PROJECT SUMMARY
Ensuring accurate predictions of the transport and deposition of orally inhaled drug products (OIDPs)
within dry powder inhaler (DPI) flow channels to the human respiratory system is crucial. This is particularly
important given the diverse variabilities such as inhaler design, drug formulation, and user-specific factors (e.g.,
disease-specific lung conditions and patient-device coordination). Such accuracy is essential for demonstrating
comparability and bioequivalence (BE) between generic and reference listed drug (RLD) inhalers, quantifying
their performances' uncertainties. Although CFPD models and discrete element method (DEM) can provide high-
fidelity spatiotemporal distributions of variables of interest (i.e., emitted APSDs, emitted doses, regional lung
deposition fractions) using 3D physiologically realistic human airway geometries, there is a drawback due to the
high computational cost for using such models to investigate how numerous user-specific factors can influence
the comparability and BE. To address such a drawback, the overall goal of this project is to develop advanced
CFD-DEM trained machine learning (ML) models, i.e., AI-empowered reduced order models (ROMs), to
improve the understanding of variability in generic OIDPs, enhance comparability, and demonstrate BE more
efficiently. Our central approach is based on the state-of-the-art CFD-DEM, elastic airway model, and self-
supervised learning (SSL), ensuring a significant contribution to accurately assessing generic inhaler
comparability and performance with the reliable fast-running ROMs as new open-source in silico tool. Limited by
the 2-year time period, the research will pursue three specific aims focusing on inhaled Tiotropium Bromide
laden lactose particles emitted from SpirivaTM HandihalerTM based DPIs in human respiratory systems with
different COPD severities: (1) Aim 1: Quantify variability of inhaler geometric design, drug formulation, and user-
inhaler coordination on emitted APSDs and doses from DPIs using CFD-DEM and develop an AI-empowered
ROM to predict emitted APSDs and doses, (2) Aim 2: Quantify variability of COPD severity on transport and
deposition of inhaled OIDPs in the human respiratory system, and develop another ROM to predict regional lung
deposition, and (3) Aim 3: Integrate the two ROMs into an “All-in-One” ROM for uncertainty quantification. In
both years, we will prepare and submit co-authored manuscripts with FDA OGD, transfer data and codes with
other deliverables, and provide technical training and tutorials to them on using and further refining the ROMs.
The proposed research is significant because the proposed ROMs will address the computational challenges
of current CFD-based models in uncertainty quantification of in silico inhaler comparability and BE studies. The
proposed research is innovative since it will integrate the state-of-the-art CFD-DEM, elastic airway, and SSL for
ROM training and testing. It will create a paradigm shift in regulatory evaluations and advance regulatory
approaches to support innovation, ultimately facilitating the development and approval of generic DPIs.
