mcda.drugis.org
User manual

Douwe Postmus, Daan Reid and Gert van Valkenhoef

Introduction

This is the manual for the MCDA user interface for benefit-risk analysis. It starts with instructions about creating a workspace (based on a dataset), followed by a brief introduction to benefit-risk analysis, and a guide to the MCDA user interface itself.

Preparing your dataset

This section only applies to the stand-alone version of MCDA, hosted on https://mcda.drugis.org/. In ADDIS (https://addis.drugis.org/), this step is not necessary.

After signing in to MCDA, you will be redirected to your personal home page. It contains a list of your previously created workspaces (which will be empty until you create one). A workspace is an abstract container for a set of evidence, plus any filterings, transformations and analyses you may have performed for that set of evidence. Any unfinished workspaces in the process of manual entry will also be shown here. Clicking the 'Create workspace' button will open a dialog that lets you choose what to base your workspace on: an example dataset selected by us, a file uploaded by you, or manual entry by you.

Data upload

Datasets can be uploaded in JSON format. The JSON should contain the following entries (note that JSON is case-sensitive):

  • title
  • description
  • criteria
  • alternatives
  • performanceTable
  • preferences

The title and description properties should contain text.

The criteria property should contain a series of properties with the names and details of the different criteria in your analysis. The following is an example of JSON describing a single criterion (HAM-D):

              "criteria": {
                "HAM-D": {
                  "title": "HAM-D",
                  "scale": [0.0, 1.0],
                  "unitOfMeasurement": "Proportion",
                  "description": "Responders on the HAM-D rating scale",
                  "pvf": {
                    "direction": "increasing",
                    "type": "linear",
                    "range": [0.0, 1.0]
                  }
                }
              }
            

The alternatives property should contain the alternatives under investigation. For example:

              "alternatives": {
                "Placebo": {
                  "title": "Placebo"
                },
                "Fluox": {
                  "title": "Fluoxetine"
                },
                "Parox": {
                  "title": "Paroxetine"
                }
              }
            

The performanceTable property should contain the actual data of the measurements for each criterion for each alternative. Performance can be measured either in relative or exact terms. If relative, the performance should contain a baseline plus a matrix of relative deviations. If exact, simply the value suffices.

An example of exact performance:

              "performanceTable": [
                {
                  "alternative": "Placebo",
                  "criterion": "Hypo",
                  "performance": {
                    "type": "exact",
                    "value": 0.02
                  }
                }
              ]
            
An example of relative performance:
              "performanceTable": [
                {
                "criterion": "HAM-D",
                "performance": {
                  "type": "relative-logit-normal",
                  "parameters": {
                    "baseline": {
                      "type": "dnorm",
                      "name": "Placebo",
                      "mu": -0.17143575198943867,
                      "sigma": 0.11299261988783152
                    },
                    "relative": {
                      "type": "dmnorm",
                      "mu": {
                        "Placebo": 0.00000,
                        "Fluoxetine": 0.4718129,
                        "Paroxetine": 0.7258847
                      },
                      "cov": {
                        "rownames": ["Placebo", "Fluox", "Parox"],
                        "colnames": ["Placebo", "Fluox", "Parox"],
                        "data":
                          [ [0.00, 0.00, 0.00],
                            [0.00, 0.013445532, 0.010394690],
                            [0.00, 0.010394690, 0.023006616] ]
                      }
                    }
                  }
                }
              }]
            

The preferences property should generally consist of an empty object. This will be filled in by performing preference elicitation within the MCDA app.

A list of complete examples is available on the MCDA repository.

Manual entry

Selecting the 'Manual entry' option in the workspace creation dialog takes you to the manual workspace creation screen. Here, you can input the title, therapeutic context, criteria, and alternatives as well as the measurement data.

Manual entry, first step.

Alternatives are nothing more than a name, therefore creating one is as simple as typing and clicking the 'Add alternative' button (or hitting enter). Criteria are more complicated. They have several properties , e.g. whether or not they are favourable, what the source of your data for this criterion is and the unit they are measured in.

Manual entry, creating a criterion.

Creating a survival criterion also allows you to set the specific summary measure to use. This can be the mean or median survival, or survival at a specific time point. Note that the unit of measurement of a criterion may change automatically (e.g. it becomes 'proportion' when indicating the criterion is dichotomous) but can always be changed manually afterwards. The data source selection determines how data is entered in the next step. At any time during the manual entry process you can click the 'Save' button and the current state of your workspace in progress will be saved. You can click this in-progress workspace from your personal homepage later, and continue where you left off.

Manual entry, second step.

The second step of the manual entry process is where you input the data for each combination of criterion and alternative. Data is entered for each cell by clicking on it and filling out the cells in the dialog that appears. Depending on the type of the criterion and the data source selected, this dialog will have different fields, (e.g. mean, standard deviation and N for continuous variables, or simply a value for exact value distributions).