Models and Data for "Data-Centric System Dynamics Modelling of General Internal Medicine Physicians in Ontario: Forecasting Supply and Workload Analysis"

This dataset contains the system dynamics models, calibration setup, and supporting files associated with the MASc thesis “Data-Centric System Dynamics Modelling of General Internal Medicine Physicians in Ontario: Forecasting Supply and Workload Analysis” by Parnian Azimzadeh. The models are implemented in Vensim Professional 10.1.3 and include four main components: 1) Calibration Model – used for optimization and identification of optimal rates for five base fractions: Base Return Fraction, Base Leave Fraction, Base OutMigration Fraction, Base CredLoss Fraction, and Base Retire Fraction. The final optimized rates, along with the final payoff report, are provided in the file Final_Flow_Coef_0.1.par.txt. 2) Validation Model – applies the optimized rates from the calibration period (2009–2019) to test the model against out-of-sample data for 2020–2023. 3) Baseline Forecast Model – uses the optimized calibration results to project the supply of general internal medicine physicians in Ontario through 2040 under baseline conditions. 4) Scenario Analysis Model – introduces adjustable policy levers to test alternative futures by modifying selected inflows and outflows. For this purpose, a separate scenario dataset with disaggregated flows is used to enable more detailed scenario analysis. Additional files included: A) Optimization trace (.tab) – contains the simulation results for each optimization run, showing how error and parameter values evolve across iterations. B) Payoff and optimizer configuration files (.vpd, .voc) – define the error measure and optimization settings used in calibration. All three core models (Calibration, Validation, and Baseline Forecast) rely on a “pushed” data series, which shifts the original longitudinal OPRC PIO dataset forward by one year. This adjustment ensures alignment with the Vensim equations that calculate physician counts annually. The recategorization and one-year push, as described in the thesis, produce only negligible changes in results, and validation tests are conducted against the actual data. Note on data loading: When loading the CSV data through Tools → Control Panel in Vensim, the setting “time runs down” must be selected, with time values recognized when col# = 1. This ensures proper alignment of time series with the model equations. Together, these files enable full reproduction of the calibration (2009–2019), validation (2020–2023), and projection analyses, as well as exploration of scenario-based futures for physician workforce supply.