Artificial intelligence and machine learning for precision warfarin dosing: a comprehensive narrative review.

Warfarin Sensitivity Genotyping: A Review of the Literature and Summary of Patient Experience

Individualised strategy of warfarin treatment for deep vein thromboembolism – case reports

Reversal of the Anticoagulant Effects of Warfarin by Vitamin K1

Influence of CYP2C9 and VKORC1 on Warfarin Dose and Complications in Kazakhstani Patients with Left Ventricular Assist Device

BackgroundWarfarin is a commonly prescribed anticoagulant, and its dosing requires careful monitoring. This study used machine learning-based models to predict optimal warfarin dosage and identify factors influencing the International Normalized Ratio (INR) status in patients with cardiovascular diseases (CVDs).MethodsThis study involved 490 patients with CVDs on warfarin therapy at a hospital. Data were collected from patient records and direct interviews using a questionnaire. The various machine learning (ML) algorithms, including Support Vector Machine (SVM), Multinomial Logistic Regression (MLR), Random Forest (RF), Decision Tree (DT), and Ensemble Models, predict the optimal warfarin dosage for patients. Data were balanced using the Synthetic Minority Over-sampling Technique (SMOTE) algorithm to address class imbalances, and model performance was evaluated using metrics such as Precision, F1 Score, Accuracy, and Area Under the ROC Curve (AUC), with all analyses performed in Python using Jupyter Notebook.ResultsAfter balancing the data using SMOTE, ML algorithms were evaluated, with RF and SVM achieving similar accuracy of 75.7% and AUC values of 94% for RF and 93% for SVM. The ensemble (RF, MLR) and ensemble (SVM, MLR, RF) models outperformed others, achieving an accuracy of 76.4%, sensitivity of 75.2%, specificity of 92.1%, and AUC values of 95% and 94%, respectively. The DT model showed the lowest performance with an accuracy of 67.8%, an AUC of 79%, and a precision of 69%. Feature importance analysis indicated that INR, BMI, and warfarin indications were the most influential factors in predicting warfarin dosage, while gender, amiodarone, and nationality had minimal impact.ConclusionEnsemble ML algorithms, particularly those combining RF and SVM, show strong potential for accurately predicting warfarin dosages in cardiovascular patients. Key predictors such as INR, BMI, and warfarin indication improved model accuracy, while less impactful factors included gender and amiodarone use, supporting RF-SVM ensembles as effective tools for personalized dosing.

On Warfarin Use in Kidney Disease: A Therapeutic Window of Opportunity?

Warfarin is the most frequently prescribed oral anticoagulant worldwide. Due to its narrow therapeutic index and inter-patient variability in dose requirement, this drug has been considered an ideal target for personalised medicine. Several warfarin dosing algorithms have been proposed to tailor the warfarin dosage in the European, Asian and African-American populations. However, minimal interest was directed towards Middle East countries. The factors affecting warfarin dose requirement could be different in patients from different geographical and ethnic groups, limiting the value of published dosing algorithms. The first objective of this study was to examine the contribution of genetic and nongenetic factors on the variability of warfarin dose requirements in the Egyptian population using an easy, cost-effective and rapid analysis of vitamin K epoxide reductase complex subunit 1 (VKORC1) and cytochrome P450 (CYP) 2C9 single nucleotide polymorphism (SNP) genotyping of patients. A second objective was to develop and validate an algorithm for warfarin dose prediction that is tailored to Egyptian patients. Eighty-four patients, 41 males and 43 females, with a median (25th-75th percentiles) age of 39 (31-48) years were recruited in this study. Fifty patients whose international normalised ratio (INR) was in the range of 2-3 were allocated to a study cohort. SYBR Green-based multiplex allele-specific real-time PCR was used for genotyping of CYP2C9 (1075A>C) and VKORC1 (1173C>T) polymorphisms. Linear regression analysis, including the variables age, gender, CYP2C9 and VKORC1 SNP genotypes, was run to derive the best model for estimating the warfarin dose that achieves an INR of 2-3. The new warfarin dosing algorithm was examined in a second cohort of patients (n=34) to check its validity. The predicted dose requirements for a subgroup of our patients were calculated according to Gage and International Warfarin Pharmacogenetics Consortium (IWPC) algorithms available at http://www.warfarindosing.org. In the study cohort, warfarin dose/week in VKORC1 TT subjects was statistically significantly lower than in VKORC1 CC/CT subjects (p=0.032), while there was no statistically significant difference in warfarin dose/week between CYP2C9*1*1 and *1*3 (p=0.925). A multivariate stepwise linear regression analysis revealed that age and VKORC1 had independent and significant contributions to the overall variability in warfarin dose with a p-value=0.013 and 0.042, respectively. Maintenance dose (mg/week)=65.226-0.422×(age) - 9.474×(VKORC1). The estimated regression equation was able to account for 20.5% of the overall variability in warfarin maintenance dose. A significant positive correlation, with sufficient strength, was observed between the predicted warfarin dose and the actual prescribed dose (r=0.453, p=0.001). In the validation cohort, after application of the dosing algorithm, correlation between predicted and actual dose was statistically significant (p=0.023). The equation was particularly successful among patients with a dose≥35 mg/week. The correlation coefficient between the actual and predicted doses for IWPC and Gage were 0.304 and 0.276, respectively. When compared with our algorithm (r=0.279), the difference was non-significant: p=0.903 and 0.990, respectively. VKORC1 (1173C>T) contributes to the warfarin dose variability. Patients' age and genetic variants of VKORC1 account for nearly 20.5% of the variability in warfarin dose required to achieve an INR of 2-3. The success of a prediction equation based on these variables was proved in a different cohort: the predicted dose correlated significantly with the maintenance dose and the equation was more successful among patients with a dose≥35 mg/week. The results of the warfarin algorithm we developed were comparable with those of the IWPC and Gage algorithms with the advantage of using one SNP (VKORC1 1173C>T) only. This represents an economic advantage in our community. Replication of this study in a larger cohort of patients is necessary before translation of this knowledge into clinical guidelines for warfarin prescription.

