What question did this study set out to answer?

To evaluate machine learning-derived models for predicting optimal doses of oral anticoagulants.

January 10, 2026Journal of Pharmacy Practice and Research0 citationsOpen Access

Machine learning models for predicting dosing of oral anticoagulants: a systematic review

MBMichael Barras LHLeila Shafiee Hanjani ISIan Scott

Key Result

Only four out of 53 studies on machine learning models for predicting optimal oral anticoagulant doses met essential criteria for model building.

Key Points

To evaluate machine learning-derived models for predicting optimal doses of oral anticoagulants.
Conducted a systematic review of relevant studies using PubMed, Embase, and other databases.
Included studies that used machine learning methods for OAC dosing predictions.
Applied CHARMS checklist for data extraction and appraisal.
53 studies met inclusion criteria, primarily focusing on vitamin K antagonists.
77% of studies targeted a 'stable therapeutic dose' for oral anticoagulants.
Only four studies met all criteria essential for building ML models.

Structured PICO

Do machine learning models accurately predict the optimal dose of oral anticoagulants in adults?

Population

53 studies evaluating machine learning-derived prediction models for determining the optimal dose of oral anticoagulants in adults.

Intervention

Machine learning-derived prediction models for oral anticoagulant dosing

Outcome

Prediction of optimal oral anticoagulant dose (predominantly stable therapeutic dose)

Current machine learning models for oral anticoagulant dosing predominantly focus on warfarin and lack rigorous external validation and prospective clinical assessment.

Main Result

Absolute Event Rate: 0% vs 0%

Abstract

Abstract Aim To identify and appraise studies of machine learning (ML)‐derived prediction models for determining the optimal dose of oral anticoagulants (OACs). Data Sources PubMed, Embase, International Pharmaceutical Abstracts (IPA), IEEE Xplore, and Web of Science were searched from inception to 31 May 2024 using key terms synonymous with 'artificial intelligence' or 'machine learning', 'prediction', 'dose', and 'oral anticoagulants'. OACs included vitamin K antagonists (VKAs) — warfarin, acenocoumarol, phenprocoumon — and direct oral anticoagulants (DOACs) — apixaban, rivaroxaban, dabigatran. Study Selection Studies published in English that used ML methods to develop and/or validate models to predict optimal OAC doses in adults, in any healthcare setting were included. A modified checklist, using the Checklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS), guided data extraction and study appraisal was performed independently by two researchers. This review was conducted and reported according to the Preferred Reporting Items for Systematic reviews and Meta‐Analyses (PRISMA) 2020 statement and was registered with the International Prospective Register of Systematic Reviews (PROSPERO) (Study = registration: PROSPERO CRD42021281150). Results Of 7791 abstracts, 164 underwent full‐text review and 53 studies met the inclusion criteria. All used supervised learning methods; all but one evaluated VKAs. Of the 53 studies, 77% ( n = 41) targeted a ‘stable therapeutic dose’; and only one DOAC study evaluated patient outcomes using apixaban, rivaroxaban, edoxaban, and dabigatran. Nine studies (17%) were prospective and two (4%) reported external validation. Of 44 retrospective analyses, 15 (34%) studies used the International Warfarin Pharmacogenetics Consortium (IWPC) dataset, with 12 (23%) reporting external validation. Only four (8%) studies satisfied all pre‐determined criteria considered essential for building ML models for OACs. While comparison across studies was challenging, two of the better performing, externally validated models were by Li et al. (2020) and Gu et al. (2022) with an accuracy of 63% and 75% and retrospectively. Conclusion Future research should train models on large datasets, including key genetic and non‐genetic predictors, undertake external validation, and prospectively assess performance using clinical data.

Ask AI

Helpful

Bookmark

View Full Paper