Getting a drug from hypothesis to market is almost impossible.

Let’s say you have 1000 drug candidates nudging about petri dishes in a lab. About 100 will survive to be tested on humans. Of those, just seven will become drugs, according to a study published in Nature magazine last year.

Not only does the immense cost of this level of failure create a burden on healthcare systems tasked with subsidising new drugs, but from an investor point of view — the people tapped to help get these treatments through the process — the sizes of the losses are intimidating.

Predicting which compounds will make it through the gauntlet and which will stall is very difficult, requiring experience, broad industry knowledge, and help from software.


Success or failure, it’s expensive

Over the decades-long process it takes to go from lab to a phase three clinical trial, the total cost to make a drug was an eye-popping $US2-3bn ($3-4.5bn) in 2013, according to a US study. It also said that cost was rising by 8.5 per cent every year.

In Australia, little data is available on the cost of drug development.

The only data available is that in 2015 industry investment in local clinical trials was about $1bn.

Most of that money is spent on failure.

The Nature study’s 90 per cent failure rate in the lab is a figure that hasn’t changed in decades.

But it says the chances of drugs making it through the clinical trial stages improved slightly.

From phase one to launch the success rate has gone up from 6 per cent to 7 per cent. From phase two to launch it’s gone from 11 per cent to 15 per cent, and from phase three to launch it’s risen from 49 per cent to 62 per cent.

In Australia, there is no data on how many clinical trials conducted here succeed and how many fail.


Bring in the computer scientists

It’s difficult to assess what might fail and what will succeed, as there are so many variables to consider alongside technical risk — something faced by a company venturing into the unknown like Actinogen (ASX:ACW) did with Alzheimer’s? — and execution risk, as with Innate Therapeutics’ weak clinical plan for how it would monitor Multiple Sclerosis patients.

Opyl (ASX:OPL), a company that uses AI to scour health data from social media, is building software that, it says, can estimate the chances of a trial succeeding between 1 per cent and 70 per cent, based on variables like trial design, recruitment, and site location.

Morgan’s analyst Scott Power says the implications are enormous: that kind of data means companies can better manage their resources, leading to fewer failures, lower costs, faster drug approval times, and more funding that can go to other drugs.

“That’s mind bogglingly important, that potentially can save the industry millions of dollars,” he said.

The Opyl technology is based on information from over 300,000 published trials and more than 60 trial variables across a wide selection of diseases and conditions.

In the US, a collaboration dubbed Project ALPHA between the MIT Laboratory for Financial Engineering and Informa Pharma Intelligence is building something similar, one that will be publicly available with an open-source licence.

The initial study on which it’s based analysed 140 features across 15 disease groups with data from 2003-2015, including trial outcome, duration, and prior approval for another indication to forecast clinical trial outcomes.

What these machine learning programs are going up against is missing data. Project ALPHA senior author Andrew Lo said in the 2017 study that before the FDA changed the rules in 2007 “it was not uncommon for investigators to release only partial information about pipeline drugs and clinical trials to protect trade secrets or simply because there was no incentive to do more”.

But while companies are encouraged to add the outcome to the US database, they aren’t forced to.

In Australia, the Therapeutic Goods Administration (TGA) notes drugs it’s approved and denied, and the Australian New Zealand Clinical Trials Registry is a database of trials generally. Companies don’t need to update their entry with a trial outcome.


Clinical trials explained

There are a few pure R&D companies on the ASX, such as Patrys (ASX:PAB) which has had success in the lab with a treatment for glioblastoma, the brain cancer that killed US Senator John McCain.

But most have one or more treatments in human clinical trials.

Phase one trials are for safety only. The Nature data shows most of those 100 drugs that make it from lab to human fail at this stage.

Phase two is to show efficacy in a small group of people, usually patients who have few other options left for treatment.

Phase three is to prove a drug can work in a large population.

Drugs that flew through too-well-targeted earlier trials can fall over here, as Starpharma’s (ASX:SPL) bacterial vaginosis treatment did in 2012. Chief Jackie Fairley partly blamed trial sites in lower socio-economic areas that included less educated women — in other words, women who couldn’t follow the instructions.

