?

Doing a common sense check using available information on the US drug market vs the ROW market I’ve found the US represents a little below 50% of the world drug market by revenue (staggering when you think about that). I’ve read estimates ranging from 42% to 45%. For Orphan drugs I could only find one source that suggested 53%.

I could dig further, however, I don't think its controversial to suggest that to a first approximation ROW has the same revenue potential as the US.

That said, it will definitely be less profitable, with lower average drug prices and regulatory approval to be obtained (and customers to be supported) in many countries, not just one like the US. But still has significant value (one reason why Acadia paid US$100m up front for it for Daybue).

I’d therefore suggest it seems reasonable to apply a ROW value for NNZ-2591 of at least 50% of the US market value. This discounts its size 50% due to lower profitability.

Adding in ROW for NNZ-2591 we get an updated ‘acquisition value’ of AU$11b – AU$17b or $90 to $138 per share. Once again I’d emphasise this is a theoretical current value (not the future value).

Now lets return to the POS again. I've have used the historical averages for CNS Orphan drugs, which is what you'd choose if you knew nothing a priori about about NNZ-2591. Based upon a historically abysmal 25% Ph II success rate and just 32% Ph III you'd need, on average, 12 drug candidates to get one to Approval.

However, lets add a bit of Bayesian reasoning. How many drug candidates has Neuren put through Ph II/III trials to date? Two I believe - Retts and Phelan-McDermid. What's Neuren's Ph II success rate so far? 100%. Its Ph III success rate? 100%. Are these independent variables? No. Both Daybue and NNZ-2591 work through the same mechanism of action - normalising IGF-1 in the brain. This means these successful results reinforce each other.

The lower the initial probability of success the greater the relative increase in confidence for the later trials after already getting a successful result. If the average success was 75% and you got two successes in a row that would tell you little about future success likelihood. However, if the average success is 25% then the likelihood of 2 consecutive Ph II successes is only 6.25%. Our a priori successes makes a huge difference.

I haven't tried putting these numbers into a Bayesian model (will look at that later), but I suspect we'd end up with the 25% Ph II increasing to >50% (2 priors) and the Ph III increasing to about 50% (1 prior).

This also passes the common sense test. Do most of us feel like the chances of the next 3 Ph II trials being positive remain just 25% each? I don't think so.

Updating our model with the higher numbers NNZ-2591 clearly warrants (the exact numbers could be argued, but they are definitely alot higher) I get an updated current day valuation of AU$25b to AU$39B, or $138 to $215 per share. Those priors make a BIG difference.

However, we still haven't finished exploring the full value of Neuren.

As we all know, and Hashan also emphasised to good effect, if NNZ-2591 is successful then these 4 indications won’t be the end of the story. There is still Fragile-X (with 3x potential patients of Retts) of course. But there's also dozens more neurodevelopment disorders that could be targeted, not to mention the still open question of Autism, and neurodegenerative disorders (open because there are some good arguments for only sticking to neurodevelopment orphan diseases).

Its not to imagine this could produce additional revenue at least equal to the current four NNZ-2591 indications being targeted. In Hashan’s words, “NNZ-2591 is a pipeline in a drug”. An acquirer will not be blind to this potential. So why should we. And I’m sure Neuren management aren’t. In fact, there has been suggestion that the current four disorders were carefully chosen to the best evidence for wider drug application. These other indications could be many worth multiples the current ones (although with patents due to expire eventually the window for exploring them will will start to close).

It seems reasonable to me this additional potential could add another 50% to our acquisition value. There is no question in my mind large acquirer would be vigorously pursuing many more diseases. Adding an uplift for this potential raises our valuation still further to AU$25b-AU$39b, or $196 - $308 a share.

So now what happens if the next 3 Ph II trials are positive? This means our model would now shift to 50% for Ph II to Approval. (As an aside, 47% is the NORMAL historical Ph II to Approval POS reported in the research for Orphan drugs once you shrug off the CNS specific Orphan failure curse, which we assuredly have).

Now we're up to $326-$515 per share. This is were we may be (with respect to calculating a fair acquisition value for Neuren) by end of this year all going to plan.

