Rivaroxaban Initiation Timing and Clinical Outcomes After Acute Ischemic Stroke in Patients with Non-Valvular Atrial Fibrillation: An IPTW-Based Cohort Study

Abstract

Background

The optimal timing of anticoagulation initiation after acute ischemic stroke (AIS) in patients with non-valvular atrial fibrillation (NVAF) remains uncertain. We evaluated the association between early versus delayed rivaroxaban initiation and clinical outcomes in this population.

Methods

In this retrospective, single-center cohort study, 401 hospitalized AIS patients with NVAF receiving rivaroxaban between January 2019 and June 2024 were classified into early (≤7 days) or delayed (>7 days) initiation groups. Primary outcomes were excellent [modified Rankin Scale (mRS) 0-1] and good (mRS 0-2) functional outcomes at 90 days and 12 months. Secondary outcomes included recurrent ischemic stroke, clinically relevant major bleeding, all-cause mortality, composite outcomes, and in-hospital length of stay (LOS). Inverse probability of treatment weighting (IPTW) was applied, followed by regression analyses. Sensitivity and subgroup analyses were performed.

Results

After IPTW adjustment, early rivaroxaban initiation was associated with a higher likelihood of achieving mRS 0-1 at 90 days (aOR 1.76, 95% CI 1.02-3.05) and 12 months (aOR 1.82, 95% CI 1.08-3.06), as well as a shorter LOS. No significant differences were observed for mRS 0-2 or other secondary outcomes. Sensitivity analyses yielded consistent results, and subgroup analyses suggested a trend toward greater benefit among patients with higher baseline stroke severity (NIHSS >7).

Conclusion

Early rivaroxaban initiation after AIS in patients with NVAF was associated with improved excellent functional outcomes without an apparent increase in adverse events, supporting its potential feasibility in routine practice and warrant confirmation in larger prospective studies.

Introduction

Acute ischemic stroke (AIS) results from an abrupt interruption of cerebral blood flow due to arterial thrombosis or embolism, remaining a leading cause of long-term disability and death worldwide. ¹ Atrial fibrillation (AF), the most prevalent cardiac arrhythmia, promotes blood stasis in the left atrium, markedly increasing the risk of thromboembolic stroke. ² Indeed, AF independently elevates stroke risk by nearly fivefold, and AF-related strokes tend to present with greater severity and higher rates of mortality, recurrence, and disability compared with other etiologies.^3,4 The burden of AF-associated AIS is particularly pronounced in China, where roughly one-quarter of AIS patients have AF, the vast majority of which is non-valvular atrial fibrillation (NVAF).^5,6 Despite the widespread use of anticoagulation, patients with AIS and AF remain at considerable residual risk, with recurrent stroke occurring in 3.0% at 3 months and 7.0% at 1 year, and mortality reaching 12.4% and 18.1% at the corresponding time points. ⁷ A prospective study of Chinese patients with AIS and AF over three years documented an 18% rate of recurrent ischemic stroke (IS) and a 38% all-cause mortality, highlighting the poor prognosis and high clinical burden of this population. ⁸

Oral anticoagulation represents the cornerstone of secondary prevention in NVAF, reducing the risk of fatal and disabling stroke by over 60% and overall stroke incidence by approximately two-thirds compared with no anticoagulation. ⁹ In our previous work, acute-phase anticoagulant therapy in patients with AIS and AF yielded superior neurological outcomes and lower risks of mortality, lower-limb venous thrombosis, and major bleeding compared with antiplatelet regimens. ¹⁰ Nevertheless, data from the GLORIA-AF registry revealed that fewer than half of Chinese patients with AF received any oral anticoagulant (OAC). Among those treated, only a quarter received non-vitamin K antagonist oral anticoagulants (NOACs). ¹¹ Rivaroxaban is one of the most commonly prescribed NOACs for stroke prevention in NVAF and has demonstrated non-inferiority efficacy with a favorable safety profile compared with warfarin in large randomized trials (ROCKET-AF, J-ROCKET-AF).^12,13

