AI could transform clinical trials for the better, unlocking the next generation of drug discovery and building new pipelines for pharmaceutical companies.
Drug research is a long, arduous process. It takes years to get to the point of clinical trials, and one misstep sends companies back to the drawing board. The lifecycle of new treatments tops out at an astonishing $2.9 trillion, so of course, companies are actively exploring a more cost-effective way to get new therapies to market. Artificial technology (AI) is being eyed to help.
Why? Clinical trials are the last stage before approval, so there hasn’t been a way to cut costs. Some studies put the cost of just a single patient at a US-based clinical trial was over $40,000. Cancer treatments are even more. However, recent advances in AI could help reduce the cost of clinical trials in surprising ways.
See also: Why a Data Fabric Could Future-Proof Clinical Trials
A desperate need for change
Right now, clinical trials are a necessary evil for pharmaceutical companies. Some studies suggest that around 14% of clinical trials end in FDA approval, even after thousands, millions, billions of dollars poured into the product lifecycle.
Traditional clinical trials often lack the clinical power to conduct new therapies quickly, putting pharmaceuticals way behind the curve even when new therapies are necessary. COVID put companies in overdrive with a race for a vaccine or a cure, and the process didn’t inspire much confidence once we had a vaccine in our hands.
We have troves of real-world data. Imagine billions of data points just sitting there, hoping for us to connect a pattern. The processing capability wasn’t there before, but now that AI can aggregate this data for valuable insight, there’s a world of possibility for faster, more efficient discovery.
- reduce heterogeneity in clinical studies providing better insights
- come large amounts of healthcare data from insurance, hospitals, treatments, and other sources
- Reduce the instance of side effects and manage cofounding factors
- Reduce time to market
- enable pharmaceutical companies to pivot quickly in times of crisis, something we never thought we’d see.
Using data and AI to find solutions
Artificial intelligence is great at taking massive amounts of data and finding patterns. This pattern-making ability goes far beyond what human observation can uncover, making AI a valuable tool for discovery.
Drug companies could make use of this pattern-finding ability to emulate the last stage of drug discovery. Ohio State University professor and director of the AI in Medicine Lab, Dr. Ping Zhang, demonstrated AI’s capabilities with a deep learning algorithm that:
- Ingested nearly 1.2 million insurance claims from de-identified patients.
- Looked at existing medications to find an unnoticed therapeutic effect on coronary artery disease.
- Found connections between metformin, a diabetes therapy, and antidepressant escitalopram, confirming existing tests that both could be beneficial.
- Identified two drugs that could have an effect if used together, lisinopril and atorvastatin.
These results are exciting. It confirmed that companies made a positive choice to move forward testing the secondary effects of a diabetes and an antidepressant drug for coronary artery disease. It also identified two drugs that worked together to have an impact on the disease.
How the AI algorithm works
The algorithm ingested claims data — assigned treatments, disease outcomes, potential cofounders, and active ingredients in each treatment.
The algorithm is a high throughput, drug repurposing framework. It first factored in the sequence of events from each patient. It started from the first diagnosis code, tracking patients for two years and accounting for confounding factors such as demographics, comorbidities, and co-prescribed drugs.
It compared their diagnoses and treatments, looking for positive correlations between certain drugs and improvements in patient outcomes. The machine then ran mocked up clinical trials for each ingredient found in these patient records.
These simulated trials allowed the research team to estimate treatment effects. Six drugs without CAD indication ranked in the top eight possible new treatments:
The potential of machine learning
Although the team hasn’t tested the algorithm for a true research trial yet, the team demonstrated AI’s true potential. Research has already been a wild success testing structural features of compounds or proteins, for example.
Zhang took it a step further by also adding in long-term, real-world data sources. The phenotypic data from patients provided a robust source of input for the algorithm, which was able to sort out the patterns and find unique new potential from existing treatments.
Health databases are rich sources of information for discovering new uses for old drugs. The problem is that human brains cannot account for all the variables possible. Now, the processing power machine learning brings to the table could use existing information to “discover” new treatments hiding right under our noses.
Researchers helped account for the complex world of health. The team combined casual inference and deep learning to explore connections while accounting for certain biases. For example, drugs prescribed to treat the symptoms of an undiscovered ailment may be blamed for causing that ailment.
Randomized clinical trials cost companies thousands per patient and can’t move fast enough. With global disruptions like Covid putting pharmaceutical companies on the line to discover treatments and cures faster than ever, these initial results are very promising.
Plans for future use of AI
Researchers have already used AI to predict smaller outcomes in high-stakes areas such as the ICU. OSU also trialed an algorithm to predict which would contract sepsis while in the emergency department.
The algorithm is available on Github as an open-source code. The researchers are hoping that collaborators will help further the algorithm’s reach. Traditionally applied to classification, the long short-term memory algorithm is a popular one. With an academic or even industry collaborator, the researchers could scale the algorithm’s purpose and eventually extend it into real trials.
There are substantial cost savings on the line, too. As pharmaceutical companies utilize existing data, we could reduce the time to market for vital treatments. Even better, we may find cures waiting for us in existing therapies. AI could unlock the next generation of drug discovery and build a new pipeline for pharmaceutical companies and researchers.
AI could transform clinical trials for the better. With less heterogeneity, shorter discoveries, and a more efficient process overall, AI applications are far-reaching. It’s another notch in the human-machine partnership.