There are three methods to start a project in Simulx: create a new project, open an existing project and import a project from Monolix. To import a project from Monolix is as simple as clicking on the button “Import from Monolix” in the Simulx home tab. It is the easiest way to build a simulation scenario, because everything to run a simulation is automatically ready. As a result, it saves a lot of time time. It is always possible to modify current simulation elements, define new ones and change the scenario so the flexibility of Simulx is not compromised.
- Simulx project structure with “Import from Monolix”
- A typical simulation workflow with a project imported from Monolix
Simulx project structure
Importing a project from Monolix creates a Simulx project with all elements that used in the model automatically generated in the Definition tab.
- Model, population parameters and individual parameters estimates and output variables are imported from Monolix.
- Occasions, covariates, treatments and regressors are imported from the dataset.
Moreover, exploration and simulation scenarios are set and ready to run. They contain one exploration group to simulate a typical individual (in the exploration tab) and one simulation group to re-simulate the Monolix project (in the simulation tab).
Default simulation elements
- model: mlxtran model with blocks [INDIVIDUAL], [COVARIATE] and [LONGITUDINAL].
- mlx_PopInit [no POP.PARAM task results]: (vector) initial values of the population parameters from Monolix.
- mlx_Pop: (vector)population parameters estimated by Monolix.
- mlx_IndivInit [no POP.PARAM task results]: (vector) initial values of the population parameters from Monolix.
- mlx_PopIndiv: (vector) population parameters estimated by Monolix.
- mlx_PopIndivCov: (table) population parameters with the impact of the covariates used in the model (but no random effects).
- mlx_EBEs: (table) EBEs (conditional mode) estimated by Monolix.
- mlx_CondMean: (table) conditional mean estimated by Monolix.
- mlx_CondDistSample: (table) one sample of the conditional distribution (first replicate in Monolix).
- covariates [if used in the model]:
- mlx_Cov: (table) ids and covariates read from the dataset.
- mlx_CovDist: (distribution) log normal distribution of covariates from the dataset with empirical mean and variance for continuous covariates and multinomial low based on frequencies of modalities for categorical covariates.
- mlx_AdmID: (table) ids, amounts and dosing times (+ tinf/rate or washouts) read from the dataset for each administration type.
- mlx_observationName: (table) ids and measurement times read from the dataset for each output of the observation model
- mlx_predictionName: (vector) uniform time grid with 250 points on the same time interval as the observations for each continuous output of the structural model.
- mlx_TableName: (vector) uniform time grid with 250 points on the same time interval as the observations for each variable of the structural model defined as table in the OUTPUT block.
- mlx_Occ [if used in the model]: (table) ids, times and occ(s) read from the dataset.
- mlx_Reg [if used in the model]: (table) ids, times and regressor values and names read from the dataset.
Default exploration and simulation scenarios
- Indiv.params: mlx_IndivInit or mlx_PopIndiv.
- Treatment: one exploration group with mlx_AdmId for all administration IDs.
- Output: mlx_predictionName for all predictions defined in the model.
- Size: number of individuals read from the dataset.
- Parameters: mlx_PopInit or mlx_Pop.
- Treatment: mlx_AdmId for all administration IDs.
- Output: mlx_observationName for all observations.
- Covariates: mlx_cov if used in the model.
- Regressor: mlx_reg if used in the model.
Interface allows to have an overview on all defined elements, modify them and create new ones as well as build simulation scenarios. But, modifying the model removes all simulation elements. In addition, removing occasions removes all occasion-dependent simulation elements.
A typical simulation workflow with a project imported from Monolix
[Demo projects: 1.overview – importFromMonolix]
This example is based on a PK-PD model for Warfarin developed and estimated in Monolix. The Warfarin dataset contains concentration and PCA(%) measurements for 32 individuals, who received different oral doses of the drug. Firstly, the goal of the Simulx project is to use the information from the Monolix project to test the efficacy and safety conditions for different treatments. Secondly, to simulate clinical trials and compare various strategies. Simulations should answer the following questions:
- Which “loading dose” strategy assures a rapid steady state without a concentration peak?
- Do multi-dose treatments meet the efficacy and safety criteria?
- What is the uncertainty of the percentage of individuals in a target due to the variability between individuals and due to the size of a trial group?
The PK model includes an administration with a first order absorption and a lag time. It has one compartment and a linear elimination. The PD model is an indirect turnover model with inhibition of the production. All individual parameters have log-normal distribution besides the Imax parameter, which is logitNormally distributed. In addition, the log-transformed scaled weight covariate explains intra-individual variability of the volume, age covariate has an effect on the clearance and sex covariate on the baseline response. Finally, the combined-1 error model is used in the observation model of the concentration, and the constant model of the response.
0. Re-simulation of the Monolix project
After importing a project from Monolix, the task buttons “simulation” and “run” in the Simulation tab re-simulate the project. After that, plots and results are generated automatically. Results are tables for outputs and individual parameters and plots display model observations as individual outputs and distributions.
