A Novel Predictive Tool for Poor Anticoagulation Control in Patients on Vitamin K Antagonists in Spain: An Exploratory Study.

Background

Given the limitations of available predictive tools for poor anticoagulation control, we aimed to build and validate a novel tool and compare its predictive ability with the SAMe-TT2R2 score, using the patients from the OBJETIVO 2024 study.

Methods

The original sample was randomly assigned into a training group (70%, n = 1982) for model development and a validation group (30%, n = 909) for model validation. Stratification of patients was performed based on the presence of diabetes and functional dependence on daily living activities (N with available data = 2891). The model was developed through binary logistic regression, with poor international normalized ratio (INR) control (time in therapeutic range <65% using the Rosendaal method) as a dependent variable. Independent variables included renal insufficiency (glomerular filtration rate < 60 mL/min), chronic obstructive pulmonary disease, diabetes, active smoker, alcohol abuse, previous ablation, hemoglobin level, HbA1c, functional dependence in daily living activities, and number of treatments received in the last 6 months.

Results

The receiver operating characteristic area under the curve (ROC AUC) was 0.579. The optimal cut-off point was 0.474 (sensitivity: 47.5%; specificity: 65.0%). Overall quality of the model for the training and validation groups was 0.49 and 0.55, respectively. The mean SAMe-TT2R2 in patients from the OBJETIVO 2024 study was 2.3. The ROC AUC for the SAMe-TT2R2 tool was 0.530. Overall quality of SAMe-TT2R2 for the present population was 0.51.

Conclusions

None of the models presently tested reached the minimum threshold considered acceptable for discriminative ability. To date, utility of different models to predict poor anticoagulation control seems far from optimal in clinical practice.