However, AI can help predict these results much earlier in phase I or II, helping to avoid unnecessary work on the ineffective medications that can occur when using current efficacy predictors available today.
“The way that we predict if drugs are effective earlier on in the pipeline is through what are called surrogate endpoints, and they’re not always reliable predictors of efficacy,” said Mitro. “There are many instances of notable trials where surrogate endpoints have underestimated or overestimated the actual efficacy of the drug.”
Mitro goes on to say that there are issues with both overestimating and underestimating drug efficacy. If drug efficacy is overestimated by surrogate endpoints, it may eventually be pulled off the market when it inevitably fails to demonstrate effectiveness, and if it’s underestimated, it may never reach the hands of those who need it – both of which, she says, are “expensive decisions.”
AI can utilize data and pattern prediction to reduce these over- and underestimations. It could even go back in time and find out if other underestimated drugs were overlooked far too soon.
“You could go back today, and you could apply our AI models on a drug or on a clinical trial that failed, and you could potentially see, was this actually an effective drug? If it was an effective drug that the study sponsor just decided not to move forward with because, according to their other information, it didn’t seem effective, maybe that’s an opportunity to pick it back up,” said Mitro.