Beyond the choice of anticoagulant, the timing of anticoagulation initiation after AIS has become a central issue in patients with NVAF, as clinicians must balance the risk of early recurrent ischemia against the potential for hemorrhagic complications. Traditionally, initiation timing has been guided by expert consensus, most notably the widely cited “1–3–6–12 days rule”. ¹⁴ However, data from the multicenter Berlin Atrial Fibrillation Registry showed that adherence to this rule was not associated with differences in 3-month composite outcomes, ¹⁵ raising questions about its applicability in routine practice. In recent years, several randomized controlled trials (RCTs) have compared early versus delayed initiation of NOACs in patients with AIS and NVAF.^16–19 These studies generally reported no excess risk of major bleeding with earlier initiation, while efficacy outcomes appeared comparable between early and delayed strategies. The RELAXED study further demonstrated that initiation of rivaroxaban within 14 days after AIS was feasible and safe, with low rates of recurrent IS and major bleeding at 90 days. ²⁰ Despite these trial data, evidence from real-world clinical settings remains limited, particularly regarding functional recovery following different rivaroxaban initiation strategies.

Against this background, this present study aimed to examine the association between early versus delayed initiation of rivaroxaban and clinical outcomes in patients with AIS and NVAF using real-world data. Patterns of rivaroxaban dosing were evaluated as a secondary, exploratory component to describe contemporary prescribing practices, and were not considered a primary exposure of interest.

Methods

Study Design

This retrospective, observational, single-center study was conducted at Nanjing Drum Tower Hospital and included patients admitted between January 2019 and June 2024. The study was approved by the Ethics Committee of Nanjing Drum Tower Hospital (Ethics Number: 2023-315-01) and was conducted in accordance with the principles of the Declaration of Helsinki.

A total of 1019 consecutive patients diagnosed with AIS and AF were initially screened. Inclusion criteria: (1) Patients with AIS, defined as an acute neurologic deficit caused by focal brain, spinal cord, or retinal infarction, in line with the tissue-based definition recommended by AIS guidelines, with the diagnosis confirmed by neuroimaging ²¹ ; (2) Patients with AF were defined as either previously documented AF or new-onset AF diagnosed by electrocardiogram (ECG) during hospitalization. ²² NVAF was defined as newly diagnosed or pre-existing, in the absence of moderate-to-severe mitral stenosis or mechanical heart valves on ECG. Exclusion criteria: (1) Did not receive rivaroxaban during hospitalization, including those treated with antiplatelet therapy only, low-molecular-weight heparin (LMWH), dabigatran, edoxaban, warfarin, or no antithrombotic therapy; (2) Patients with an unknown onset time; (3) Initiation time of rivaroxaban more than 30 days after AIS onset; (4) Lost to follow-up. After applying these criteria, the final study cohort consisted of patients with AIS and confirmed NVAF who received rivaroxaban during hospitalization.

Patients were categorized into two groups based on the timing of rivaroxaban initiation after AIS: (1) Early-initiation group: Patients who started rivaroxaban ≤7 days after AIS onset; and (2) Delayed-initiation group: Patients who started rivaroxaban >7 days after AIS onset.

Data Collection

Patients’ baseline data were extracted from the electronic medical records system, including demographics, baseline scores, past medical history, comorbid conditions, and admission laboratory results. Baseline variables were defined as those measured or occurring before rivaroxaban initiation. In-hospital treatment strategies focused on key acute-phase management decisions, including reperfusion therapy with mechanical thrombectomy and/or intravenous thrombolysis, as per current guidelines. As these interventions occurred before the initiation of rivaroxaban, they were treated as baseline clinical variables. Detailed definitions of all study variables and their inclusion in the propensity score analysis are provided in Supplementary Table 1. Concomitant antiplatelet therapy was defined as the use of antiplatelet agents (primarily aspirin or clopidogrel) in combination with rivaroxaban at discharge. The initial rivaroxaban dose (≤10, 15, or ≥20 mg/d) was recorded descriptively but was not treated as an exposure or adjustment variable in analyses of initiation timing, as dose selection occurred concurrently with or after treatment initiation. Other routine in-hospital treatments (eg, statin therapy, neuroprotective agents, and supportive care) were provided according to standard practice and were not considered exposure variables. Concurrent infection (co-infection) was defined as any new-onset infection diagnosed during hospitalization after AIS onset, recorded descriptively and excluded from adjusted analyses because its timing relative to rivaroxaban initiation could not be reliably determined. Hemorrhagic transformation was defined as intracerebral hemorrhage confirmed by neuroimaging occurring after AIS onset ²³ but before the initiation of rivaroxaban therapy. Stroke severity at admission was assessed using the National Institutes of Health Stroke Scale (NIHSS), ²⁴ with scores of 0-7 considered to represent mild stroke. ²⁵ Thromboembolic and bleeding risks were evaluated using the CHA₂DS₂-VASc-60 and HAS-BLED scores, respectively.^26,27 Admission laboratory results included coagulation markers, inflammatory indicators, metabolic parameters, hemoglobin (HB), and renal markers.

