A Cancer ‘Moonshot’ Needs Big Data | Hacker News Reader

A Cancer ‘Moonshot’ Needs Big Data

wsj.com

49 pointsaficionado54 comments

More importantly, we need FDA regulatory reform. There is absolutely no reason that approved anti-cancer drugs in one cancer type and stage should be limited to just those cases. You should be able to spin up a direct Phase II study without going through Phase I. And more importantly you should be able to get these drugs when all other options are exhausted.

People need to read the book "The Death of Cancer" by the former NCI chief Vincent DeVita. I agree that we need a cancer czar and that the NCI needs to run the show including drug approvals. In addition, academia needs to get with the program.

> There is absolutely no reason that approved anti-cancer drugs in one cancer type and stage should be limited to just those cases.

Can you expand on that?

Mutations in the BRAF gene, to name one druggable example, are a cause of many different types of cancer - melanoma, some breast, small-cell lung cancer, etc. Under the current paradigm, each type of cancer is treated as a different disease and requires a different clinical trial. This is bad.

What should happen (and is beginning to happen, with "basket trials") is that cancers are grouped by their genetic profile, instead of their tissue of origin. The BRAF V600E mutation is targetable with Vemurafenib, and so that drug should be used regardless of whether you have breast cancer or lung cancer.

Perhaps, but it is also true that these targeted therapies have different success rates among different cancers. Vemurafenib, famously approved for melanoma, hasn't been as successful in papillary thyroid.

Certainly it makes sense to consider genetic profile in treatment, but doesn't the statement "that drug should be used regardless of whether you have breast cancer or lung cancer" ignore the local tumor microenvironment and possible differences in drug resistance expression between cell types, i.e., isn't reducing cancer down to genetic profile alone potentially as problematic as reducing cancer down to its tissue of origin?

Your points are right on - I oversimplified while trying to explain the concept.

This isn't really that much of an issue because oncologists are not always idiots and can prescribe drugs off-label, with cause. Identifying a braf mutant in a disparate cancer type would probably qualify.

It is a huge issue. They risk FDA censure and it usually requires IRB approval as well. It will change with precision medicine but it won't change fast enough. People really need to read DeVita's book. All of my comments are taken from his arguments. DeVita was the chief of the NCI for 30 years.

I'm sorry that you think academia isn't "with the program". There's lots of great basic and translational research going on in the academic sphere. Can you at least elaborate, if you're going to paint us all with a broad brush?

Academic cancer centers need to focus on improving clinical outcomes, not just basic research. That cancer centers are associated with primary research institutions universities is largely a compromise between with the NIH and the NCI when the national cancer act was implemented. Tenure is based on basic research publications and on not clinical trials. So therapies are not optimized.

How are we supposed to cure diseases we don't fully understand? How can radical new solutions be tested if we don't even know if they're feasible? Basic research answers these questions.

If you would have waited until you fully understood a disease in order to treat it then kids would still be dying of leukemia today instead of having a 90% remission rate today. People should read about Emil Freireich and the development of chemotherapy.

>"kids would still be dying of leukemia today instead of having a 90% remission rate today"

Do you know a good source for the historical data? I am wondering how they rule out that in the 1950s/60s they saw only much sicker patients, while today a bigger population gets diagnosed and treated. Related, there may be lead time bias: https://en.wikipedia.org/wiki/Lead_time_bias

Any source with discussion of these methodological issues would be helpful.

Where is the money supposed to come from for this? Keeping basic research labs afloat in the current funding environment is hard enough -- clinical trials etc are orders of magnitude more expensive, and not practical for most academics, even those interested in translational research.

For better or worse, our current model is for academics to do basic and early-stage translational research (mostly animal models), then sell/spin off to companies to do the actual trials and reap the profits. I don't see what academics are supposed to do about this.

It would come from contracts from the NCI.