1. Exploration of the loading dose strategies
Exploration tab simulates a typical individual. After importing a project from Monolix, individual parameters equal population parameters estimated by Monolix or their initial estimates. Using several exploration groups allows to compare different treatments. The goal of this example is to test how many days of a “loading dose” are neccesary to reach a steady state without a peak of the concentration. Dosing regimens are:
- 14 days with a 4mg single dose OD
- 1 day with a load dose 8mg OD followed by 13 days with a 4mg single dose OD
- 2 days with load doses 8mg OD followed by 12 days with a 4mg single dose OD
“Loading dose” treatments are of manual types (with time of a dose and amount), but multi-dose elements are of regular types. The latter is an easy way to specify a treatment period, inter-dose interval and number of doses. Treatments combination takes place directly in the exploration tab.
Output is the concentration prediction on a regular time grid over the whole treatment period (t = 0:1:336) and the plot displays all exploration groups together. The two-days loading dose gives a high concentration peak and without a loading dose the steady state is reached too slowly. As a result, further analysis focuses on treatments with one day of a loading dose.
2. Treatment comparison: percentage of individuals in the target
Simulation of a population of individuals allows to compare the percentage of individuals in the target for different treatment arms. As before, simulation scenario uses specific treatment and output elements.
- BID treatment: One day of a “loading dose” with 4mg or 6mg dose twice a day (BID), followed by 26 doses of 2mg or 3 mg respectively each 12 hours.
- OD treatment: One day of a “loading dose” with 8mg or 12mg dose once a day (OD), followed by 13 doses of 4mg or 6 mg respectively each 24 hours.
- Outputs: manual type using model predictions (Cc and R) at time equal 336h.
In the simulation tab, the button “plus” adds a new group and “arrows” move the treatment element from the shared section to the group specific section (green frames). Each group has a specific combination of treatments (blue frame). In addition, the option “same individuals among groups” removes the effect of intra-individual variability between individuals (red frame). As a consequence, the differences between groups are only due to the treatment and simulation can use a smaller number of individuals.
The comparison between groups is in terms of the percentage of individuals in the efficacy and safety target:
- Efficacy: PCA at the end of the treatment should be less than 60%.
- Safety: Ctrough on the last treatment day should be less then 2ug/mL.
Simulx treats these types of criteria as binary (true/false) outcomes. It is a post-processing of simulation outputs, which takes place in the outcome&endpoint section of the Simulation tab. Definition of new outcomes includes selecting an output computed in the simulation scenario and post-processing methods.
What is more, outcomes can be combined together (blue frame below), and an endpoint summarizes “true” outcomes over all individuals in groups.
Similarly to the task button “Simulation”, which runs the simulation, the “Outcomes&endpoint” button calculates the efficacy and safety criteria and the number of individuals in the target in this example. Moreover, running the task generates automatically the results in the “endpoint” section and plots for outcomes distributions.
In this example, the outcome distribution compares the number of individuals in the target between groups. The highest number of “true” outcomes is in the group with the low level dose twice a day (gr_BID_lowDose). Failure to satisfy the safety criteria by the two groups with the high dose level causes large numbers of “false” outcomes. In fact, the endpoint results show that for these two groups less then half of the individuals have Ctrough below the safety threshold (blue frame below).
3. Clinical trial simulation and uncertainty of the results
The previous step shows that the high dose level violates the safety criteria in more then 50% of individuals. Therefore, the simulation of a clinical trial focuses only on one dose level with the BID and OD administration. The goal is to analyse the effect of uncertainties due to variability between individuals, measurement errors and number of individuals in a trial.
Simulation scenario has four groups with different dosing regimens (BID or OD) and different group sizes (30 or 100). Outputs are model observations at the end of the treatment. Most importantly, the option “replicates” simulates the scenario several times (green frames). As a result, the endpoints summarize the outcomes not only over the groups but also over the replicates.
[In the demo project, change number of replicates to 100, and add two groups: one with treatment as for GR_BID, other as GR_OD. Move “number of ids” as group specific, and set N=30 for one BID-OD pair and N=100 for the other. Change names to indicate group sizes.]
After running both tasks, plots display the endpoint distribution as a box plot with the mean value (dashed line) and standard deviations for each group. As expected, the uncertainty is lower for trial with more individuals. However, the mean values remain at similar levels.
“Group comparison” option in the Outcomes&endpoints section compares the endpoints values across groups (blue frame). Selecting different reference groups and hypothesis (through the odds ratio) allows to change the objectives. Moreover, calculation of new outcomes or endpoints do not require re-running a simulation because the post-processing is a separate task.
The statistical test checks if any of the dosing regimens, BID or OD, is better then the other. Firstly, in the smaller trial so the gr_BID_N30 group is the reference and the hypothesis states that the odds ratio is different from one. For each replicate, if the p-value is lower then selected 0.05, then a test group (GR_OD_N30) is a “success”. Results summary in the “group comparison” section shows that only 3% of the test group replicates are successful (left image below), which suggests that OD dosing regimen is not significantly better then BID. Larger trial size might give different results. Change reference group to gr_BID_N100 et re-run the outcomes&endpoints task. Percentage of successful replicates for gr_OD_N100 is higher, 9%, (right image below), but it is comparable to the previous result.