One step closer to no CTs: Predicting success of clinical trials and marketing approval.

Pharmaceutical companies face huge costs in order to launch a new drug to the market. These costs are associated with expensive and timely clinical trials that seek to prove safety and effectiveness of a new drug. In particular, Phase III of a clinical trial accounts for the biggest share of cost. The Tufts Center for the Study of Drug Development (CSDD) estimated the cost of developing a prescription drug that gains market approval at $2.6 billion, a 145% increase over the estimate the Center made in in 2003. Adding an estimate of post-approval R&D costs increases the estimate from $2.6 billion to $2.87 billion (2013 dollars). Several studies have also estimated the likelihood of approval of a new drug or device. These estimates change, for instance, depending on the disease or type of molecule studied. However, on average, new drugs have an 8% – 10% chance of approval, for all indications. Drugs which fail part-way through this process often incur large costs, while generating no revenue in return.

Despite decades of experience conducting clinical trials and vast improvements in knowledge about disease mechanisms, drug development professionals have been unable to improve the low clinical approval success rates for new molecular entities (NMEs) and new biologic entities (NBEs). Therefore, the need to increase new compound success rates and reduce drug development risk has intensified in the last decade. As described in a previous week note (week Note 2015 – 25), predictive models can be an approach to tackle this issue. The model that was described in the above mentioned week note was able to predict completion of clinical trials with a relatively high accuracy. However, we were not able to relate completion with success of approval. This is a huge gap since no such algorithm exists for assessing the probability of success (POS) for regulatory marketing approval (note that these models are widely used in other areas of healthcare).

To tackle this gap, our team is in the process of developing a new version of our prediction model which will actually be able to predict whether a trial will be approved or suspended, for any phase and disease group. Also, the model will be extended to predict marketing approval for a drug considering its past clinical trial history. Despite the fact that obtaining the likelihood of approval can serve as a cost reduction strategy for pharmaceutical industries (and hence reduce the cost of drugs in the long run), we will be able to obtain the factors that are associated with success of drugs, and hopefully, we will be able to improve the design of clinical trials.

The table below shows preliminary results of the model under construction for predicting approval of phase III trials. In order to validate results, we compared two distinct algorithms; Random Forest and Support Vector Machine. Random forest shows a slightly better performance (based on accuracy). We are able to predict with almost 90% of accuracy (sensitivity) those trials that are approved in phase III and with almost 80% (specificity) trials that are suspended.

In this preliminary study, some of the factors that seem to be highly associated with obtaining FDA approval are the target enrollment, number of countries, number of arms, disease group, pivotal, and whether the trial has a data monitoring committee.

Results so far seem promising. However, there is much more to do. Here is a “to do list” developed by the team:

  • Categorize by disease group and indication. Then we could see which disease groups do better and why. What phase do they get caught up, what are the characteristics of the trials that get suspended for each disease group?
  • Pull out Phase II and Phase III for autoimmune, endocrine, and respiratory- What happens to create the big jump up in phase success rates (from 2 to 3) that have been described by other authors and studies?
  • Find what goes wrong in Phase II and/or III for oncology and cardiovascular disease? They have the lowest phase III success rates, why?
  • Look at time to approval or suspension. Can we see which trials take longer and the time to approval? Look at the drugs that get suspended at Phase III or NDA/BLA to identify some common characteristics.
  • Do trials that use biomarkers including genetic markers actually have higher approval rates?.

Felipe Feijoo, PhD