Abstract
Background: Total hip arthroplasty (THA) is a well-tolerated and effective procedure that can improve a patient's mobility and quality of life. A main concern, however, is the release of metal ions into the body due to wear and corrosion. Commonly reported ions are Co and Cr, while others, such as Ti, Mo, and Ni, are less frequently studied. The objective of this study was to characterize compartmentalization and time-dependent ion behaviors across serum, whole blood, and urine after hip prosthetic implantation. The goal of using Random Forest (RF) was to determine whether machine learning modeling could support temporal trends across data. Methods: Data was gathered from the literature of clinical studies, and we conducted a pooled analysis of the temporal kinetics from cohorts of patients who received hip prosthetics. Mean ion concentrations were normalized to µg/L across each fluid and weighted by cohort sample size. RF was used as a study-level test of predictive accuracy across ions. Results: For serum and whole blood, Co and Cr displayed one-phase association models, while Ti showed an exponential rise and decay. Ions typically rose quickly within the first 24 months postoperatively. Serum Co and whole blood had similar patterns, tapering off just under 2 µg/L, but serum Cr (~2.02 µg/L) was generally higher than that of whole blood (~0.99 µg/L). Mean urinary Co levels were greater than those of Cr, suggesting a larger, freely filterable fraction for Co. RF was implemented to determine predictive accuracy for each ion, showing a stronger fit for Co (R(2) = 0.86, RMSE = 0.57) compared to Cr (R(2) = 0.52, RMSE = 0.50). Conclusions: Sub-threshold exposure was prevalent across cohorts. Serum and whole blood Co and Cr displayed distinct kinetic profiles and, if validated, could support fluid-specific monitoring strategies. We present a methodology for interpreting ion kinetics and show potential for machine learning applications in postoperative monitoring.