In Search of Pharma’s Moore’s Law

Will AI usher in a new era of high productivity for clinical trials?

By Sagar Anisingaraju on Jul 5, 2018

Artificial Intelligence (AI) has made great strides in the past five years. We are probably using an AI-based system without even realizing it in our daily lives today.

There is a lot of active research going on in AI in both academia and in the pharma industry. Within pharma, several opportunities open up as we think of AI making an impactful difference; lead candidate identification, drug development, compound repurposing, clinical trials, post approval, commercial markets, etc. Let us look at one area closely, that of planning and conduct of clinical trials.

Costs of bringing a drug to the market over the last three decades.

Moore's lore
In my view, if there is one area that is ripe for disruption, it is this one. With the rising cost of clinical trials and associated processes, drug development costs and time have exponentially gone up over the last thirty years, from approximately $150m in 1970 to about $2.6bn this decade.^1,2

In 1965, Gordon Moore from Silicon Valley famously stated that the number of transistors in an integrated circuit would double every year, a phenomenon known as Moore’s Law. Collectively, the semiconductor industry has proved him correct for several decades, dramatically increasing in power and decreasing in relative cost at an exponential rate.

What is pharma’s own Moore’s Law? Can AI-enabled technology help pharma to continuously ‘learn’ from its own clinical trials and that of peers to bring recursive savings in both time and cost? Can shared learnings help reduce the safety and efficacy issues inherent in drug development? Can we use such recursive learnings to bend that cost curve of drug development over the next decade while improving safety?

The answer is a qualified ‘Yes’. A potential equation for achieving pharma’s version of Moore’s Law can be thought of as: Clinical trials time & cost = e^{-(dv/dt. Min(clinical trials))}.

The below figure shows the impact of a 20% year-over-year improvement in costs over a decade.

Pharma’s Moores’s Law to bend the cost curves over next decade.

How to make the above even remotely possible? Where do we start? Just like in the semiconductor industry, it will take the collective intelligence of the pharma industry, data sharing for research and the ecosystem, to bring the curve down.

We can discuss a comprehensive digitization and data sharing platform for research in later article but, here, let us review some specific areas where the pharma industry should start thinking of applying AI, and potential implementation approaches.

Predicting success
During the clinical trial planning phase, an AI engine can first start with an understanding of inclusion and exclusion criteria defined in the natural language. It can then find clinical trials that are currently running to match the basic criteria and identify historical trials from public and syndicated sources, such as clinicaltrials.gov.

Collective insights generated from these historical trials with natural language understanding (NLU) offers huge value for modifying current and future protocol designs. Insights from past failed trials can be very useful to avoid repeating common errors, omissions and safety concerns.

For example, the shared insights above can be used for identifying principal investigators. Choosing the right principal investigators for a site is a key decision for a clinical trial. AI-based NLU engines can mine public literature to find out about principal investigators, categorize and rank them in their area of research.

Combining this intelligence with insights gained from AI-based mining of operational data can be used to predict the likelihood of a site’s success in running a clinical trial.

Site selection, principal investigator identification, and patient matching will all get an uplift from these AI models. The presence of protected health information (PHI) has often been a hindrance to effective use of patient data for clinical trials. AI can help in appropriately scrubbing the PHI elements.

Needle in an AI-stack
Only 5% of patients in the total pool of patients take part in a clinical trial. Identifying patients for a clinical trial is like the problem of finding a needle in a haystack.

Traditionally, structured data from real-world evidence datasets are used for patient identification. Relevant entities like diagnosis, procedure, lab values, etc. can be extracted from inclusion/exclusion criteria to match with structured and unstructured real-world evidence datasets.

An AI-enabled PHI scrubber can make doctor notes a lot more effective for patient matching. Nudge networks or patient influencer physician networks can be identified for better enrollments.

Data standardization for shared use has also been a plaguing problem. AI systems can solve this by being taught to ‘learn’ rules from the data.

Finally, the conventional experience of querying and reporting the complex clinical data streams can also get an uplift from current AI technologies. Google has recently shown us how an AI-enabled virtual assistant such as Duplex will bring productivity gains in our daily tasks such as scheduling appointments, restaurant reservations, etc. With the available AI technology in the open domain, we can easily build intelligent virtual assistants to help study managers and other clinical trial personnel generate and act on contextual insights.

Huge productivity gains are in offing for the pharma industry with these intelligent virtual assistants, ranging from clinical trial monitoring, safety, and event management to supply chain notifications and other areas.

The above examples demonstrate how AI techniques can be implemented for productivity gains across clinical trial processes and, as a result, liberate the clinical data across the pharma ecosystem for potential cost and time savings.

However, the pharma industry, academia, the FDA, and technology companies such as Saama need to actively collaborate and share these efforts collectively to bring pharma’s own Moore’s Law to fruition.

References
1 https://csdd.tufts.edu/csddnews/
2 https://www.forbes.com/sites/matthewherper/2017/10/16/the-cost-of-develo...

Sagar Anisingaraju is Chief Strategy Officer at Saama Technologies.