Mechanical ventilation is both an essential tool of the respiratory critical care environment and a major risk in the development of Acute Respiratory Distress Syndrome (ARDS). The human lung- ventilator system (LVS) includes the interaction of complex anatomy with a mechanical apparatus. The form of this system imposes difficulties on modeling with sufficient flexibility and physiological fidelity to support patient individualization in clinical informatics applications. This work proposes a

more »

... pothesis-driven strategy to LVS modeling, in which robust personalization is achieved by exchanging explicit physiological process resolution for a pre-defined parameter basis. Model inversion, here via windowed data assimilation, forges patient data into interpretable parameter values intended for data characterization rather than physiological fidelity. Application to human LVS data, including sequences of normal and dyssynchronous breaths, illustrates robustness of the inferential model. Parameter summaries distinguish types of breath sequences, with implications for downstream machine learning applications such as tracking lung injury and ARDS development, breath classification, and phenotyping. Fidelity of reconstructed waveforms from characterization is tied to parameter definitions and resolution, factors that must be considered in conjunction with application objectives.