Partial least squares regression: a valuable method for modeling molecular behavior in hemodialysis

Area of Science:

• Biomedical Engineering
• Clinical Chemistry
• Data Science

Background:

• Hemodialysis requires precise monitoring of molecular kinetics.
• Predicting post-dialysis urea concentration is crucial for patient management.
• Complex biological processes necessitate robust statistical modeling.

Purpose of the Study:

• To apply Partial Least Squares Regression (PLS) for modeling complex molecular kinetic data.
• To develop predictive statistical models for urea concentration post-hemodialysis.
• To assess the utility of PLS in clinical settings and patient monitoring.

Main Methods:

• Utilized Partial Least Squares Regression (PLS) to analyze molecular kinetic data.
• Developed statistical linear models to predict equilibrated urea concentration.
• Validated model accuracy using a cross-center study.

Main Results:

• Achieved models with an average relative prediction error (RPE) below 0.05%.
• Cross-center validation demonstrated a model RPE of less than 3%.
• The PLS model proved robust to variations like sampling time and identified outlier patient patterns.

Conclusions:

• PLS is effective for modeling complex kinetic processes in hemodialysis.
• The developed models offer accurate prediction for bedside monitoring.
• PLS facilitates the identification of key co-variables and patient-specific molecular dynamics.