Primary neurological functional outcomes assessed using the modified Rankin Scale (mRS), ²⁸ included excellent functional outcome (mRS 0-1) and good functional outcome (mRS 0-2). ²⁹ Secondary outcomes, defined as clinical events, included recurrent IS, clinically relevant major bleeding (CRMB), all-cause mortality, and composite outcomes. Recurrent IS was defined as new focal neurological deficit confirmed by imaging evidence of acute focal ischemia on CT or MRI. ³⁰ CRMB was defined according to the International Society on Thrombosis and Hemostasis (ISTH) criteria as clinically significant bleeding meeting ≥1 of the following criteria: fatal bleeding, a hemoglobin drop of ≥ 2 g/dL, transfusion of ≥ 2 units of red blood cells, or intracranial hemorrhage. ³¹ All-cause mortality was defined as death from any cause during the follow-up period. The composite outcomes were defined as the occurrence of at least one of the following events, including recurrent IS, CRMB, or all-cause mortality. These outcomes can be reliably assessed through outpatient follow-up and telephone interviews at 90 days and 12 months, making them well-suited for retrospective cohort studies. In-hospital length of stay (LOS) was also collected as an additional outcome.

Statistical Analysis

All statistical analyses were conducted using R software (version 4.5.2). Missing data were addressed using multiple imputation by chained equations (MICE) with the “mice” package. Twenty imputed datasets were generated (m = 20) over 50 iterations (maxit = 50) using predictive mean matching (method = “pmm”) for continuous variables. A fixed seed (seed = 123) was set to ensure reproducibility. Convergence and the distribution of imputed values were checked to ensure plausibility of the imputed data. The 20th imputed dataset was selected for subsequent analyses. The “BaylorEdPsych” package was used to test whether the data were missing completely at random (MCAR), and the “VIM” package was used to visually inspect missingness patterns.

Continuous variables were assessed for normality using the “pastecs” package. Normally distributed variables were presented as mean ± standard deviation (SD) and compared using the t-test, while non-normally distributed variables were summarized as median with interquartile range (IQR) and analyzed using the Wilcoxon rank-sum test. Categorical variables were presented as counts and percentages and compared using the chi-square test or Fisher's exact test. Restricted cubic spline (RCS) analyses were performed on the original data using the “rms” package to explore the association between rivaroxaban initiation time and neurological functional outcomes.

To reduce confounding and selection bias, inverse probability of treatment weighting (IPTW) was applied using the “survey” package. Propensity scores were estimated using a logistic regression model incorporating baseline demographic, clinical, and laboratory variables measured before rivaroxaban initiation. Balance between groups before and after IPTW was assessed via absolute standardized mean differences (SMD). Variables with SMD > 0.1 were considered unbalanced and adjusted for in multivariable analyses using the IPTW-weighted data. The mRS score distributions at 90 days and 12 months were displayed as stacked bar charts before IPTW.

Univariate and multivariate logistic regression models were used to assess the associations between rivaroxaban initiation groups and neurological functional outcomes at 90 days and 12 months. Additionally, univariate and multivariate Cox proportional hazards models were conducted using the “survival” package to evaluate time-to-event outcomes, including recurrent IS, CRMB, all-cause mortality, and composite outcomes. LOS was analyzed as a continuous variable using linear regression models. An exploratory dose-stratified analysis was conducted, with event rates compared between early- and delayed-initiation groups within each rivaroxaban dose (≤10, 15, or ≥20 mg/d).

Covariates with p-values <.1 in univariate analysis were entered into multivariate models. Multicollinearity was assessed using the variance inflation factor (VIF) from the “car” package, and the final models included variables with VIF < 3. Kaplan-Meier curves and cumulative incidence plots after IPTW were generated using the “survival” and “survminer” packages, and group differences were assessed by the log-rank test. Subgroup analyses were performed using the “jstable” package, stratified by age, NIHSS at admission, endovascular therapy, and cerebral artery occlusion to examine the consistency of treatment effect. An interaction p value <.10 was considered indicative of effect modification between rivaroxaban initiation timing and subgroup variables. ³² Sensitivity analyses were conducted by performing a complete-case analysis excluding patients with missing data, and a separate analysis excluding patients with hemorrhagic transformation to evaluate the robustness of the findings. A two-sided p-value <.05 was considered statistically significant.

