Our advanced predictive models can now be used to evaluate and optimise the clinical trial design for the comparison of any two drug regimens. The user simply selects the treatment regimens and the proposed trial design (slide the green bars to change!), and then hits the "Simulate" button. When the simulations are completed, numerous results are then generated, including the Probability of Technical Success (i.e. P<0.05 in the primary endpoint analysis).

Read More

The outcome of a clinical trial depend on the drug regimens being compared and the design of the trial. That is, the treatment effects, and the treatment difference (the "delta"), are not fixed values, but depend on multiple aspects of the trial (e.g. treatment duration, baseline HbA1c, patient population etc.). The trial design information provided below is combined with our advanced HbA1c and weight models, and 100 trials are simulated across a range of different sample sizes (yielding 100 "virtual" trials). The code returns the % of successful trials (i.e. P<0.05 in favour of the experimental treatment), the predicted treatment effects (i.e. changes from baseline), and the predicted treatment difference. Additional options can be found under the "Advanced" button.

If the Simulator seems to be unresponsive, please refresh/reload the page to reset everything back to normal.

Historically the sample size for clinical trials have been based on power statements for HbA1c such as

"Assuming a treatment difference of 0.3%, a total of 230 patients per arm are required to provide 90% power to demonstrate superiority."

The problem here is the very large assumption regarding the true treatment difference (the "delta" of 0.3%). What evidence is there for it? If the true difference is much larger, the trial will recruit far more patients than are necessary, incurring additional costs and time (e.g. read our LinkedIn article regarding the semaglutide SUSTAIN 8 study here ). Worse still, if the true treatment difference is smaller than expected, the trial may not yield a positive outcome at all (e.g. DURATION 6, where 900 patients failed to show superiority of exenatide 2 mg weekly versus liraglutide 1.8 mg daily). Thus 90% Power ≠ 90% Probability of Success (indeed "90% Power" may equate to a 0% Probability of Success). In addition, the power statement ignores how the design (e.g. the treatment duration) can influence the "delta". For example, 14 mg of oral semaglutide would find it very difficult to show superiority over glyburide over 16 weeks (even with a total sample size of 800), but would have a very high Probability of Success (90%) with only 300 patients if studied over 52 weeks (with a baseline HbA1c of 8.5% in patients on background metformin), since the HbA1c "delta" is much larger at 52 weeks compared to that after only 16 weeks.

In summary, it is time for a more "science-based" approach using "Probability of Technical Success" to both evaluate and optimise our clinical trial designs. This "Model-Informed Drug Development (MIDD)" approach will save time, money, and lead to higher success rates in our trials.

In addition to further graphics/metrics of the simulated "virtual" trials (for both HbA1c and weight), the plan is to extend the tool above to incorporate hypoglycemia and (initial) combination therapy.

Read More

The developed models for both HbA1c and weight have been successfully extended to describe and predict (initial) combination therapies (using 68 combination arms across 41 studies), allowing novel combination therapies to be simulated and evaluated/optimised. Thus an important goal is to extend the simulation engine to show predictions for both current and novel combinations.

At the moment, the code above only runs 100 "virtual" trials, as the goal is to demonstrate this particular application of our models to improve drug development. However the code has been successfully tested using a full 5000 "virtual" trials, and can be refined further (e.g. allowing the user to specify their preferred estimate of the Standard Deviation (SD) used in the simulations). We hope you can see how this work will lead to smarter, leaner and more successful clinical trials.

If you have any comments or questions on any of the above (e.g. how we can improve the Trial Simulator, any bugs!), or would like to know more, please do not hesitate to contact us!