What if Neuren takes all four through to Approval? This might take 3-5 years. But if Neuren goes it alone, is successful, and then sells, we're now looking at $630-$1002 per share.

CSL here we come .

Believe it or not there are still more areas left we could argue for yet further upside valuation in the above analysis. A few examples:

Hashan states “if NNZ-2591 delivers comparatively superior efficacy with a similar or better safety profile, it can be priced at a premium to Daybue.” As he mentions, the initial results for NNZ-2591 suggests strongly that it may have better efficacy and almost certainly a better side-effect profile, thus potentially justifying a higher price.

Yet in his model (and in the analysis above) the pricing has been kept at $375k. Based upon his reasoning shouldn’t it be higher? Maybe $475k? or $575k after discounts? The pricing seems in the model seems to be, based upon the evidence, on the conservative side.

Hashan also talked alot about long term persistence rates, converting the 37.7% of patients reporting improvements on Daybue into a 37.7% long term market penetration for Retts. Putting aside whether this logic stacks up (it doesn't...) there is the question of long term market penetration. Hashan has modeled 30% for NNZ-2591.

I can't find good data on the historical penetration rates for Orphan drugs. I found just 1 paper, which showed penetration rates from 1% to 77%, an enormous range.

I would expect it must depend upon a huge number of variables - disease severity, ease of diagnosis, drug effectiveness, side effect profile, etc. NNZ-2591 looks quite favourable in these respects, and also appears likely to be a better performing drug than Daybue based upon initial results. For example, although 37.7% of Retts patients have apparently recorded symptom improvement, the Phelan McDermid Ph II trial recorded improvement in 16 of the 18 subjects which is 89%. This all argues for a higher penetration rate than Daybue. Of course, we don't know what the final penetration rate of Daybue is yet either.

In conclusion I can't find any data on which to base a possible range of penetration rates. However, I can look at what would happpen if the penetration rate was 50% instead of 30%, which seems a potentially plausible upside scenario. This would increase the peak sales by 40%. This increases the risk adjusted (for today's situation) valuation from $99 to $153 per share to $151 - $236 per share. Add in ROW & an uplift for additional drug indications and we increase from the previous $196-$308 a share to $313-$495 a share. If all Ph II positive the previous $325-$515 per share increases to $529-$840 per share. If Neuren brings all to approval, $630-$1002 now becomes $1040-$1650 a share.

So increasing the penetration rate makes a BIG difference. The key point is there is still a great deal of upside in the model.

But wait there's more!

With the recent revolution in Whole Genome Sequencing (WGS) Hashan's 2x overestimate of market size could end up having some validity. I'll put up another post on this shortly, but this has the potential to be revolutionary for genetically based rare diseases, and the timing of this revolution is perfect for Neuren.

A few final comments.

Firstly, whether price-to-peak-sales that Hashan has used is the best valuation methodology is an open question. I think DCF is still the best method. I haven't had time to update my Neuren DCF model with all the latest data above, but will do so sometime soon and post what I get out the other end.

Secondly a comment on analysts. It seems like I've been doing a beat-up on Hashan. However Hashan is still my favourite analyst by far. He's one of the few willing to properly explore the value of Neuren AND explain his modelling. Most analysts make vague comments about using DCF models, but seldom present any details of the analysis for others to critique. The more time I've spent researching pharmaceutical valuations the less confidence I have in these analysts.

For example, how on earth can you put real numbers into any model and magically spit out the $29.01 target share price Petra Capital has published. I'd say this is not possible. The chances your model outputs something just a few $ shy of the current share price is zero. Furthermore, any modelling has enormous uncertainties and if you're doing it properly you'll get a range, not an exact share price target accurate to 1c.

Perhaps this explains why most analysts don't reveal details of their modelling. And it leads to my feeling that analyst 'target share prices' are actually based upon a formula more along the lines of [Current Share Price] x (1 + 0.25 x F) where F is a number between 0 and 1 selected based upon how I'm feeling today combined with yesterday's weather forecast.

Finally, I've criticised Hashan for his errors, but I probably have errors in my analysis too!
Feel free to critique me also