Results

Baseline Characteristics

The study flow chart is illustrated in Figure 1. A total of 401 patients diagnosed with AIS and NVAF, all receiving rivaroxaban for secondary stroke prevention, were included. Missing values were observed in 1.2% of patients (n = 5) for selected laboratory variables, including low-density lipoprotein cholesterol (LDL-C), fasting blood glucose (FBG), C-reactive protein (CRP), prothrombin time (PT), activated partial thromboplastin time (APTT), fibrinogen (FIB) and D-dimer. Little's MCAR test confirmed that these missing values were missing completely at random (p = .771). The distribution of missing data is presented in Supplementary Figure 1.

Figure 1.

Flow chart of the study. Abbreviations: AIS, acute ischemic stroke; AF, atrial fibrillation; LMWH, low-molecular-weight heparin; NVAF, non-valvular atrial fibrillation; IPTW, inverse probability of treatment weighting; mRS, modified Rankin Scale; IS, ischemic stroke; CRMB, clinically relevant major bleeding; LOS, length of stay.

Before IPTW, the early-initiation group had a higher proportion of patients with NIHSS ≤ 7 (70.9% vs 44.0%, p < .001) and higher CHA2DS2-VASc-60 scores (median 7 vs 6, p = .036). Regarding medical history, the early-initiation group had a significantly greater prevalence of previous stroke (48.1% vs 33.7%, p = .005). Newly diagnosed NVAF occurred more frequently in the delayed-initiation group than in the early-initiation group (35.0% vs 13.9%, p < .001). Conversely, the delayed-initiation group more frequently had indicators of greater acute severity or subsequent complications, including cerebral artery occlusion (45.3% vs 34.2%, p = .029), hemorrhagic transformation (14.8% vs 3.2%, p < .001), and co-infection (44.4% vs 25.3%, p < .001). Consistent with these baseline differences, laboratory findings showed higher levels of white blood cell count (WBC), LDL-C, and D-dimer, along with shorter PT and APTT in the delayed-initiation group (all p < .05). Endovascular therapy was less frequently performed in the early-initiation group (17.7% vs 39.5%, p < .001). Regarding the initial rivaroxaban dose, the early-initiation group had a higher proportion of patients receiving the ≥20 mg/d dose (17.7% vs 11.5%), whereas lower doses (≤10 mg/d) were more frequent in the delayed-initiation group (29.1% vs 35.8%). The proportion of patients receiving the 15 mg/d dose was similar between the two groups. Following IPTW adjustment, baseline demographic and clinical characteristics between the two groups were largely balanced, as summarized in Table 1 and in Supplementary Figure 2. Variables with absolute SMD >0.1 after IPTW included hemorrhagic transformation, age ≤75 years, newly diagnosed NVAF, platelet count (PLT), fasting blood glucose (FBG), NIHSS at admission ≤7, and HAS-BLED score.

Table 1.

Comparison of Baseline Characteristics Between Early and Delayed Rivaroxaban Initiation Before and After IPTW in AIS Patients with NVAF.

