Why do so many new drugs fail in clinical testing?

ine of every ten new drugs fail in clinical testing. According to consultant Cheryl Barton, a drug in phase III testing has 32% chance of failure. Only 21% of drugs that enter phase I testing ever make it to market [1]. Even in Phase I, 37% fail. Most drugs fail in phase II. (Goodman and Gilman estimate the success rate in the three phases to be 50%, 30% and 25-50%, or an overall success rate of between one in 13 and one in 26). The percentage of drugs for neurological diseases that fail is even higher. For example, over 200 drug candidates for Alzheimer's disease have failed so far in clinical testing.

Why is this? People who develop drugs are not morons. It costs, by one estimate, nearly 900 million dollars to develop and test a new drug. The drugs have been tested in rats or mice, and they clearly work, or the company would not have moved into clinical testing. So why do so many of these drugs suddenly stop working when tested in people? Phases of Clinical Testing
Phase I = small study = healthy volunteers = test toxicity and dosing
Phase II = medium = patients = test of efficacy
Phase III = big = patients = search for side effects

It's probably fair to say that 99 of every 100 drugs that work in the test tube never even make it to large-scale animal testing. Most of the time, a drug is dropped because of suspected toxicity. It doesn't take much. Maybe a liver enzyme showed a slight elevation in one animal. One random fluctuation in one of a hundred tests can be enough to seriously dampen enthusiasm for a new drug. To chemists, it's frustrating because they know the test is probably wrong. Chemists do tens of thousands of tests during their careers, and develop an intuition about these things--but if you can't prove the test is wrong, it does no good to argue. Many useful drugs are lost simply because a company is overly cautious.

Once a drug makes it out of the lab, chances are it will still fail. There are a number of other reasons. Some might be obvious upon reflection, others are less so.

Laboratory rat and mice strains are too genetically homogeneous.

The way drug testing is done today is by administering it to a single strain of genetically uniform lab rats or mice, typically all the same sex. That's like testing a drug on a population consisting entirely of a thousand Olsen kilo-tuplets, all of whom have identical physical attributes and equal amounts of talent. If it works against the laboratory animals, then it's tested against a real population of humans, consisting of Frank, the fat cigar-smoking butcher, Delicia, the 90-pound crack addict skank, Clayton, the six-foot-two inch black basketball player, Lance, the guy from accounting who talks funny, and 996 other random people. A substantial number of them are already taking other prescription or recreational drugs, which could counteract the drug under test or make it toxic. Another substantial percentage have undiagnosed pathologies, which means they're more likely to get side effects. All of them have slightly different biochemistry. Is it any wonder the drug suddenly fails?

Studying the effects of a drug in a laboratory strain like Sprague-Dawley rats (top) makes no more sense, pharmacologically, than a clinical test using a hypothetical group of Olsen kilo-tuplets (bottom). [3]

I've done proteomic analyses on both rat and human samples, and the difference is astounding. The rat samples are all virtually indistinguishable. The human samples are so different from each other as to be almost un-analyzable. The brains of some patients are in a state of reactive gliosis, or runaway inflammation. Some patients, who appear normal, have huge amounts of certain proteins. Others appear to lack those proteins altogether. The biochemical differences between individuals in human populations are huge. No wonder the idea of five hundred pairs of Olsen twins is so appealing.

Researchers use laboratory strains of rats and mice for three reasons: because they're easier to handle, because the results are more easily compared to other studies, and because their uniformity makes it easier to see both beneficial and harmful effects. Ironically, these very qualities may also be making it too easy for us to see the desired effect, and also make it too easy to unfairly knock out potentially valuable drugs. Using a population of captured wild rats, or at least a mixed population of several different strains, would make animals a closer match to human populations. The result would be fewer false positives and less spurious toxicity. It would save millions in the long run.

Rats and humans have different metabolic pathways.

Rats can get drunk on methyl alcohol (wood alcohol), which is deadly poison to humans, because rats have much less of one particular enzyme that converts methanol to formaldehyde. It is the conversion to formaldehyde, and its rapid subsequent oxidation to formic acid, that makes methanol toxic. For the same reason, chocolate, which is harmless to most humans, is deadly poison to some types of dogs. There are many other "gotchas" that routinely trip up new (and often over-enthusiastic) investigators. For example, many chemical compounds protect rats and mice against ischemic stroke in the lab. But no neuroprotective drug has ever been effective in treating acute ischemic stroke in humans [2]. The reason is that the brain biochemistry of humans is different from that of rats.

This doesn't mean that animal testing is useless. The similarities between rats and humans are very strong. It just means that we need to know more about how they're different.

There are unforeseen side effects.

Many drug failures are caused by unforeseen side effects. There's not much that can be done about this, other than more basic research to uncover how the body works. Eventually, we'll have enough data to permit the use of computer simulations instead of animal testing. But we're nowhere near that point yet. Those who would like to see an end to animal testing, if they were serious, would be in Washington fighting for more support for basic research, or working to create more private sources of research funding. Where are they?

The animal model is wrong, or the theory behind the disease is wrong.

