Using a State-Space Representation and Derivative Control to Model Influenza Outcomes in NY, CA, OH, MD, MN, and FL


Summary

Objective

To develop a model using Derivative control that predicts vaccine utilization at a state level in order to provide insights for policymakers and public health officials.

Methods

Logistic Regression was used to identify key variables, from a large dataset, correlated to vaccine utilization which would be used to train the model. Model parameters were constructed using an Ordinary Least Squares Regression on annualized, state-level data. Root Mean Squared Error was used to assess model fit to the training and testing datasets. 

Results

Using derivative control, the model’s vaccine utilization output was within five percentage points of the observed values for training and testing data sets. The standard deviation in RMSE between training and testing datasets was less than 1%. Both outcomes are within a functional tolerance, indicating a strong fit and predictive capability. Out of the States used to test the fit, Ohio and Maryland show the greatest increase in vaccine utilization. 

Conclusion

This model is an apt tool for predicting vaccine utilization at the state level. Policy tests may be easily implemented and adjudicated using RMSE or other best fit techniques. There is ample opportunity to adjust model scope to predict how specific subpopulations will utilize the vaccine as a result of different policy implementations.