Robust K-PD model for activated clotting time prediction and UFH dose individualisation during cardiopulmonary bypass. (February 2022)

• Record Type: Journal Article
• Authors: Chaux, Robin; Lanoiselée, Julien; Magand, Clément; Zufferey, Paul; Delavenne, Xavier; Ollier, Edouard
• Is Part Of: Computer methods and programs in biomedicine. Volume 214(2022)
• Publication Date: 2022-02
• Subjects: Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2021.106553
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: Activated clotting time (ACT) is a point-of care test used to monitor UFH effect. ACT can return outlier values leading to potentially inappropriate dosing adaptation. A K-PD model with a residual error following a student's t distribution can provide ACT predictions which are robust to outlier ACT measurements. Robust real time UFH dosing individualization is obtained with Bayesian estimation. Abstract: Background and objective: Activated clotting time (ACT) is a point-of-care test used to monitor the effect of unfractionated heparin (UFH) during cardiopulmonary bypass (CPB). This test sometimes returns aberrant values, which can lead to the administration of an inappropriate dosing regimen. The development of a population-robust K-PD model of UFH could allow the individualisation and automation of UFH therapy during CPB. Methods: We conducted a prospective observational study to collect ACT measurements from patients undergoing surgery using CPB. The ACT data were split into a development and validation cohort. The development cohort was used to estimate a standard and robust population K-PD model characterised by a residual error following a normal distribution and student's t -distribution. The ACT prediction performance using Bayesian estimates of individual K-PD parameters was evaluated by comparing predicted versus observed ACTs. Using estimates of the robust K-PD model, a Bayesian individualisation strategy to automate UFH administration was proposed and evaluated using Monte Carlo simulations. Results: A total of 295 patients were included in the study, and 1561 ACTs were collected. In patients without outlier values, Bayesian estimates (based on four ACT measurements) from both standard and robust K-PD models had similar performances, with a median prediction bias close to 0 s. In patients with outlier measurements, the use of the robust K-PD model greatly improved the prediction bias and root-mean-square error (RMSE), with a mean prediction bias of 3.25 s, IQR = [-19.9; 46.03] versus -86 s IQR = [-135.7; -63.8] for the standard model. Monte Carlo simulations showed that the robust Bayesian individualisation strategy allowed the ACT to be maintained above the target using only two to three ACT measurements. Conclusions: The use of a robust K-PD model reduced prediction bias and RMSE in patients with outlier ACT measurements. The Bayesian individualisation strategy using robust estimates of individual parameters may help automate UFH dosing regimens. Proper clinical validation is warranted before its use in daily clinical practice.