Phase four comes once a drug hits the market and consists of real world data and surveillance of side effects.

In 2018, the powerful US Federal Drug Administration (FDA) halved the dose for women of insomnia medication zolpidem — the one Tiger Woods took to spice up sex with his mistress — after real world data showed the female body metabolises it much more slowly than the male body.


The experts’ rules of thumb

Experience in what to look for in trial design and reading between the lines in company announcements can provide clues to how a trial will end.

Recruitment is possibly the biggest inhibitor to a well run trial as delays in getting the right candidates can wreck timelines and budgets.

“A lot of people get caught out by patient selection, or making sure they get the best patients for the trial,” Morgan’s Power says.

A tighter inclusion criteria can mean it’s harder to find participants but can lead to higher chances of success — which may then cause problems for later trials if the drug doesn’t work in a more varied population, or because it’ll only be approved for a small population.

Bioshares editor David Blake says companies are in between a rock and hard place — they want good results but also the widest application.

For example, the FDA approved a new HIV drug last year, but for use by men only because the drug’s maker, Gilead Sciences, only tested it in men and transgender women. One of the biggest populations of HIV-positive people in the world are women in Africa.

The gold standard is a double-blinded (where neither researchers nor participants know which arm the subjects are part of), randomised (any participant has an equal chance of ending up in any arm of the trial), controlled trial (where one group is not tested at all), to prove whether the drug works against best practice or nothing at all.

To find enough subjects companies may need enough sites — one site may be able to deliver two patients a month but need to consider 300 just to find those two.

A company which says it is adding more sites after recruitment has started is saying it’s struggling to find subjects, and therefore timelines and budgets may not be met.

Blake says a good sign is when a company says recruitment is going as expected. If there is silence on that topic, it could be time to become suspicious.

Opthea (ASX:OPT) is the current case study for how to undertake a well-designed, well-run clinical trial.

It used 113 sites in 10 countries to enable it to recruit 366 participants ahead of time and finished six months early. The company was testing a drug against an eye condition called wet age-related macular degeneration (wet-AMD).

READ: Can the myth hold for Opthea’s scientist CEO?

But there are elements with the trial design that can make the gold standard more or less shiny.

Best practice is to test a drug against a placebo, and many phase two trials do this. But testing against a current standard of care earlier, for example as Botanix (ASX:BOT) did with its recent failed cannabis-acne trial, can help make a useful decision about whether to keep spending on the drug.

Botanix’s acne treatment didn’t perform any better than products already on the market, although it is persisting with a phase three trial for it.

Also needed are enough participants to power the trial.

Pot stock Zelira (ASX:ZLD) may have only had 23 people in its recent insomnia study, but it was a cross-over study meaning the participants are switched throughout to all treatments.

CEO Richard Hopkins says in effect this means they had 46 subjects and therefore enough to deliver statistically significant results.

Good trials, whether they are destined to succeed or fail, will attract star investigators and key opinion leaders.

The clinical trial marketplace is competitive. There are limited numbers of patients who are only available through trial sites.

Opthea had success because there are very few new drugs being tested for wet-AMD so ophthalmologists were very excited about joining.

Anti-viral biotech Biotron (ASX:BIT), however, was trying to find patients for a phase 2a hepatitis C trial and in 2011 had to conduct it in Thailand, because at the time the disease was so popular large companies were sucking up all available patients.

But sometimes even with the best protocols and patients, some trials simply fail.

Coughing app maker ResApp’s (ASX:RAP) observational study of children in the US failed in 2017 because the cough recordings were picking up background noise and protocols weren’t being followed — nurses coughing to give children an example of what to do were being recorded instead.

Actinogen with Alzheimer’s is trying to cure a disease scientists don’t understand yet while Bionomics (ASX:BNO) is still trying to get some statistical data out of trials for a PTSD and anxiety drug — conditions that are extremely difficult to measure an outcome for.