What we need to do is get the price down of cancer drug development. The fundamental problem is cancer is really many diseases so that any targeted drug will only be specific for a very small subset of cancers. This means the market for any new targeted drug is very small and so the ludicrous cost of bring as new drug to market is too great to be supported by the small number of people that will use it.

The cost of these new targeted drugs is basically too high for any healthcare system to support, but the we need to make the drugs even more specific if we are going to get all the cancer types.

Smaller and smaller market with every rising development costs = failure.

Simply eliminating Phase I is the wrong solution. Entering a Phase II study before Phase I is a great way to maim or kill people. Anti-cancer drugs are often severely toxic; even the new spate of immunocentric therapies proved highly dangerous when first administered. Several early recipients died due to clueless drug administration protocols. THAT's the reason we have Phase I. Humans usually respond differently to a drug than the animals who have gone before.

That said, I think the FDA should approach cancer differently than other diseases, probably by establishing a new experimental model entirely (eliminating the Phase I/II/III system). Given that canger drugs and patients and protocols must ALWAYS be administered sensitively, circumstantially, and adaptively, the current 3 phase system seems to be serving cancer patients poorly, locking them into ineffective treatments for too long, based on too much emphasis on inflexible trial policy. Experimental protocols that are more adaptive and personalized are needed. Of course, this will be very hard technically and insanely expensive clinically.

How to begin this? Like Obama, I'd love to see a 'moon shot' reinvention of the cancer business. Perhaps a project-based redesign of the system is possible, by engaging the most innovative cancer research clinicians and data analysts to explore alternative models for cancer research and drug trials. If that's what Obama has in mind, I support his wish. Now how do we make it work?

I meant in cases where the drug is already FDA approved e.g. Phase I trial has already been completed in other cancer types. There is no need to do another Phase I trial just because it is another cancer type.

The restrictions on using an approved drug for a new cancer type or stage is silly.

The original NCI program for treating Leukemias worked pretty well. The FDA is extremely risk adverse which is not what you want for terminal cancer treatment.

"Cancer" isn't a single disease or problem domain, but what could be categorized as a syndrome that shares some basic commonalities.

The governmental "War on Cancer" began under Nixon and was pitched as a "moonshot" type of program, which promptly ran aground on the realities of cancers.

If only there were some sort of documentation for this kind of effort... perhaps they could call it "The Emperor of All Maladies" or something.

Excellent PBS documentary you ironically refer to...

http://www.pbs.org/show/story-cancer-emperor-all-maladies/

Quite a beast to conquer, for the reason you stated...errant DNA replication, probably the single most powerful software on Earth, gone beserk...

Obama's "moonshot" reference likely refers to a proposed attempt at re-centralizing the efforts of earlier research projects, to consolidate and benefit from known gains...

"Moonshot" is probably simply the hyperbole choice of a speechwriter, designed to inflate the magnitude and importance of the endeavor...

The book is excellent as well

http://www.amazon.com/Emperor-All-Maladies-Biography-Cancer/...

For those interested in reading this book: Your mileage may vary.

I wanted to like it, but I couldn't get over the writing. It had so many flourishes and unnecessary descriptions that I stopped reading halfway through.

>'"Cancer" isn't a single disease or problem domain, but what could be categorized as a syndrome that shares some basic commonalities.'

Cancer is characterized by having cells with an abnormal number of chromosomes and genetic instability. It is one disease. I don't know the origins of it, but this "many disease" claim sounds like an excuse for lack of progress in targeting aneuploid cells.

"In contrast to normal cells, aneuploidy--alterations in the number of chromosomes--is consistently observed in virtually all cancers." https://www.ncbi.nlm.nih.gov/pubmed/15549096

Of course cancer is a result of genetics gone amok, but that's about as useful as saying that death is caused by hypoxia... of course it is, all death is the result of lack of oxygen to the brain.

Cancer as a disease, in its expression, effects, and treatment, is not a monolith and has been utterly resistant to attempts to treat it as such.