	Original				IPTW
	Early (N = 158)	Delayed (N = 243)	p value	SMD	Early (N = 450)	Delayed (N = 391)	p value	SMD
Demographic data (n, %)
Age (median [IQR])	77.00 [70.00, 84.00]	75.00 [68.50, 83.00]	.336	0.080
Age ≤ 75	69 (43.7)	127 (52.3)	.102	0.173	255.5 (56.8)	194.1 (49.6)	.327	0.143
Gender (Female)	76 (48.1)	101 (41.6)	.217	0.132	207.6 (46.1)	177.6 (45.4)	.929	0.014
Baseline scores (median, IQR)
NIHSS at admission	5.00 [2.00, 9.00]	9.00 [4.00, 15.00]	<.001	0.486
NIHSS at admission ≤ 7 (n, %)	112 (70.9)	107 (44.0)	<0.001	0.564	215.6 (47.9)	211.7 (54.1)	.405	0.131
CHA2DS2-VASc-60	7.00 [6.00, 8.00]	6.00 [6.00, 7.00]	.036	0.155	7.00 [6.00, 7.00]	7.00 [6.00, 7.00]	.444	0.037
HAS-BLED	3.00 [2.00, 3.00]	3.00 [2.00, 3.00]	.279	0.143	3.00 [2.00, 3.00]	3.00 [2.00, 3.00]	.237	0.103
Past medical history (n, %)
Smoking history	24 (15.2)	48 (19.8)	.287	0.120
Drinking history	17 (10.8)	42 (17.3)	.084	0.189
Hypertension	123 (77.8)	176 (72.4)	.242	0.126	340.9 (75.7)	292.7 (74.8)	.887	0.020
Diabetes	50 (31.6)	67 (27.6)	.431	0.089	115.4 (25.6)	111.0 (28.4)	.641	0.061
Coronary heart disease	47 (29.7)	52 (21.4)	.075	0.192	105.9 (23.5)	90.5 (23.2)	.953	0.009
Thromboembolism	12 (7.6)	17 (7.0)	.845	0.023	40.1 (8.9)	29.9 (7.6)	.756	0.049
Heart failure	39 (24.7)	46 (18.9)	.172	0.140	75.2 (16.7)	79.0 (20.2)	.446	0.085
Previous stroke	76 (48.1)	82 (33.7)	.005	0.295	161.1 (35.8)	146.8 (37.5)	.793	0.036
Renal insufficiency	10 (6.3)	17 (7.0)	.841	0.027	25.0 (5.5)	22.8 (5.8)	.924	0.011
Bleeding history	19 (12.0)	37 (15.2)	.381	0.093	52.0 (11.6)	52.4 (13.4)	.679	0.054
Malignancy	13 (8.2)	25 (10.3)	.601	0.071	36.3 (8.1)	36.4 (9.3)	.715	0.043
Newly diagnosed NVAF (n, %)	22 (13.9)	85 (35.0)	<.001	0.505	163.7 (36.4)	109.9 (28.1)	.308	0.198
Laboratory results on admission (median, IQR)
WBC (109/L)	6.70 [5.50, 8.45]	7.40 [6.00, 9.00]	.010	0.238	7.00 [5.43, 9.50]	7.10 [5.77, 8.70]	.903	0.061
LYM (109/L)	1.50 [1.10, 1.80]	1.40 [1.00, 1.90]	.463	0.021	1.50 [1.00, 1.80]	1.40 [1.10, 1.90]	.899	0.018
HB (g/L)	134.50 [124.25, 147.00]	136.00 [125.00, 151.00]	.289	0.107	134.37 [119.71, 151.38]	134.00 [124.00, 149.02]	.767	0.045
PLT (109/L)	180.00 [140.25, 219.75]	177.00 [145.50, 218.00]	.903	<0.001	189.00 [142.28, 228.89]	176.00 [146.43, 218.00]	.364	0.112
TG (mmol/L)	1.02 [0.80, 1.33]	1.06 [0.78, 1.37]	.737	0.077	0.98 [0.71, 1.30]	1.05 [0.75, 1.36]	.609	0.088
LDL-C (mmol/L)	2.05 [1.58, 2.65]	2.28 [1.75, 2.95]	.019	0.228	2.06 [1.54, 2.84]	2.24 [1.72, 2.86]	.580	0.063
FBG (mmol/L)	5.20 [4.67, 6.19]	5.39 [4.62, 6.75]	.289	0.125	5.18 [4.67, 5.98]	5.27 [4.53, 6.55]	.559	0.111
eGFR (mL/min/1.73m2)	89.70 [73.58, 108.88]	89.70 [75.85, 105.15]	.756	0.036	90.05 [69.92, 114.92]	89.70 [75.65, 103.33]	.662	0.036
CRP (mg/L)	4.55 [2.80, 11.07]	5.40 [3.30, 15.90]	.076	0.047	5.20 [3.50, 9.89]	5.20 [3.16, 14.52]	.848	0.050
PT (s)	11.90 [11.33, 12.78]	11.60 [11.20, 12.20]	.001	0.178	11.80 [11.30, 12.30]	11.60 [11.20, 12.18]	.073	0.013
APTT (s)	27.35 [25.83, 29.50]	26.10 [24.90, 27.75]	<.001	0.458	26.45 [24.59, 27.89]	26.43 [25.14, 28.20]	.555	0.090
FIB (g/L)	2.90 [2.50, 3.50]	3.00 [2.50, 3.50]	.862	0.053	3.00 [2.60, 3.30]	3.00 [2.60, 3.50]	.844	0.004
D-dimer (mg/L)	0.65 [0.32, 1.30]	1.02 [0.50, 2.35]	<.001	0.019	0.68 [0.33, 1.39]	0.95 [0.44, 2.13]	.100	0.052
Comorbidities (n, %)
Co-infection	40 (25.3)	108 (44.4)	<.001	0.410
Hemorrhagic transformation	5 (3.2)	36 (14.8)	<.001	0.416	62.8 (14.0)	41.1 (10.5)	.645	0.123
Cerebral artery occlusion	54 (34.2)	110 (45.3)	.029	0.228	203.1 (45.1)	160.9 (41.2)	.614	0.082
In-hospital treatment strategies (n, %)
Endovascular therapy	28 (17.7)	96 (39.5)	<.001	0.497	158.0 (35.1)	123.7 (31.6)	.671	0.079
Concomitant antiplatelet therapy at discharge	17 (10.8)	20 (8.2)	.480	0.086
Initial rivaroxaban dose
≤ 10 mg/d	46 (29.1%)	87 (35.8%)	.141	0.201
15 mg/d	84 (53.2%)	128 (52.7%)
≥ 20 mg/d	28 (17.7%)	28 (11.5%)

