Application of Artificial Intelligence & Machine Learning in Clinical Research (Part-2)

There are two ongoing revolutions (one on Technology side and other on Biology side). When they meet then it’s going to be very powerful revolution for Healthcare and Pharma R&D.

"Imagine a world in which we can cure previously incurable diseases by editing an individual’s genetic fabric. The foundational technologies that could make all this possible largely exist. Rapid and ever-cheaper DNA sequencing has deepened our understanding of how biology works and tools such as CRISPR are now being used to recode biology to treat diseases or make crops less vulnerable to climate change. This is what we call the Bio Revolution. ” This is the opening statement in a recent publication from McKinsey.

The McKinsey report estimates that bio innovations could alleviate between 1% and 3% of the total global burden of disease in the next 10 to 20 years from these applications — roughly the equivalent of eliminating the global disease burden of lung cancer, breast cancer, and prostate cancer combined. Over time, if the full potential is captured, 45% of the global disease burden could be addressed using science that is conceivable today. This Bio Revolution has the potential to be as transformative to business and economies as the Digital Revolution that proceeded it, creating value in every sector, disrupting value chains, and creating new business opportunities.

On Technology side, all of us have seen wider adoption of Virtual Trials during COVID-19. We are seeing more and more investments in companies involved in eSource, providing medical care at home, telemedicine etc. Trial designs are getting more and more complex on one hand and we are getting more sophisticated computing, data analytics, and artificial intelligence technologies on the other hand.

Once both of these two ongoing revolutions will converge then miracle is going to happen.

We spend 2-3 Billion dollars over a period of 10 years in Pharma R&D on running a trial and get only one read out from the data i.e. the endpoint which we are testing. We need to think of entire R&D process generating the extraordinarily valuable source of data on what drugs work and what we see about patients during that process because Machine Learning applied to those rich raw datasets can really reduce the R&D slowness, the R&D cost and expedite drug to the market. Now really is the time we need to start making those changes if there is recognition that data capture is worth the cost.

We are seeing changes in Early stages of Drug Development by leveraging AI/ML and even using RWD to identify mathematical methods and what needs to be done in real world of discovery. It’s not going that fast from Phase-II onwards for multiple reasons. Pharma does not have a perfect process for traditional drug development because that talks of reproducibility and in most of the cases we are not able to reproduce.

If we had a clean slate and designing a clinical development process that will keep patients safe and all the patients who could benefit from the drug get the drug then what will that look like. It certainly will not look like what we have today.

Potential application of AI in Pharma.

1.   Drug Discovery: Pharma companies have data of millions of patients from hundreds of trials. AI can help in mining that data and new usage of existing drugs. Computer based proteins using AI data coming from failed and successful trials. Using this to get an idea of potential drugs, binding between receptor and drugs else we will have off target adverse events. That's the base. Disease are not due to one mechanism (Physiology). They are for 2-3 reasons. So we can start developing one molecule targeting 2-3 indications. We can try getting Biological Pathways using AI as we can have multiple allergies say in an Immunology trial because of same biological pathway. Machine Learning has algorithms, which can run on data, but biggest question is “Do Pharma companies have right data collected in right format on which those algorithms can be run”.

2.   Clinical Development: There is an opportunity to start using Real World Data. Trying to get good definition of Endpoints in randomized clinical trials using AI because it takes several months to get Endpoints for drugs related to unmet Medical needs as we don't have any precedence. Using Real World Data helps us to get agreement with Regulators on those endpoints saving lot of time. We can apply virtual drugs on virtual Patient Population through simulation. We might reach to a stage where Regulators are more ready to review outcomes of such virtual drugs applied on virtual patients.

3.   Shift from transactional system to outcome based system: Several governments as well as other organisation are proposing that Pharma gets payed based on outcome of the drug. Today the agreement between the payer and the provider is not very sophisticated as there are lot of variables which we don't understand as RWD is not very structured. AI can help us in making sense from that unstructured data. 25-30% is waste based on transactional model if we switch to outcome-based model and pay for only based on outcome then that 25-30% will be saved.

Food for Thought

·        If Google was running a Clinical Trial then do you think they will measure all data points and later will figure out how to use that data to repurpose the drug or learn something for a new drug. Don’t we think lot more data can be collected in 10 years compared to what we collect in traditional clinical trials.

·        If yes then why is Phama not doing that and is collecting say only 10% of what Google would collect and even for this 10%; experts are saying that too much data is being collected unnecessarily delaying the trials and Pharma can save some of the cost and time by collecting only what is needed in the trials. Are experts saying this because traditional data capture tools, traditional data collection methods, traditional data cleaning methods turn out to be costly (if we look at cost per data point in Pharma R&D v/s cost on other industries). Do we need to get there in terms of economies of scale?

·        Do we need more sophisticated data even on the provider side to show that what they are paying for is what they are getting the outcome? Should we use that as an excuse for not getting started. Can we start moving to crude measures of outcome based payments. For Ex: can we say that this cohort of patients who will take this drug will have 25% less hospitalization as an outcome (this is outcome focused and not volume of service based). Then we say that Pharma will be paid based on its ability to deliver this outcome to the patients.

Let's continue thinking on what all needs to change from Pharma, Patients, Payers, Technology and Regulators end to further accelerate adoption of AI/ML in Clinical Research?

Reference

https://www.mckinsey.com/mgi/overview/in-the-news/the-bio-revolution-is-changing-business-and-society#