If you could count the chromosomes in every cell of a patient's body and specifically kill cells with/without certain numbers, you could cure the cancer. For practical purposes it is one disease with a single identifying characteristic and molecular target. Only technical limitations are preventing a cure.

No disease manifests exactly the same way in every person, so I don't get the point of calling cancer multiple diseases.

> you could cure the cancer

Or you could vaccinate against a virus and prevent the cancer. Not all cancers need to be attacked the same way.

>"Not all cancers need to be attacked the same way."

Ok, but what does that have to do with the (apparently new) idea of calling cancer "multiple diseases"?

Not all cancers have obvious chromosomal aberrations

I was very surprised to learn a few years ago that this is <10%, possibly much less depending on measurement error (see my ref above). So then there are two diseases: Aneuploid cancer (>90%) and non-aneuploid cancer (<10%).

I'm not a huge fan of "war on x" mentality, and even less of government funding these scientific endeavors in general, but it's hard to argue that we haven't made progress fighting cancer. Much of the improvement in treatments have hit the common clinic in the last ten years. A reasonable goal, say, being able to put ~80% of patients who present with cancer into remission, seems possible in the next 20 years, loftier goals, say 95% or "80% with nearly no side effects", I think will be hit by 50 years.

That's a very enlightening and depressing book.

I don't agree with much that Tom Coburn says, but he's right that 'data silos' are a big problem in cancer research. To elaborate on a few points:

- While enrolling 500 people with breast cancer in a study might be easy enough at one academic hospital, there are lots of rare cancer types. It's fiendishly hard to gather enough cases to do real research on these, and samples are like gold, so people are resistant to sharing.

- There's lots of duplication of effort. While studies funded by the NIH are required to deposit their data, what's being deposited is raw sequence data, which requires lots of computation, disk, and skill to turn back into useful information about the mutations in cancer.

- The clinical data (phenotype) that is released with studies is often incomplete or vague enough to be useless

- Getting data into and out of these access-controlled repositories is, to put it mildly, a pain in the ass.

We do need to protect people's privacy, but removing some of these hurdles would go a long way towards accelerating the field.

Well, better / more uniform access to patient records would be nice, but more money into basic research is probably the best bet (as it has always been). It makes me think of this Gregory Petsko commentary (open access):

http://www.genomebiology.com/2001/3/1/comment/1001

With the great quote:

"...If a hundred years of cancer research has taught us anything, it is that if you must get cancer, you want to be a mouse, because we can cure cancer in mice."

Maybe it is cheaper to keep trying and there is less oversight so more BS gets through...

>”...because we can cure cancer in mice.”

Actually as far as I know we can’t cure mice of cancer - by cancer I mean natural cancers that mice get as they age just like human cancers. I have looked through the literature in great detail and I can’t find a single recent paper where anyone has even tried curing mice of cancer. If anyone know of one I would love to read it.

The way we go about trying to cure cancer is insane. We basically take inbred mice, inject them with a cancer cell line to give them cancer, and then treat them with experimental drug X. This process is nothing like how real human cancers arise. On top of this we only test new drugs in people that have failed all other treatments and have very high tumour burdens. We then demand that the new drug have a significant effect in these patients despite us needing drugs that work in newly diagnosed patients. If we get a drug that works it is more by chance than design.

>"I have looked through the literature in great detail and I can’t find a single recent paper where anyone has even tried curing mice of cancer."

It's interesting how your perspective will change after reading the literature. It is the same thing with the various mutations claimed to cause cancer. For example, in this thread someone claims mutations to BRAF cause cancer. I am nearly certain that inspecting the literature will not result in any studies where that gene in a normal cell was mutated, and then it took on the properties of a cancer cell.

Maybe we will see some data on many cell generations later and in conjunction with various other treatments, but not the direct experiment of "non-cancer cell + BRAF mutation = cancer cell" while "non-cancer ell + control treatment = non-cancer cell".

Cancer in humans is never driven by one mutation - if it were there is no way we could reach adulthood. Cancer is the slow process of accumulating mutations in cell lineages - the time between the first mutation in say BRAF and a metastatic cancer is 20 to 40 years. Lots of other mutations have to happen before you get clinical cancer.