A drug can also fail because the disease does not work the way the investigator thinks. A good example here is Alzheimer's disease. Anti-Alzheimer drugs are tested in a special type of transgenic mouse that produces large amounts of a peptide called beta-amyloid. Apart from the fact that these mice are hideously expensive (over $500 apiece), this mouse model makes the assumption that Alzheimer's disease is caused by a build-up of beta-amyloid. If that theory turns out to be wrong--and there's a lot of evidence suggesting that it might be--the results of all the drug tests that used these transgenic mice are invalid. Any drug based on that theory would fail in clinical testing.

Many diseases that affect humans are so-called lifestyle diseases. Normal lab rats and mice don't get lifestyle diseases, because their diet is far too healthy. We can't feed rats a diet of potato chips, pepperoni pizza and soda--it would be cruel. It would be tantamount to treating them like computer programmers. So it is tough to come up with a good animal model for these diseases. The temptation again is to use genetically-altered transgenic mice. But, as with Alzheimer mice, a transgenic mouse that is a genetically fat and lazy TV-watching slob that never exercises, weighs twice as much as its friends, and writes software that always crashes is the way it is, not because it has the disease we are studying, but because of the genes we put into it. If those genes are not what causes the disease in humans, we are only treating the symptoms, and the most valuable drugs--those that actually cure the disease--are being thrown down the toilet.

The link between high cholesterol and atherosclerosis is another example. There is strong evidence that lowering cholesterol can protect against coronary heart disease. But that doesn't mean we understand why. There's a tendency to think of some biomolecules, like cholesterol, as "bad". But cholesterol is essential for life. It's particularly important for the brain. Drugs that block cholesterol synthesis, known as statins, must be prevented from crossing into the brain. If they did, they would cause depression, interfere with learning, and who knows what else. By treating high cholesterol instead of the real problem, which is whatever cholesterol does to the arteries, we may be treating only the symptoms.

Another example is cancer. In cancer, the preclinical mouse model is a chunk of human tumor tissue, known as a xenograft, growing in immunodeficient nude mice. The mice have to be immunodeficient to prevent rejection of the human cells. But this also means that if the drug works by activating the immune system, its effects will probably be missed. Researchers also sometimes assume all cancers are the same, and expect that a drug that works against one type of cancer will work against all of them. This is not necessarily true.

The pharmacology is done incorrectly.

Careless pharmacology is more common than many people realize. For example, many years ago, some investigators decided to test a class of drugs known as PKC downregulators against cancer. The drugs failed miserably. Yet, from the benefit of hindsight, it was clear that only a few of these studies simultaneously monitored PKC activity. Those who did obtained data that didn't make sense. "Pharmacology research is tough!" --Barbie It turns out that PKC downregulators, at low dosages, and at some time points, have an effect that is opposite to downregulation: they activate PKC. So, did the drugs fail because because the investigators gave the wrong dosage, or because the theory was wrong, or because they didn't fully understand the mechanism by which the drug works? The urgency felt by these early investigators in testing a possible cure, ironically, may have prevented us from learning the answer.

Many times, short-cuts are taken because it's easier. Experimenters don't want to get bitten by wild rats, so they use tame lab rats. They leave out critical controls to reduce the number of animals they need, only to have a flawed study that has to be repeated or is ignored by their colleagues--or even worse, not ignored, and sets research back by several years. In other studies, investigators measure a surrogate marker for the disease because it is easier. High blood pressure and blood cholesterol are often used as surrogates for cardiovascular disease. In these cases, the value of the results depends on how solid the connection is between the surrogate and the disease. In this case, drug companies are usually well aware of this risk. Other times, the risk is not apparent until it's too late.

Companies often include three or more groups in their clinical tests: a placebo, the drug under test, and an existing drug. The existing drug serves as a positive control. It tells you whether the new drug works better than existing drugs, but it also increases costs. It also increases the probability of failure, because if the new drug doesn't work significantly better than the existing drug, it's usually abandoned. That's another reason a drug might fail: it does the same thing as something we already have.

Wishful thinking.

Sometimes researchers engage in wishful thinking. This may be understandable, since they're only human. (Or their boss, who decides whether they remain employed, may engage in wishful thinking.) For example, one paper once claimed that patients on their drug improved their score on some psychiatric test by 4.6±7.1, while the placebo group improved by only 2.1±6.5. You don't have to be a statistician to know that, even with over 400 patients, this is not a real effect.

(Just in case you are a statistician, let me add that the numbers were means±SE. With 200 patients in each group, this means the P-value was 0.79.)

Conclusion

Academic scientists sometimes make disparaging remarks about the drug companies--that their work is not creative enough, or that it's too commercial--but their comments are also mixed with envy, because a drug is the ultimate culmination, or the ultimate refutation, of all disease-oriented research. As John L. LaMattina of Pfizer said, “Only pharmaceutical R&D discovers, develops, manufactures, tests, and demonstrates the properties of compounds that prove or disprove medical hypotheses.” As potential patients, we need the drug companies, and their products, to succeed.

References

[1] Mark Moran, Psychiatric News. http://pn.psychiatryonline.org/content/38/15/25.full
[2] A. Richard Green, Clinical and Experimental Pharmacology and Physiology Volume 29 Issue 11, Pages 1030 - 1034
[3] No similarity is meant to be implied between these two.