Abbreviations: IPTW, inverse probability of treatment weighting; AIS, acute ischemic stroke; NVAF, Non-valvular Atrial Fibrillation; SMD, standardized mean differences; IQR, interquartile range; NIHSS, National Institute of Health Stroke Scale; CHA2DS2-VASc-60, congestive heart failure, hypertension, age 65 years or older, diabetes mellitus, previous stroke/transient ischemic attack, vascular disease, age 60 to 65 years, female; HAS-BLED, hypertension, abnormal renal or liver function, stroke, bleeding history, labile international normalized ratio, age 65 years or older, and drugs that increase bleeding tendency or alcohol use; Co-infection, concurrent infection; WBC, white blood cell; LYM, lymphocyte count; HB, hemoglobin; PLT, platelet count; TG, triglyceride; LDL-C, low-density lipoprotein cholesterol; FBG, fasting blood glucose; eGFR, estimated Glomerular Filtration Rate; CRP, C-reactive protein; PT, prothrombin time; APTT, activated partial thromboplastin time; FIB, fibrinogen.

Initiation Timing of Rivaroxaban

RCS analyses depicted the association between initiation timing of rivaroxaban and neurological functional outcomes at 90 days in the original data (Supplementary Figure 3). The spline curves for mRS 0-1 and mRS 0-2 exhibited inflection points at 6.02 days and 6.65 days, respectively, indicating a change in trend near the end of the first post-stroke week. These results support the pre-specified 7-day threshold used to distinguish early from delayed rivaroxaban initiation.

Clinical Outcomes

The incidence of clinical outcomes according to rivaroxaban initiation timing is summarized in Table 2. Among the 401 patients analyzed, favorable functional outcomes were common: at 90 days, 46.6% achieved mRS 0-1 and 64.6% achieved mRS 0-2, with similar proportions observed at 12 months. Patients in the early-initiation group had significantly higher rates of favorable functional outcomes than those in the delayed-initiation group. At 90 days, early initiation was associated with higher proportions of mRS 0-1 (57.6% vs 39.5%, p = .001) and mRS 0-2 (72.8% vs 59.3%, p = .008). Similar patterns were observed at 12 months. The distribution of mRS scores at 90 days and 12 months is shown in Supplementary Figure 4.

Table 2.

The Incidence of Clinical Outcomes According to Early Versus Delayed Initiation of Rivaroxaban.

