Direct oral anticoagulants (DOACs) are the treatment of choice for the prevention of cardioembolic stroke and the treatment of venous thromboembolism (Steffel et al., 2021; Stevens et al., 2021). Compared with vitamin K antagonists, they offer the advantage of not requiring therapeutic monitoring due to their relatively wide therapeutic range (Rahmat and Lip, 2015). However, outside of pivotal trials, which typically exclude patients with multiple comorbidities and those with significant drug–drug interactions, real-life patients may accumulate several risk factors that increase the risk of under- or overdosing (Terrier et al., 2022). Although no therapeutic interval has yet been validated (Moner-Banet et al., 2020), a growing body of evidence from both pivotal and real-life studies suggests an association between DOAC plasma concentrations and clinical outcomes (Eikelboom et al., 2017; Testa et al., 2024; Lin et al., 2023; Suwa et al., 2023; Wu et al., 2025; Ain et al., 2024). High peak (C-max) and trough (C-trough) concentrations of DOACs have been linked to an increased risk of bleeding, while low C-trough levels have been associated with ischemic events (Eikelboom et al., 2017; Testa et al., 2024; Lin et al., 2023; Suwa et al., 2023; Wu et al., 2025). A therapeutic range, thus, appears to be emerging and could be useful in guiding treatment for patients at higher risk of clinical events. In this study, we present the results of clinical follow-up of patients enrolled in the OptimAT study (NCT03477331), whose primary objective was to validate pharmacokinetic (PK) models in a hospitalized population treated with antithrombotic agents. The present analysis represents a secondary, exploratory, and hypothesis-generating endpoint of a PK-focused study aiming to describe the association between apixaban or rivaroxaban plasma concentrations and clinical events in this cohort over a 2-year prospective follow-up. Accordingly, this analysis was not designed to support causal or outcome-based inference.

A total of 100 apixaban- and 100 rivaroxaban-treated patients hospitalized in the Geneva University Hospitals were included in this study. The mean age of the population was 74.7 years, with 39% of the population being women, and the main reasons for apixaban or rivaroxaban treatment were atrial fibrillation (AF) in 76.5% vs. VTE in 23.5% of the cases (Table 1). An apixaban- or rivaroxaban-based dose consistent with the Summary of Product Characteristics (SmPC) was observed in 76% and 87% of the patients treated with apixaban and rivaroxaban, respectively. Among patients whose dose was not consistent with the SmPC, 23 of 24 (96%) patients receiving apixaban were prescribed a lower dose than recommended, compared with 7 of 13 (54%) patients treated with rivaroxaban. As shown in Supplementary Table S1, apixaban and rivaroxaban plasma concentrations exhibited wide distributions among patients. For apixaban, coefficients of variation were 38.8% for AUC 0 h–8 h, 38.8% for C-max, and 43.1% for C-trough concentrations. For rivaroxaban, coefficients of variation were 33.8% for AUC 0 h–8 h, 32.7% for C-max, and 55.1% for C-trough concentrations. Interquartile ranges and the minimum–maximum values for all pharmacokinetic parameters are reported in Supplementary Table S1. Reasons for loss to follow-up in patients without clinical events are provided in Supplementary Table S2. During the 2-year follow-up, 71 patients had their data censored due to treatment modifications (dosage changes or changes in treatment), 23 due to discontinuation of anticoagulation, 18 died (with only 1 death related to a thrombotic complication), and 2 were lost to follow-up or withdrew from the study. Changes to another oral anticoagulant, to a different apixaban or rivaroxaban dosing or discontinuation of anticoagulation occurred for the following reasons: as planned for 27 patients according to VTE treatment guidelines, arbitrarily by a treating physician for 15 patients, due to renal or hepatic failure for 13 patients, and for other reasons for 38 patients. Respectively, 23% and 28% of patients had been treated with apixaban and rivaroxaban for more than 1 week at the time of inclusion. The mean follow-up duration was 397 (standard deviation (SD) = 305) days. A total of 59 patients reached the end of the study. A total of 27 clinical events were observed (Supplementary Table S3), corresponding to an incidence rate of 9.3 per 100 patient-years for any bleeding event (95% CI: 5.57–14.44) and 3.9 per 100 patient-years for ischemic or thromboembolic events (95% CI: 1.68–7.67). These included one major bleeding event recorded (apixaban) (0.5%), 18 minor bleeding events (9%) (N = 8 for apixaban, N = 10 for rivaroxaban), 6 ischemic events (3%) (N = 3 for apixaban, N = 3 for rivaroxaban), and 2 thromboembolic events (1%) (N = 1 for apixaban, N = 1 for rivaroxaban). Patient clinical events by percentile and molecule are presented in Supplementary Tables S4, S5. The Kaplan–Meier survival curves illustrating the cumulative incidence of clinical events (bleeding, ischemic, and thromboembolic) according to apixaban and rivaroxaban PK parameters are shown in Figure 1. As previously defined, for hemorrhagic events, patients in the highest quartile of plasma concentrations (>75th percentile) for AUC 0 h–8 h, C-trough, and C-max were compared with those in the lower quartiles (≤75th percentile). Conversely, for ischemic and thromboembolic events, patients in the lowest quartile (<25th percentile) were compared with those in the higher quartiles (≥25th percentile). Cox regression analyses showed no significant association between the plasma PK parameters of patients in the lowest/highest quartiles of apixaban and rivaroxaban and the occurrence of ischemic/thrombotic or bleeding events, respectively, even after adjusting for the HAS-BLED score for bleeding events (Figure 1; Table 2). Power (%) for various assumed survivals under the alternative hypothesis and the comparison between two independent groups of 100 patients (per group) can be found in Supplementary Table S6. In the present cohort, the power was 20% to show a 5% statistically significant difference, with, for example, 95% vs. 90% of event-free survivals at 2 years between two independent groups [e.g., the highest quartile of plasma concentrations (>75th percentile) vs. the lower quartiles (≤75th percentile) for bleeding events]. The present study had 90% power to detect a 20% absolute difference in the incidence of events at 2 years (Supplementary Table S6).

