Therapeutic Drug Monitoring of Direct Oral Anticoagulants and Its Association with Clinical Outcomes: A Systematic Review and Meta-Analysis

Scheme 1. PRISMA flow diagram.

Scheme 2. Risk of Bias Summary (Newcastle–Ottawa Scale) Traffic-light plot generated using the ROBVIS tool illustrating domain-specific and overall risk-of-bias judgments for all included studies (n = 19). D1 = Selection; D2 = Comparability; D3 = Outcome. Green (+) = Low risk of bias; Yellow (−) = Unclear risk; Red (X) = High risk; Red (!) = Critical risk. Judgments reflect reviewer consensus based on the Newcastle–Ottawa Scale, where higher risk categories denote greater methodological limitations related to study design, confounding, or outcome ascertainment. (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Testa, 2018 [11]; Testa, 2019 [20]; Miklič, 2019 [13]; Bernier, 2020 [12]; Fuentebella, 2025 [5]; Lin, 2025 [4]; Stretton, 2025 [6]; Al-Aieshy, 2024 [7]; Zhang, 2024 [15]; Palareti, 2024 [14]; Lim, 2024 [18]; Bozic, 2023 [16]; Lin, 2023 [2]; Chen, 2022 [17]; Siedler, 2022 [3]; Ballerie, 2021 [1]; Mavri, 2021 [9].)

Figure 1. Pooled Major Bleeding Event Rate (Random-Effects Model). Forest plot showing individual study proportions and 95% confidence intervals (CIs) for major bleeding events among patients on direct oral anticoagulants (DOACs). The pooled estimate (diamond) represents a major bleeding rate of 8% (95% CI: 0.05–0.11) with substantial heterogeneity (I2 = 95.6%, p < 0.001). Weights are from a random-effects (DerSimonian–Laird, Freeman–Tukey transformation) model. The black squares represent individual study event proportions, with square size proportional to study weight; horizontal lines indicate 95% confidence intervals. The vertical dashed line denotes the pooled estimate, and the blue diamond represents the overall random-effects summary with its 95% confidence interval. (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Testa, 2019 [20]; Miklič, 2019 [13]; Bernier, 2020 [12]; Fuentebella, 2025 [5]; Lin, 2025 [4]; Palareti, 2024 [14]; Lim, 2024 [18]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 2. Major Bleeding Rate by Therapeutic Drug Monitoring (TDM) Type. Subgroup forest plot stratified by assay methodology used for DOAC concentration measurement. Pooled bleeding rates were 13% (95% CI: 0.05–0.21) for anti-Xa/apixaban assays, 17% (95% CI: 0.03–0.31) for anti-Xa only, and 6% (95% CI: 0.03–0.08) for trough/peak measurements. Between-group heterogeneity approached statistical significance (p = 0.067), indicating variability across assay types. (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Testa, 2019 [20]; Miklič, 2019 [13]; Bernier, 2020 [12]; Lin, 2025 [4]; Palareti, 2024 [14]; Lim, 2024 [18]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 3. Major Bleeding Rate by DOAC Indication (Atrial Fibrillation vs. Venous Thromboembolism). Subgroup forest plot comparing major bleeding event rates between studies enrolling patients with atrial fibrillation (AF) and venous thromboembolism (VTE). The pooled rate for AF studies was 12% (95% CI: 0.05–0.20; I2 = 96.5%), and for VTE studies 7% (95% CI: –0.05 to 0.18; I2 = 86.6%). Between-group heterogeneity was not statistically significant (p = 0.43). (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Miklič, 2019 [13]; Bernier, 2020 [12]; Lim, 2024 [18]; Siedler, 2022 [3]; Mavri, 2021 [9]; Fuentebella, 2025 [5]; Chen, 2022 [17]).

Figure 4. Pooled Thromboembolic Event Rate (Random-Effects Model). Forest plot showing individual study and pooled thrombotic event proportions. The pooled rate was 7% (95% CI: 0.05–0.09) with high heterogeneity (I2 = 87.3%, p < 0.001). Highest event rates were observed in Fuentebella 2025 [5] (26%) and Lim 2024 [18] (21%), while the lowest were Lin 2025 [4] (3%) and Palareti 2024 [14] (4%). (Nguyen, 2021 [8]; Miklič, 2019 [13]; Bernier, 2020 [12]; Fuentebella, 2025 [5]; Lin, 2025 [4]; Palareti, 2024 [14]; Lim, 2024 [18]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 5. Meta-Regression: DOAC Concentration vs. Major Bleeding (log rate) Bubble plot demonstrating the relationship between mean DOAC plasma concentration (ng/mL) and log-transformed major bleeding event rate across studies. The regression slope (red line) was not statistically significant (β = −0.00021, 95% CI −0.00067 to 0.00026; p = 0.35; Adj R2 = 0%), suggesting no meaningful correlation between drug level and bleeding risk. (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Testa, 2019 [20]; Miklič, 2019 [13]; Fuentebella, 2025 [5]; Lin, 2025 [4]; Lim, 2024 [18]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 6. Meta-regression of drug level vs. mortality. No clear association was observed between mean DOAC concentration and all-cause mortality (β = 0.00008 [95% CI −0.00033 to 0.00049], p = 0.69; Adj R2 = 0%). (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Miklič, 2019 [13]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 7. Meta-Regression: Drug Level vs. Thromboembolic Events (log rate) Scatter plot showing the relationship between mean direct oral anticoagulant (DOAC) concentration (ng/mL) and log-transformed thromboembolic event rates. The association between drug level and thromboembolic events was not statistically significant (β = 0.00005; 95% CI −0.00041 to 0.00051; p = 0.78; Adj R2 = 0%). (Nguyen, 2021 [8]; Miklič, 2019 [13]; Fuentebella, 2025 [5]; Lin, 2025 [4]; Lim, 2024 [18]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 8. Pooled Mortality Event Rate (Random-Effects Model) Forest plot showing study-specific and pooled mortality proportions. The pooled mortality rate was 3% (95% CI: 0.03–0.04) with negligible heterogeneity (I2 = 0%, p = 0.761). Mortality estimates were consistent across studies, ranging from 3% to 5%. (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Miklič, 2019 [13]; Bernier, 2020 [12]; Palareti, 2024 [14]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).

Figure 9. Mortality Rate by DOAC Indication (AF vs. VTE) Subgroup analysis comparing mortality rates by clinical indication. Among atrial fibrillation studies, the pooled mortality rate was 3% (95% CI: 0.03–0.04; I2 = 0%), while venous thromboembolism studies showed a similar rate (3%, 95% CI: −0.01 to 0.07; I2 = 0%). Between-group heterogeneity was non-significant (p = 0.79). (Jakowenko, 2020 [10]; Nguyen, 2021 [8]; Miklič, 2019 [13]; Bernier, 2020 [12]; Chen, 2022 [17]; Siedler, 2022 [3]; Mavri, 2021 [9]).