Can we predict if we will complete the trial?
I did not know the concept of pharmacometrics till today. It basically considers decision-analysis tools to obtain information about trials that have not started yet. But the term “information” is too broad. For instance, we can think in terms of economic resources, expected trial duration, simulation of results, or whether the trial will be completed or not.
I read an interesting paper entitled “Modeling Successful Phase III trials”. The paper introduces the concept of pharmacometrics and how it can be used to optimize the use of Phase II data to support Phase III success. This is certainly a great idea that enhances communication within a sponsor company, as well as between sponsors and regulators, and provides a fully-informed, scientific basis for decision making. Drug sponsors that have invested in pharmacometrics experience fewer surprises and a more certain path to success—as well as early recognition of failures, for quicker no-go decisions. Even though pharmacometrics have focused mostly on the preclinical-clinical data to measure success of trials, it can be also possible to use this concept to optimize the trial design. Here is an example.
Our research group has developed a statistical model capable of predicting, with an average accuracy of 74%, if a clinical trial will be completed or terminated (note that termination may be due to “good” or “bad” reasons). The information (or data) that the model uses to make decision is 100% based on the trial protocol design. For example, we can predict if a trial that is randomized, double-blinded, and with a certain enrollment target will be completed or not. This type of model can be extremely beneficial for the stakeholders in the CT system. Here is a summary of some of the results we obtained using the AACT database.
Another interesting feature of our study is that our model is capable of determining the designs and factors that are most important for trials in different phases. For instance, our model corroborates that enrollment is the main factor that determines if a trial can be completed or not (red color indicates higher importance). A second factor that is also important for trials in all four phases is “Endpoint”. Also, we can find some factors that play a big role for trials in Phase 1 and Phase 4, but not for trials in Phase 2 and 3. Some of these factors are “number of countries” where the trial is taking place and whether the trial accepts “Healthy Volunteers” or not (see color scale in the table below).
The fact that we can obtain this information by looking at the design of trials only is significantly interesting. If data of trials that are already being performed (clinical data) is introduced to our model, we may obtain even better estimates regarding completion or termination. In such scenario, this can be a tool that can be utilized by sponsors and CROs to continuously test the likelihood of their trials being successful or not, and hence, lead to significant better designs and cost reduction.
We look forward to keep improving this model and getting new findings!