Related Experiment Video
Updated: Sep 25, 2026
Leveraging Turbidity and Thromboelastography for Complementary Clot Characterization
Published on: June 4, 2020
A clinical prediction model to support risk-adapted fibrinogen monitoring during tigecycline therapy: Multicentre
Jinlin Guo1, Xinfeng Cai2, Yayu Huo3
1School of Pharmacy, Shanxi Medical University, Taiyuan, Shanxi, China; Department of Pharmacy, Shanxi Provincial People's Hospital, Taiyuan, Shanxi, China.
Objectives:
Uniform fibrinogen monitoring during tigecycline therapy may be inefficient, whereas conventional absolute thresholds can miss deterioration in patients with a marked relative decline from a high baseline. We developed and validated an initiation-time clinical prediction model to support risk-adapted fibrinogen monitoring during tigecycline therapy.
Methods:
In this multicentre prognostic study across tertiary hospitals in China, retrospective model development and external validation were followed by prospective temporal validation. Eligible adults received tigecycline for at least 72 h, had baseline fibrinogen of at least 2.0 g/L, and underwent at least one on-treatment reassessment. The primary outcome was study-defined fibrinogen deterioration, defined as a decline of at least 50% from baseline and/or a nadir below 1.5 g/L. A logistic regression model using variables available at treatment initiation was evaluated for discrimination, calibration and decision-analytic utility.
Results:
Among 1411 eligible patients, 332, 760 and 319 were included in the development, external validation and prospective validation cohorts, respectively; strict case/control analysis sets comprised 222, 514 and 234 patients. The final 4-variable model included age, tigecycline daily dose, total bilirubin and baseline fibrinogen and was intended for monitoring-intensity stratification, not treatment selection. Optimism-corrected development, external validation and prospective validation AUCs were 0.755, 0.731 and 0.714, respectively. In the full prospective eligible cohort (N = 319), discrimination attenuated (AUC 0.693) and calibration drift supported local recalibration before implementation. After recalibration, decision-curve analysis favoured model-guided monitoring at prespecified threshold probabilities.
Conclusions:
This simple initiation-time model may support risk-adapted monitoring, but local recalibration should precede threshold-based use in new settings.