(M-111) Cystatin C complements but does not exceed serum creatinine as a renal function marker: evidence from vancomycin pharmacokinetic model performance in a clinical context
Monday, November 11, 2024
7:00 AM – 5:00 PM MST
Jasmine Hughes, PhD – Director of Data Science, InsightRX; Ron Keizer, PharmD PhD – Chief Scientific Officer, InsightRX
Dominic MH Tong, PhD: No relevant disclosure to display
Objectives: Cystatin C (CYSC) and serum creatinine (CR) are two biomarkers used in estimating glomerular filtration rate (GFR), a marker of kidney function. CYSC has been shown to be more accurate for older patients and patients with lower muscle mass [1-2], whereas CR is more widely available at a lower cost. For population pharmacokinetic (popPK) models, an accurate estimate of kidney function is critical in determining drug clearance. In vancomycin, popPK models using CR [3-7] or CYSC [1-2] have been developed and used to tailor dosing for individual patients clinically. Here, we compare CR-based and CYSC-based popPK model performance for adult patients treated with vancomycin, and evaluate whether a refit model using one or both biomarkers improves vancomycin prediction.
Methods: De-identified retrospective EHR data describing 513 adult patients treated with vancomycin at 16 health systems were used in this analysis. The best performing popPK models describing adult vancomycin PK using CR were identified from previous work, while models using CYSC were identified from literature and evaluated for fit for use in this population [8]. Metrics used to evaluate iterative prediction performance were accuracy (defined as predictions within 20% or 2.5 mg/L of the measured value), mean percent error (MPE), and normalized root mean square error (nRMSE). We then refitted the Hughes model [3] using different GFR equations on 70% of the data and evaluated its performance on the other 30%. GFR was evaluated using the Hoek equation [8] for cystatin C GFR, the CKD-EPI equations for GFR with and without cystatin C [9], and combining Cockcroft-Gault eGFR with a cystatin C term directly for vancomycin clearance.
Results: CR-based models generally outperformed CYSC-based models in predicting vancomycin concentrations with the best models for each biomarker being UVM and Liu (accuracy: 44.3%, 43.0% Liu; MPE: -4.9%, 3.3%; nRMSE: 41.1%, 42.1% for UVM and Liu, respectively). The refitted models that use both CYSC and CR, whether via CKD-EPI (Accuracy: 53.2%) or CYSC directly (53.2%) performed better than CR alone (46.1%) or CYSC alone (48.9%) .
Conclusions: Clinically proven CR-based popPK models were superior to published CYSC-based models in predicting vancomycin levels. Refitting models with CYSC and/or CR showed CYSC is not a useful substitute for CR. A combined model including both CYSC and CR improved vancomycin prediction performance slightly, though this improvement may not be clinically relevant.