VKORC1 and CYP2C9 genetic polymorphisms may not accurately predict warfarin dose requirements. We evaluated an existing warfarin dosing algorithm developed for Malaysian patients that was based only on VKORC1 and CYP2C9 genes. Five Malay patients receiving warfarin maintenance therapy were investigated for their CYP2C9*2, CYP2C9*3, and VKORC1-1639G>A genotypes and their vitamin K-dependent (VKD) clotting factor activities. The records of their daily warfarin doses and international normalized ratio (INR) 2 years prior to and after the measurement of VKD clotting factors activities were acquired. The mean warfarin doses were compared with predicted warfarin doses calculated from a genotypic-based dosing model developed for Asians. A patient with the VKORC1-1639 GA genotype, who was supposed to have higher dose requirements, had a lower mean warfarin dose similar to those having the VKORC1-1639 AA genotype. This discrepancy may be due to the coadministration of celecoxib, which has the potential to decrease warfarins metabolism. Not all patients' predicted mean warfarin doses based on a previously developed dosing algorithm for Asians were similar to the actual mean warfarin dose, with the worst predicted dose being 54.34% higher than the required warfarin dose. Multiple clinical factors can significantly change the actual required dose from the predicted dose from time to time. The additions of other dynamic variables, especially INR, VKD clotting factors, and concomitant drug use, into the dosing model are important in order to improve its accuracy.

Warfarin Dose Reduction vs Watchful Waiting for Mild Elevations in the International Normalized Ratio

Lack of Clinically Significant Interactions Between the Subcutaneously Administered Direct Thrombin Inhibitor Desirudin and Orally Administered Warfarin Upon the International Normalized Ratio.

Until recently, most patients taking warfarin (brand name Coumadin) had to visit a laboratory and/or clinic every few weeks for an international normalized ratio (INR) blood test and adjustment of their warfarin dose. It is now possible for a patient to measure his/her INR (self-testing) with a finger-stick drop of blood with use of a small, portable, battery-powered device. Some self-testing patients adjust their dose of warfarin (self-dosing) based on a set of instructions. Even more recently, online systems have been developed to facilitate and improve self-testing and self-dosing. Patients who use self-testing have described it as life changing. A video on the ease and benefits of self-testing by a physician-patient named Dr Michael Schwartz can be viewed on ClotCare at www.clotcare.org/inrselftestingvideo.aspx. Yes. The additional benefits are why Medicare and other insurance companies started paying for self-testing for most patients in March 2008. To understand the other potential benefits, however, one needs some background information. Warfarin is used to prevent blood clots that cause strokes, heart attacks, or other life-threatening conditions. If the dose of warfarin is too small, the INR will be low, and a patient may get a blood clot. If the dose is too large, the INR will be high, and a patient …

CORR Synthesis: When Should the Orthopaedic Surgeon Use Artificial Intelligence, Machine Learning, and Deep Learning?

