The parameter Houlihan: A solution to high-throughput identifiability indeterminacy for brutally ill-posed problems

Area of Science:

• Physiological modeling
• Machine learning
• Data assimilation

Background:

• Clinical forecasting often uses complex physiological models.
• Data assimilation offers advantages for sparse, non-stationary clinical data.
• Parameter identifiability issues can arise in nonlinear physiological models.

Purpose of the Study:

• To introduce a novel method, the parameter Houlihan, for selecting model parameters in data assimilation.
• To minimize forecasting error while addressing parameter identifiability challenges.
• To enhance the application of data assimilation in clinical settings.

Main Methods:

• Combining traditional machine learning with data assimilation techniques.
• Developing the parameter Houlihan algorithm for parameter selection.
• Evaluating glucose forecasts using data assimilation with Houlihan-selected parameters.

Main Results:

• The parameter Houlihan method effectively minimized forecasting errors compared to traditional methods.
• Data assimilation with Houlihan-selected parameters improved glucose forecast accuracy.
• Lowest forecast error did not always guarantee physiological accuracy, indicating areas for improvement.

Conclusions:

• The parameter Houlihan method offers a viable approach to combine machine learning and data assimilation.
• This methodology provides a lower-threshold entry point for using data assimilation with clinical data.
• Further advancements are needed to improve the physiological fidelity of forecasts.