Designing robust clinical trials with the help of AI technologies
Over the past several years, there’s been an exponential increase in the level of clinical trial complexity, ironically amidst technological advancement. Only one in five candidates makes it from phase I trials to FDA approval, and nearly half of the compounds are abandoned before phase III trials for lack of efficacy. Of 5,000 to 10,000 chemical compounds put to preclinical testing, just 10 qualify for clinical stage.10 According to Johns Hopkins Bloomberg School of Public Health, approximately $6 million are spent in Phase I & Phase II, with an $70 million hike at phase III.11 According to Policy & Medicine, this amounts to $2.6 billion spent on drug development, including large spends on failed compounds.3
Here are two intriguing facts:
- By 6th March, 2020, the number of registered studies on clinicaltrial.org went up from 2,119 in 2000, to 325,850.4
- According to J.P. Morgan5, the FDA approved 20–25 new drugs, on an average, per year in the past two decades, however this number has increased by 50% in the last three years.
In 2019, 48 drugs were approved, 13 of which work by a novel mechanism of action. Other drugs cater to long-neglected diseases such as migraine, epilepsy, and endometriosis, or orphan diseases that affect fewer than 200,000 people. Peter B. Bach, director of Memorial Sloan Kettering Cancer Center’s Center for Health Policy and Outcomes is concerned about the increasing focus on rare diseases that affect few people.7
Here it can be mentioned that the timeline for drugs that target complex or rare diseases, the approval rate can be als low as 0.1%.8 For instance, the failure rate for drugs developed between 2002 and 2012 targeting Alzheimer’s disease was 99.6%. Although this might be an extreme example it remains true that for certain diseases the success rate of clinical trials is very low.
With failing faster becoming the new mantra for Pharma, efficient drug development is losing its grip. “For a variety of policy and scientific reasons, companies are finding it more efficient to target conditions that affect very few people but have underlying molecular genetic drivers which are scientifically easier to target—and the time horizon on success or failure is shorter,” says Bach.7
Designing for change:
In reflecting on the distinction between successful and unsuccessful trial designs, several factors appear as essential components that cast the structure of the success or failure of studies. Largely all these factors are in control of the study designers, however they are often overlooked, resulting in many studies which fail even though a drug has a favorable disease outcome.
Only one in five candidates makes it from phase I trials to FDA approval, and nearly half of the compounds are abandoned before phase III trials for lack of efficacy. Of 5,000 to 10,000 chemical compounds put to preclinical testing, just 10 qualify for clinical stage.10 According to Johns Hopkins Bloomberg School of Public Health, approximately $6 million are wasted in Phase I & Phase II, with an $70 million hike at phase III.11
Leveraging AI effectively:
AI technologies can help analyze data for better prediction about which drugs are likely to be successful in trials, thus, cutting costs as well as accelerating drug development. With data from previous trials, learning could help in designing clinical trials to improve success. AI can also help investigators identify and optimize the parameters of the trials.
One of the biggest benefits of leveraging AI technologies is by utilizing outcomes at various stages in the trial and modifying the course of clinical trials, making them more flexible. Such designs will lead to more efficient results. Altering the course of the design corresponding to pre-specified rules can help generate better utilization of resources in possibly less number of study participants.
AI-powered trial comparators can be employed to understand the success and failure causes as well as to translate trials that may potentially fail due to design constraints into successful studies. Early 2019 saw a breakthrough in domain-specific AI when Innoplexus leveraged its Clinical Trial Prediction engine correctly estimating Aducanumab’s failure.15 The Biogen’s drug failure was attributed to some trial sites, execution risks, as well as drug issues. According to Innoplexus’ CEO Dr Gunjan Bhardwaj Innoplexus’s CTP engine could be a tremendous tool (capturing nearly 350 parameters) to plan and predict clinical trials outcome. Implementing the “learnings” coming from the CTP analysis would help to increase the success rate of the respective trials, which would save millions for the industry and help to bring drugs earlier to the patients waiting for it.
Discovery and clinical trials make up the largest share of costs in getting a drug to market. Efficacy gains in these areas would have a direct impact, measurable not only in time to market but also in savings.
AI-enabled efficiency gains can be:
- Dynamic sample size estimations to adapt clinical trial protocols real time
- Monitoring of trial centers and automated re-recruitment in new or better performing sites
- Unified data platforms to allow automated real time data entry QA
- Identification of untapped development opportunities for molecules which have gone through the preclinical phase successfully
- Easily establish new clinical development pathways and clinical protocols