The effectiveness and safety of warfarin require maintaining an international normalized ratio (INR) within the therapeutic range. To identify predictors of nontherapeutic INR results in patients receiving warfarin. A retrospective study was conducted using 350 ambulatory care patients from a broad geographic region, all receiving long-term warfarin therapy and followed in a tertiary-care cardiology clinic. Possible predictors of nontherapeutic INR results (gender, age, body weight, body mass index, height, race, tobacco use, alcohol use, warfarin dose, therapeutic indication, regimen intensity, INR monitoring frequency/category, interacting medications, adverse events) were assessed with logistic regression models. Subset analysis involved 146 patients concurrently monitored with capillary whole blood INR (CoaguChek). As measured on venous specimens, 52% (182/350) of the patients had subtherapeutic INR results and 13% (44/350) had supratherapeutic INR results despite frequent (< or =4 wk) monitoring in 75% of the patients. Due to the small sample size, supratherapeutic INR results could not be further analyzed. Of 19 predictors tested, only daily warfarin dose (p < 0.02) and regimen intensity (p < 0.03) were significant independent and additive predictors of subtherapeutic results. Patients on the high-intensity regimen (INR 2.5-3.5) and receiving warfarin < or =6 mg/day had >50% risk of having subtherapeutic INR results. Subtherapeutic CoaguChek results were independent predictors of subtherapeutic venipuncture INR results in the subset (p = 0.001). In the absence of readily identifiable predictors, only higher warfarin dosing and/or more frequent monitoring (possibly with point-of-care/home monitoring devices) may minimize the time that INRs are subtherapeutic, especially in patients receiving low-dose and/or high-intensity anticoagulation therapy.

Warfarin dosing in cardiac surgery patients is complicated by a heightened sensitivity to the drug, predisposing patients to adverse events. Predictive algorithms are therefore needed to guide warfarin dosing in cardiac surgery patients. This study aimed to develop and validate an algorithm for predicting the warfarin dose needed to attain a therapeutic international normalized ratio (INR) at the time of discharge in cardiac surgery patients. We abstracted variables influencing warfarin dosage from the records of 1031 encounters initiating warfarin between April 1, 2011, and November 29, 2019, at St Michael's Hospital in Toronto, Ontario, Canada. We compared the performance of penalized linear regression, k-nearest neighbors, random forest regression, gradient boosting, multivariate adaptive regression splines, and an ensemble model combining the predictions of the 5 regression models. We developed and validated separate models for predicting the warfarin dose required for achieving a discharge INR of 2.0-3.0 in patients undergoing all forms of cardiac surgery except mechanical mitral valve replacement and a discharge INR of 2.5-3.5 in patients receiving a mechanical mitral valve replacement. For the former, we selected 80% of encounters (n=780) who had initiated warfarin during their hospital admission and had achieved a target INR of 2.0-3.0 at the time of discharge as the training cohort. Following 10-fold cross-validation, model accuracy was evaluated in a test cohort comprised solely of cardiac surgery patients. For patients requiring a target INR of 2.5-3.5 (n=165), we used leave-p-out cross-validation (p=3 observations) to estimate model performance. For each approach, we determined the mean absolute error (MAE) and the proportion of predictions within 20% of the true warfarin dose. We retrospectively evaluated the best-performing algorithm in clinical practice by comparing the proportion of cardiovascular surgery patients discharged with a therapeutic INR before (April 2011 and July 2019) and following (September 2021 and May 2, 2022) its implementation in routine care. Random forest regression was the best-performing model for patients with a target INR of 2.0-3.0, an MAE of 1.13 mg, and 39.5% of predictions of falling within 20% of the actual therapeutic discharge dose. For patients with a target INR of 2.5-3.5, the ensemble model performed best, with an MAE of 1.11 mg and 43.6% of predictions being within 20% of the actual therapeutic discharge dose. The proportion of cardiovascular surgery patients discharged with a therapeutic INR before and following implementation of these algorithms in clinical practice was 47.5% (305/641) and 61.1% (11/18), respectively. Machine learning algorithms based on routinely available clinical data can help guide initial warfarin dosing in cardiac surgery patients and optimize the postsurgical anticoagulation of these patients.

A Randomized Controlled Trial of Empiric Warfarin Dose Reduction with the Initiation of Doxycycline Therapy