 Hello, my name is Klaus Detlefsen and I work at the Danish pharmaceutical company Novo Nordisk. On behalf of the group of developers, I'll present a tool for communicating, determining and documenting sample size in clinical trials. It's the shiny app we call NN Sample Size. Please note my disclaimer that views and opinions expressed are those of the speaker, not necessarily Novo Nordisk. When we are going to determine the sample size in a clinical study, it typically goes on like this. A medic comes to the physician with some assumption and a trial design and wants us to calculate the number of subjects needed to show the expected effect sizes. We go back, calculate the numbers and return to the medic who may again discuss and come up with new choices of parameters and then the process reiterates. But why is it important? So we need to strike a balance. The regulators demand statistical significance to have claims in the label. But it's also costly in terms of time and money to include too many subjects. So we also need a strike a balance on the ethical side. It's unethical to expose more than needed because of unwanted side effects that may be for the experimental drug. But it's also unethical if we are not able to detect a clinical relevant effect size. It's an efficient chance of detecting the relevant effect size. It's, in principle, very simple to calculate the number of subjects needed. Here is a call to the function power.t.test in R where we have set the treatment difference to be one standard deviation to be 1.2 and has chosen a power of 90% and implicitly the significance level is set to 5%. These are related to the probability of making a type 1 error and 1 minus the probability of making a type 2 error. In this case, we need 32 subjects in each of the two groups we are comparing. Here are some types of clinical trials. On the left-hand side, we see a parallel group design, which may be event driven so that subjects are randomized into one of two treatments and the trial terminates when you have sufficient number of events. We also have a crossover design on the right-hand side where subjects are randomized into a sequence of two treatments. In reality, it gets complicated to calculate the number of subjects in a trial. For example, inclusion-exclusion criteria may affect the rate of events in a trial. Duration of treatment may affect the effect size and the choice of comparators. The expected effects and variation need to be considered when calculating the number of subjects needed. Also, if you have more than one hypothesis to be tested, you need to have multiplicity considerations. And then the amount of missing data and the patterns of missing data are important also. So the key features of our Shiny app in ensemble sizes to have access control so we can limit who can see the calculations also within the company. They may be confidential. Then there's the ability to make dynamic visualizations on how the parameters affect the sample sizes. And then finally, auto generation of documentation so it's easily updated and ensures we can reproduce the results and trace it back and make it transparent. Here's an example of how the sample size app works. At the top, you see how the number of subjects needed changed when you changed the assumptions on the left-hand side. You can also click on the documentation button to download SAS program or web documentation of the sample size calculation. In the next example, we have a parallel group designed with two hypotheses that are adjusted for multiplicity in a hierarchical testing strategy. And again, you can change the numbers, the assumptions, and you can follow in the top line how many numbers are needed to be screened and randomized and what is the effective power of the whole hierarchy and what is the power of the primary point. The source code is available on GitHub and you're welcome to comment on it. Thank you very much for your attention.