About OHDSI Viewer
OHDSI Analysis Viewer
Table of contents
Introduction
This is an interactive shiny app for exploring standardized outputs for OHDSI analyses including:
- characterization (descriptive studies)
- population-level effect estimation(causal inference)
- patient-level prediction (inference)
Full details of all the analysis tools can be found on the HADES website
How to use the viewer
Please use the left hand menu to select the type of analysis to explore (click on a button). This show the results that can be interactively explored.
Analysis types
Characterization
The OHDSI community have developed a suite of tools for conducting characterization studies including:
- incidence rate calculation
- baseline characterization
- treatment pathways
- and more
Population-level effect estimation
The OHDSI community have developed several packages that enable users with data in the OMOP common data model to perform causal inference studies.
Patient-level prediction
The OHDSI community have developed several packages that enable users with data in the OMOP common data model to develop and validate patient-level prediction models.
Prediction Viewer
Model Designs Summary
This shows the different model designs that have had models developed or validated.
Each model design has a model type (e.g., logistic regression or decision tree), a target cohort (the patients the model is developed to be applied to), an outcome (what is being predicted) and a TAR (time-at-risk when the outcome is being predicted). The AUROC summaries give the min/mean/max performance for the model design across multiple databases. The number of databases tell you how many different databases there are where the model design was diagnosed, developed or validated.
- Click 'View Diagnostics' to see the diagnostic results for the model design across databases (this highlights any potential issues in the design that may cause bias).
- Click 'View Results' to see the summary results for all the development/validations across databases.
- Click 'View Report' to view a summary report containing all the results for the given model design.
All Database Results For Selected Model Design
This shows the summary model development or validation performance results for the selected model design across databases.
Click 'View Result' to explore the performance for a specific database in more detail.
Full Result Explorer
This shows the settings and results for a specific model design when model development is done in the selected database or validation of a model with the model design is done in the selected database.
- Design Settings: shows the settings used for the model design - it is possible to reproduce the model development with these setting and the database.
- Model: shows the covariates in the model, the variable importance (value) and how often the covariate occurs for paients with and without the outcome.
- Threshold dependant: A plot showing the sensitivity, specificity and precision for different cut-offs. There is also an interactive explorer.
- Discrimination: You can view the test/train/cross validation (or any subset) discrimiantion performance here. Select to see the plots.
- Calibration: You can view the test/train/cross validation (or any subset) calibration performance here. Select to see the plots.
- Net Benefit: View the net benefit performance for any subset
- [Not always shown] Validation: If exploring a model development, you will see the validation tab that lets you see any external validation performances for the model.
Settings Dashboard
Probability threshold plot:
Cutoff Slider:
Dashboard
Cutoff Performance
Summary
Click view to see the corresponding plots:
Summary
Click view to see corresponding plots:
Select net benefit type to view:
Net Benefit Plot
Summary
Select one or more rows to generate comparison ROC and calibration plots