So which combinations of mutations cause cancer then? Replace BRAF in my pseudo-equations above with "combination of mutations".

There are hundreds of thousands of mutation combinations that lead to cancer. This is why it is such a difficult problem.

Has anyone done a proof of principle though. You only need to test one of the combinations. I don't think anyone has been able to do this.

This is a good question for which I don’t have the answer. We know from epidemiological research that you need at least 5 independent mutations to get to metastatic cancer. With the new CRISPR technology it should now be possible to make cells containing just 5 mutations which should then be able to be studied.

The bigger problem is these 5 mutations can be drawn from hundreds of possible genes so the combinations are almost endless.

>"We know from epidemiological research that you need at least 5 independent mutations to get to metastatic cancer."

This is all based on the Armitage and Doll 1954 theory, they write: "This result will be valid for large values of t (of the order of a human lifetime) provided that p_l x t, P_2 x t, ..., p_r x t are all sufficiently small (as could be assumed in an application of this theory to human cancer)" http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2007940/

Here, the "the probability of occurrence of the r-th change, in that line of descent, is p, per unit time [t]". So obviously if p or t is large enough, the probability of dying from cancer goes above 1, which is non-physical. There is a simplification being made here that was reasonable in the pre-computer age. What happens if you implement the same theory without that simplification?

>"the time between the first mutation in say BRAF and a metastatic cancer is 20 to 40 years"

Actually do you have a ref for this? I can't easily think of how they would date the mutation.

I haven’t got one off the top of my head, but it is pretty well known. An really good example was the atomic bombs in Japan. They tracked all the survivors and when they got cancer and the delay was in decades.

You're right, of course. BRAF mutations alone don't cause cancer. They're caused by the accumulation of key mutations coupled with clonal expansion. I oversimplified a bit while making a point about basket trials.

That's not what I meant. I said "and in conjunction with various other treatments", which was perhaps poor wording since by treatment I meant causing other mutations/etc. To my knowledge, no one has ever been able to take a normal cell, do X to it (where X is usually some combination of mutations), and see it becomes a cancer cell.

So we mostly use mice as soil to cure remission of already existing cancerous cells/tissue ? Not that similar to understanding cancer, at least cancer onset... Odd.

It is even worse than this as by controlling the number of cells injected you can make it very easy to cure the cancer.

The reason why this is done is it is much easier to do than the experiments that should be done. It is like the old joke about the drunk looking for his keys under the street light because the light is better despite have lost them in dark alley behind.

I hope organ generation/printing will help avoid these tricks. Isn't it the most promising way to make real progress ?

The most promising way would be to stop funding research if it is not done correctly and stop allowing people to publish papers with “play” research. No more use of inbred mice lines, more more injecting known cancers (only use natural cancers), no more testing of treatments that have no chance of working in humans.

The best suggestion I have seen is to make better use of all the millions of pets we have. Dogs and cats get cancer that is very similar to human cancer (much more so than mice) and these are all natural cancers. We need to encourage people to sign up their pets for cancer treatment and conduct many more trials in pets. The fact that we don’t do this at any scale is a sign that we are not too serious about curing cancer.

Involving people to treat their pet with current medicine looks like borderline impossible. If people think their dog will suffer they will skip that.

And research funding is probably conservative, since mice is the 'less bad' idea, money flow this way. Some competition may help changing habits.

There pet is already suffering - it has cancer after all. The alternative in most cases is to put the dog down.

More importantly this question has already been asked by researchers and most people are happy for their pets to participate in such trials especially when they know they are really helping cure cancer.

Aight then, my folk-understanding was off. Hoping it will help making steps.

I think we should be talking about 'uranushots' rather than moonshot because 1. the moon is close and now relatively ho-hum while uranus is in another category altogether; 2. For these things to happen a little good luck is required.

Reply on news.ycombinator.com