Abstract: FR-PO982

Engineering Design of Robust Ultrafiltration Profiles in Hemodialysis

Session Information

Category: Bioengineering and Informatics

  • 101 Bioengineering and Informatics

Authors

  • Abohtyra, Rammah M., UMass, Florence, Massachusetts, United States
  • Hollot, Christopher V., Department of Electrical and Computer Enginering, University of Massachusetts Amherst, Amherst, Massachusetts, United States
  • Horowitz, Joseph, University of Massachusetts, Amherst, Massachusetts, United States
  • Chait, Yossi, University of Massachusetts, Amherst, Massachusetts, United States
Background

Fluid removal during hemodialysis (HD) by ultrafiltration can lead to intradialytic hypotension, which is associated with an increase in morbidity and mortality. We design a robust, individualized ultrafiltration rate (UFR) profile to achieve a target volume removal under constraints on maximal UFR and maximal hematocrit (HCT) levels.

Methods

The fluid volume dynamics during HD was described by a validated nonlinear model comprising intravascular and interstitial pools, microvascular refilling/filtration, and lymphatic flow. Model parameters, red blood cells volume (Vrbc), plasma protein mass (mp), and filtration coefficient (Kf), were estimated using 30 minutes of UFR and HCT clinical data and validated using remaining data. Anticipated parameter changes during HD and estimation errors were accounted for by allowing parameter uncertainty ranges of +/- 5%. The profile was designed using an optimization algorithm to remove 2.8 L in 4 hrs, with UFR not exceeding 10 ml/hr/kg, and HCT not exceeding 110% of initial HCT, and to guarantee performance over the entire range of parameter uncertainty.

Results

Model parameters (Vrbc =2.24 L, Kf =0.006 L/min/mmHg, and mp=179.58 gr) of a 75 kg patient were estimated based on the initial 30 minutes of the HD session (Figure, top left) and validated over an additional 170 minutes of the session (Figure, bottom left). Simulation of the designed UFR profile (Figure, top right) over the range of model parameter uncertainty confirmed that all specifications were achieved (worst case HCT shown in Figure, bottom right); the worst case of HCT is simulated for a particular set of parameters within the parameter uncertainty range of the model.

Conclusion

The divergence of the validated response later in the HD period demonstrates that the underlying patient’s response can deviate from model prediction due to, for example, autonomous adaption not captured in the model. This motivated the use of robust design of UFR profiles. Our UFR profile is designed to satisfy HCT and UFR constraints, and to guarantee that performance criteria are met over the entire model parameter uncertainty range.

Funding

  • NIDDK Support