Based on the OptimAT cohort of 200 patients treated by apixaban or rivaroxaban, we failed to detect a statistically significant association between the lowest/highest quartiles of plasma concentrations for three key PK parameters (AUC 0 h–8 h, C-trough, and C-max) and the occurrence of ischemic/thrombotic or bleeding events, respectively. These neutral results contrast with several recent publications showing such an association in pivotal studies and real-world settings (Eikelboom et al., 2017; Testa et al., 2024; Lin et al., 2023; Suwa et al., 2023; Palareti et al., 2024). In these latter studies, high C-trough and C-max levels of DOACs were associated with an increased risk of bleeding, while low C-trough levels were associated with thrombotic events. Different hypotheses can be proposed to explain this discrepancy. First, the sample size of the OptimAT study was calculated for the primary objective of validating PK models. As a result, with an observed rate of 4% for ischemic/thromboembolic events and a lower-than-expected rate of 0.5% for major bleeding events, the present study was underpowered to detect such an association, as shown by our power calculation (Supplementary Table S6). In comparison, the MAS study by Testa et al. (2024) involved 1,657 patients with AF treated with DOACs, displaying similar characteristics compared to the OptimAT population, and it reported a rate of 1.3% thrombotic complications and 3.1% bleeding events (1.9% major bleeding and 1.2% clinically relevant non-major bleeding). The study by Lin et al. (2023) included 859 patients and reported a rate of 1.3% per year for thrombotic events and 1.6% per year for major bleeding events. Second, extreme low and high concentrations were infrequent in our cohort compared to those reported in other studies. Using the threshold applied by Lin et al. (2023), only 1% of OptimAT patients had lower C-trough levels and 0.5% had higher C-trough levels, compared to 14.6% and 9.0%, respectively, in the Lin et al. cohort. Third, in the MAS trial (Palareti et al., 2024), patients started DOACs within a month before the DOAC concentration measurement (Palareti et al., 2024), which is a known period of higher bleeding risk. This was not the case for our patients, as approximately a quarter had already been receiving apixaban or rivaroxaban for more than 1 week at the time of inclusion. Fourth, half of the follow-ups were interrupted before 6 months for rivaroxaban and 1 year for apixaban, respectively. Approximately one-third of the interruptions were due to treatment modifications, more than half of which were justified for planned reasons, such as VTE management guidelines or renal dysfunction. In comparison, only 1.7% of patients per year had a treatment modification or discontinuation in the MAS trial, and none of the patients changed their DOAC dosage. In other studies and across various databases, the rate of DOAC discontinuation ranges from 13% to 64% (Cools et al., 2021; Sabaté et al., 2021).

Beyond this study’s lack of power, another key limitation, shared with other related studies on the same topic, is that apixaban and rivaroxaban concentrations were measured on a given day, which may not correlate with the timing of the clinical event. We cannot guarantee that the concentrations were the same at the time of the clinical event. This is especially true in the context of hospitalization, where several factors, such as drug–drug interactions and inflammation, can affect drug concentrations (Lenoir et al., 2021). Furthermore, patients were administered different dosages based on their comorbidities and therapeutic indications. The objective, however, was to evaluate the relationship between plasma concentrations and clinical events for apixaban or rivaroxaban, irrespective of the indication or dosage. Therefore, a percentile-based approach was preferred over a threshold-based approach for the analysis. Despite these limitations, our study has several strengths. The prospective design ensured the collection of high-quality data. Additionally, clinical events were verified using medical records for validation. We also chose to censor data at the date of treatment modifications, which was not the case in other studies. This ensures that clinical events were recorded only in relation to the initial apixaban or rivaroxaban dose and the initial plasma concentration.

The surprisingly high rate of apixaban and rivaroxaban modification or discontinuation in our real-life setting, coupled with a relatively small sample size of a PK study, did not allow the detection of a significant risk of ischemic/thromboembolic or bleeding events associated with low or high apixaban or rivaroxaban plasma exposure, respectively. However, monitoring clinical events within the context of such a study provides valuable descriptive data on their incidence, which is comparable to that observed in landmark trials. Although underpowered to detect small differences, this study highlights the challenges of conducting long-term pharmacokinetic–clinical outcome studies in hospitalized populations and emphasizes the value of reporting such results to reduce publication bias associated with the underreporting of neutral studies in future meta-analyses.