A major challenge in developing blood-contacting medical devices is mitigating thrombogenicity of an intravascular device. Thrombi may interfere with device function or embolize from the device to occlude distant vascular beds with catastrophic consequences. Chemical interactions between plasma proteins and bioengineered surface occur at the nanometer scale; however, continuum models of blood predict local shear stresses that lead to platelet activation or aggregation and thrombosis. Here, an iterative approach to blood flow path design incorporating in silico, in vitro, and in vivo experiments predicted the occurrence and location of thrombi in an implantable hemofilter. Low wall shear stress (WSS) regions identified by computational fluid dynamics (CFD) predicted clot formation in vivo. Revised designs based on CFD demonstrated superior performance, illustrating the importance of a multipronged approach for a successful design process.

An implantable artificial kidney using a hemofilter constructed from an array of silicon membranes to provide ultrafiltration requires a suitable blood flow path to ensure stable operation in vivo. Two types of flow paths distributing blood to the array of membranes were evaluated: parallel and serpentine. Computational fluid dynamics (CFD) simulations were used to guide the development of the blood flow paths. Pressure data from animal tests were used to obtain pulsatile flow conditions imposed in the transient simulations. A key consideration for stable operation in vivo is limiting platelet stress accumulation to avoid platelet activation and thrombus formation. Platelet stress exposure was evaluated by CFD particle tracking methods through the devices to provide distributions of platelet stress accumulation. The distributions of stress accumulation over the duration of a platelet lifetime for each device revealed that stress accumulation for the serpentine flow path exceeded levels expected to cause platelet activation while the accumulated stress for the parallel flow path was below expected activation levels.