Cognitive Intelligence, Interviews, Machine Learning

AI Application in Clinical Trial Process Optimization – Bayer Interview

Kevin Hua

Kevin Hua
Sr. Manager A.I./Machine Learning Development – Bayer Digital Innovation US / iHub, Bayer

Kevin Hua

Kevin Hua
Sr. Manager A.I./Machine Learning Development – Bayer Digital Innovation US / iHub, Bayer

Prior to Cognitive Health in November, we.CONECT spoke with speaker Kevin Hua, Sr. Manager A.I./Machine Learning Development – Bayer Digital Innovation US / iHub at Bayer.
Being trained as an AI researcher, Dr. Hua has worked in the industry for over twenty years on application of AI/Machine learning to various real world problems, such as in accounting, healthcare, insurance, banking, retails and manufacturing industries. Pharmaceutical is an industry where AI/Machine learning was not as widely used as in other industries. The Pharmaceutical industry offers a lot of opportunities for AI, for example in clinical trials or early stage development of drugs. It is fascinating to see how much we can improve the way we have been doing drug development for over 100 years. As a Sr. Manager of AI/ML at the Digital Health Intelligence group, Dr. Hua leads a few initiatives on machine learning and data-driven decision making for clinical trials. The group has made significant progress in helping transform the firm into a digital enterprise, especially with promoting AI/ML in clinical trials.

we.CONECT: Kevin, one of the first topics you will address during your presentation are the difficulties in clinical trial planning and scheduling due to uncertainties. In your opinion, which uncertainties are most challenging and the biggest obstacles?

Kevin: The two main uncertainties are time and cost. It is difficult to estimate the exact duration of each cycle of a clinical trial. Neither can you forecast exactly how much a trial is going to cost. The nature of this process is stochastic. In fact, if you conduct the same trial again and again, each time it will take a different amount of time and money. The biggest obstacle is that the trials are not independent. Delays of one trial may have a cascading effect on delays of many other trials because of resources constraints. Similarly, budget shortfall in one trial may result in delay or cancellation of other trials. It is a system/global optimization problem.

we.CONECT: Your company has been approaching the topic of Clinical Trial Optimization with advanced analytics on historical clinical trials. Some questions to that:

1) How many trials, would you say, would a company need to actually come out with reliable results?

Kevin: In statistics, a common belief is that you need at least 30 data points to derive a distribution. However, based on Bayesian theory, any amount of data can help. It is just an issue of the level of confidence you will achieve. This is also area that machine learning models can help.

2) What were the key risk factors you were taking into account for your analysis on potential for optimization?

Kevin: Sites and principal investigators are among the top risk factors. Some other key risk factors cannot be optimized, such as disease type. They are the property of the trials.

3) Was all the data in the format you needed to apply your analytics tools, or did you have to work on the formatting before the actual analysis?

Kevin: Data is always a challenge in most machine learning projects as for most problems the data collected is usually for other purposes than for machine learning. Transforming the data and mitigating the missing data or outliers are necessary before any analysis.

we.CONECT: You applied Monte Carlo simulation to help estimate benefits of optimization strategies. Did you find significant deviations when comparing local to global level? And how did this affect your strategy building?

Kevin: For schedule optimization, the goal of Monte Carlo simulation is to identify critical paths of the processes. The problem may be over constrained when allocating the risk coverage from local to global. Our solution is to relax the constraints based on how critical a path is.

we.CONECT: How exactly did you and your team apply machine learning in your organization?

Kevin: There are many places where you can apply machine learning models in an optimization problem. For example, you can use models to derive potential risks of new sites you have never worked with before.

we.CONECT: Can you name some benefits that derived from the application of AI in the Clinical Trial process?

Kevin: One major benefit of AI/ML is that we can make historical clinical trials’ data useful for scheduling and planning for future trials, so that clinical sciences teams can make their decisions more accurately. AI and Machine Learning help transform pharmaceutical companies into data-driven enterprises.

we.CONECT: Your case study at Cognitive Health is referring to optimization through machine learning in Bayer’s cardio vascular clinical trials. According to your knowledge, is this currently the only area where machine learning techniques are being applied?

Kevin: In fact we used data for all therapeutic areas. We tested Cardio Vascular trials separately because we have more trials in this therapeutic area.

we.CONECT: What other areas would you see as a potential fit for optimization through AI application (machine learning and beyond)?

Kevin: In the pharmaceutical industry, there are many other areas where we can apply AI/Machine Learning besides clinical trial process optimization, such as drug candidate selection, drug repurposing, prediction of potential adverse effects, demand forecast, marketing effectiveness analysis or even optimization of chemical synthesis processes or supply chain.

we.CONECT: What expectations do you have regarding the Cognitive Health event? Which outcomes and benefits do you expect to gain from the exchange with the participants?

Kevin: I would expect to meet major players in the industry who are applying AI/ML to health or drug developments problems. I would also hope to see participants from academia, who are working on basic research on AI/ML. I would hope as a community we can identify a few areas where we can apply AI/Machine Learning and make a bigger impact to the industry.

we.CONECT: Which topics or questions are you striving to discuss or get answered by Cognitive Health attendees?

Kevin: As it has been discusses a lot in many occasions, data is a big challenge to many AI projects. I would like to know if it is possible to set up a consortium by the industry to create a common platform and share data among the members.

we.CONECT: Thank you Kevin for taking part in this interview.