True vs. Observed Effect – Part One

At Captario, we are doing some pretty cool modeling for our clients, which we call effect modeling. The idea originates from the fact that efficacy is the most significant risk in drug development and that most terminations are due to a lack of efficacy. With effect modeling, we can predict PTRS and support decisions about clinical trial design.


Read also: True vs. Observed Effect – Part Two

How It Works

Every compound has an inherent efficacy that we try to find with clinical trials. However, because of small sample sizes and variations in patient response, what we see in a study is not necessarily the true effect. Therefore, in our predictive models, we create two distributions to represent true and observed effects:

  • The true effect is modeled based on what we know about the compound.

  • The observed effect is naturally based on the true effect, but we add an error factor based on what we know about the study and patient group

These two effect variables are correlated but are not the same.


We then run simulations to sample data, and the chart shows an example of the correlation between the true and observed effects after 10,000 simulations.

The colors indicate two things: On the x-axis, the color shift shows the threshold we need to reach for this to be a viable product. On the y-axis, the color shift shows what we need to see in the trial to declare success. This is the threshold we set up as drug developers, so it is our decision. Remember, what we see is not the same as the truth!


This means that samples in the lower left and upper right in the chart indicate that we made the right decision based on the trial outcome. Dots in the upper left show instances where we see a positive result, but the underlying compound is inferior. And dots in the lower right indicates that we are incorrectly terminating a promising compound.


If we run 10,000 simulations, and 9,500 of them end up in the lower right, we know we should probably change something in the trial design. Perhaps add more patients to get a result that is closer to the truth. Or maybe lower the threshold for success, allowing more compounds to be tested in the next phase instead of terminating now.


Conclusively, this modeling type can help a drug development team understand the dynamics of their clinical trial design!