KerusCloud User Guide

Introducing KerusCloud

KerusCloud is a clinical trial simulation software that has been designed with a logical, linear workflow and modular design to optimise user experience, application performance and memory usage. There are three key functional tasks, shown in Figure 1 below:

Figure 1: Work-flow schematic

1. Virtual Population Creation of a Virtual Population
2. Study Design and Analysis Specify aspects of the study design including caps on recruitment from specific sites and adaptive designs and required analyses.
3. Decision Criteria Summaries of the analyses results to give a probability of success based on specified decision criteria

Enter the information you need, when you need it

Due to the modular design of KerusCloud, Virtual Population can be constructed and run prior to setting up the features for the Study Design and Analysis or Decision Criteria. Or you can create just a Virtual Population and download the data for your own bespoke analyses

Task

At each stage a KerusCloud task is run on the cloud computing service on the pushing of the Go button; this action makes the system generate and save output files. Each task has a prerequisite of the previous stage and cannot run unless the previous stage(s) have successfully completed. If an intermediate stage is changed (e.g. Study Design and Analysis) it is not necessary to re-run the tasks for the prior stages, however tasks for the subsequent stages which are dependent on the stage which has been changed need to be re-run (by clicking on the Go button again) to ensure KerusCloud uses the updated parameters and files.

KerusCloud facilitates three speed options when running Virtual Population, Study Design and Analysis and Decision Criteria tasks. You can review the credit usage for each option and select the one that best balances speed and cost for your needs

Speed Option	Baseline Computation Power	Speed Gain	Example Task Time
Eco	0.25	0.25	240 minutes
Normal	1x	1x	60 minutes
Fast	4x	4x	15 minutes

• Baseline computation power is based on 16 cores on an AWS EC2 instance.**Speed gain for each option can be up to a maximum of the stated factor.

Find Out More

To find out more about each KerusCloud Task, click into the relevent section below:

Virtual Population

The Virtual Population is created through identifying variables of interest, selecting a distribution for each variable, and then entering parameters to define the distribution. For example, a variable of normal distribution can be created by entering parameter values to represent the mean and standard deviation of the variable.

Once distribution parameters are defined for all variables then a request for a simulation will randomly generate observations from the defined distributions to create a Virtual Population for each simulation specified within the project set-up.

Note

A unique feature of KerusCloud is the ability to additionally specify a correlation structure between variables whatever their marginal distribution. If no correlations are specified, then the data for each variable will be independently simulated without any consideration for how values of one variable relate to values for other variables.

Variables can be generated with a range of parameter estimates corresponding to different potential scenarios that might be expected to occur in the “real world”. For example, there may be uncertainty over the variability of a normally distributed variable and hence data may be simulated from a distribution with a low standard deviation as well as from one with a higher standard deviation. These scenarios are labelled separately within KerusCloud so that analyses can be carried out independently on each set of simulations corresponding to the various scenarios.

Once the Virtual Population has been fully defined, pressing Go on the Virtual Population Review page results in the generation of a large dataset of simulated data.

Study Design & Analysis

Statistical analyses are defined in functional task 2. Multiple analyses can be specified, and each analysis may include any variables that have been generated in stage 1. Stage 2 first defines the sample size or range of sample sizes that the analyses are to be based on. The observations used within each analysis are randomly selected from the overall Virtual Population. The specified analyses are then run for each simulation on each scenario, based on sample sizes defined in the analysis function. Once the Analyses have been defined, pressing Go on the Analysis Review page results in the running of the analyses.

Decision Criteria

Functional task 3 is the Decision Criteria section and resulting probability of success. The results across all simulations for each scenario and sample size selection are summarised to produce probabilities for certain outcomes. For example, the required outcome may correspond to significance of treatment within a hypothetical clinical study for a variety of sample sizes and scenarios. Multiple decision criteria can be defined based on different required outcomes. Once all criteria have been defined, pressing Go on the Decision Criteria Review page results in the collation of results into probabilities of success for study outcomes.

The primary output of KerusCloud is a heatmap, which displays the probability of success for every combination of scenarios from the Virtual Population (variables, correlations, advanced options and estimands), Design choices (sample size and allocation), Analysis (fixed or adaptive) and Decision Criteria. The heatmap is colour-coded and can be filtered to allow identification of the optimal combination of design features.