Model-based meta-analysis of anticoagulant therapy based on individual patient data

Model-based meta-analysis of anticoagulant therapy based on individual patient data

Anticoagulant is a drug class used for prevention of thromboembolism (TE) in patients with atrial fibrillation (AF) or venous thromboembolism (VTE). In treatment using anticoagulant drugs, it is important to control the risk-benefit balance because excessive anticoagulation effect could cause adverse effects such as intracranial bleeding. Therefore, for the dose adjustment of warfarin, which is a conventional oral anticoagulant used for over 60 years, a routine blood test has been required to check if the drug effect is within the appropriate range. In such context, direct oral anticoagulants (DOACs), including dabigatran, rivaroxaban, apixaban, and edoxaban, recently emerged as novel oral anticoagulants. DOACs have an advantage over warfarin because they can be used at a fixed dose without the routine blood test. Indeed, large global trials have demonstrated that they are as efficacious and safe as dose-adjusted warfarin even without routine monitoring. However, recent studies have shown the "Off-Label Dosing" of DOACs. The fixed dose of DOAC is chosen from standard or low dose according to the dose-selection criteria written in their labels. Nevertheless, in actual clinical practice, a part of patients deviated these criteria. This is extremely important issue because the incidences of both TE and bleeding were higher in such patients compared with patients taking the recommended dose. However, solving the off-label dosing is a very challenging work. The off-labeled patients have more risk factors, in terms of age, gender, medical history, and so on. Therefore, we cannot judge easily which of the presence of such risk factors and the off-label dosing is responsible for higher event incidence. To further enhance the clinical benefit of anticoagulant therapy, it is essential to clarify the relationship between risk factors and event risks during anticoagulant treatment in more detail, and such knowledge should be used to personalize the current anticoagulant therapy.   In this study, we will try to utilize artificial intelligence (AI) to analyze big data derived from multiple clinical trials. This novel analytic approach introducing AI allows us to take full advantage of accumulated clinical trial data. As a result, we may overcome complex problems that could not be dealt with by conventional approaches adopting traditional statistical methods. Particularly, we will focus on interactions between/within two factors - potential risk factors (e.g. well-known risk factors, as well as factors that is unestablished but could be risk factors for TE and/or bleeding complications) and treatment factors (e.g. the type of anticoagulant drug used and its dosage and dosing schedule). We will use R for AI analysis primarily. SAS and WinBUGS may also be used for comparative analyses. The algorithms to be tested in AI analysis include k-neighbor classifier, decision tree, random forest, gradient boosting classifier, support vector machine, multi-layer perceptron (i.e. neural network). The data may also be evaluated using analysis of variance (ANOVA) and conventional model-base analyses for comparison. To extract the important features learned by the AI analysis, sensitivity analyses will be performed systematically by using virtual data. Through such analyses, we aim not only to find out novel and definitive risk factors against TE and bleeding, but also to clarify how the presence of such risk factors affects the therapeutic effects of anticoagulation therapy. Once such complex interactions are successfully clarified by AI, we may apply it to various situations; for example, we may use the AI to propose optimal treatments for special populations that are difficult to enroll into actual clinical trials, such as patients who have multiple risk factors. These outcomes could help provide more efficacious and safer anticoagulant treatment for individual patients and contribute to enhancing the precision medicine of antithrombotic therapy.

[{ "PostingID": 1741, "Title": "GSK-AR2103413", "Description": "An International Randomized Study Evaluating the Efficacy and Safety of Fondaparinux Versus Control Therapy in a Broad Range of Patients With ST Segment Elevation Acute Myocardial Infarction." },{ "PostingID": 1743, "Title": "GSK-AR1103420", "Description": "An international, randomized, double-blind study evaluating the efficacy and safety of fondaparinux versus enoxaparin in the acute treatment of unstable angina/non ST-segment elevation MI acute coronary syndromes" },{ "PostingID": 2596, "Title": "BI-1160.47", "Description": "Secondary Prevention of Venous Thrombo Embolism (VTE)." },{ "PostingID": 2597, "Title": "BI-1160.46", "Description": "Phase III Study Testing Efficacy & Safety of Oral Dabigatran Etexilate vs Warfarin for 6 m Treatment for Acute Symp Venous Thromboembolism (VTE)" },{ "PostingID": 2598, "Title": "BI-1160.53", "Description": "Efficacy and Safety of Dabigatran Compared to Warfarin for 6 Month Treatment of Acute Symptomatic Venous Thromboembolism" },{ "PostingID": 2599, "Title": "BI-1160.26", "Description": "Randomized Evaluation of Long Term Anticoagulant Therapy (RE-LY) With Dabigatran Etexilate" },{ "PostingID": 3033, "Title": "BI-1160.71", "Description": "RELY-ABLE Long Term Multi-center Extension of Dabigatran Treatment in Patients With Atrial Fibrillation Who Completed RE-LY Trial" }]