Outcomes	Total (N = 401)	Early (N = 158)	Delayed (N = 243)	p Value
Primary outcomes
mRS 0-1 at 90d	187 (46.6%)	91 (57.6%)	96 (39.5%)	.001
mRS 0-2 at 90d	259 (64.6%)	115 (72.8%)	144 (59.3%)	.008
mRS 0-1 at 12m	194 (48.4%)	92 (58.2%)	102 (42.0%)	.002
mRS 0-2 at 12m	248 (61.8%)	109 (69.0%)	139 (57.2%)	.023
Secondary outcomes
Recurrent IS at 90d	23 (5.7%)	9 (5.7%)	14 (5.8%)	1.000
Recurrent IS at 12m	54 (13.5%)	20 (12.7%)	34 (14.0%)	.816
CRMB at 90d	20 (5.0%)	9 (5.7%)	11 (4.5%)	.771
CRMB at 12m	24 (6.0%)	9 (5.7%)	15 (6.2%)	1.000
All-cause mortality at 90d	22 (5.5%)	7 (4.4%)	15 (6.2%)	.600
All-cause mortality at 12m	50 (12.5%)	18 (11.4%)	32 (13.2%)	.710
Composite outcomes at 90d	59 (14.7%)	23 (14.6%)	36 (14.8%)	1.000
Composite outcomes at 12m	106 (26.4%)	42 (26.6%)	64 (26.3%)	1.000
In-hospital LOS, days	13 (10-16)	11 (8-14)	14 (11-18)	<.001

Abbreviations: mRS, modified Rankin Scale; IS, ischemic stroke; CRMB, clinically relevant major bleeding; LOS, length of stay. Categorical variables were compared using the chi-square test or Fisher's exact test, and continuous variable (LOS) were compared using the Wilcoxon rank-sum test.

During follow-up, rates of recurrent IS, CRMB, all-cause mortality, and composite outcomes did not differ significantly between the early- and delayed-initiation groups at either time point (all p > .05).

The median LOS was 13 days (IQR 10-16), and was significantly shorter in the early-initiation group compared with the delayed-initiation group (median 11 [IQR 8-14] vs 14 [IQR 11-18] days, p < .001).

Primary Outcomes

Figure 2 presents the associations between rivaroxaban initiation timing and clinical outcomes, based on univariate, multivariate, and IPTW analyses. In the univariate analysis, patients in the early-initiation group were more likely to achieve mRS 0-1 and mRS 0-2 at both 90 days and 12 months (all p < .05). After multivariable adjustment, the associations remained significant for mRS 0-1 at 90 days [adjusted OR (aOR) 1.72, 95% CI 1.04-2.83, p = .034] and at 12 months (aOR 1.62, 95% CI 1.01-2.62, p = .047), whereas associations with mRS 0-2 were no longer significant. Similar results were observed in IPTW-adjusted analyses, with the early-initiation group being associated with mRS 0-1 at 90 days (aOR 1.76, 95% CI 1.02-3.05, p = .043) and at 12 months (aOR 1.82, 95% CI 1.08-3.0, p = .025), with no significant differences observed for mRS 0-2.

Figure 2.

Comparisons of clinical outcomes between the early- and delayed-initiation rivaroxaban groups using univariate, multivariate, and IPTW analyses. Abbreviations: IPTW, inverse probability of treatment weighting; mRS, modified Rankin Scale; OR, odds ratio; HR, hazard ratio; aOR, adjusted odds ratio; aHR, adjusted hazard ratio; CI, confidence interval; IS, ischemic stroke; CRMB, clinically relevant major bleeding.

Secondary Outcomes

Before IPTW adjustment, neither univariate nor multivariate Cox regression analyses showed significant differences between the early- and delayed-initiation groups in the risks of recurrent IS, CRMB, all-cause mortality, or composite outcomes at 90 days or 12 months (all p > .05). After IPTW adjustment, the results remained consistent, with no significant associations between rivaroxaban initiation timing and any clinical events at either 90 days or 12 months. Detailed hazard ratios with 95% confidence intervals are shown in Figure 2. Kaplan-Meier survival curves for clinical events, based on IPTW-weighted data, showed no statistically significant differences between the early- and delayed-initiation groups at 90 days (all p > .05; Supplementary Figure 5) or 12 months (all p > .05; Supplementary Figure 6).

Early initiation of rivaroxaban (≤7 days) was associated with a shorter hospital stay in univariate analysis (B = −4.15, t = −3.99, p < .001; Table 3). This association remained significant after adjustment for relevant covariates in the multivariate model (B = −3.36, t = −3.15, p = .002) and in the IPTW-adjusted analysis (B = −4.25, t = −4.81, p < .001). These findings suggest that early initiation of rivaroxaban was consistently associated with reduced LOS.

Table 3.

Association of Rivaroxaban Initiation Timing with in-Hospital Length of Stay Using Univariate, Multivariate, and IPTW Analyses.

Initiation Time ≤7 Days	Univariate	Multivariate	IPTW
B	−4.15	−3.36	−4.25
t value	−3.99	−3.15	−4.81
p value	<.001	.002	<.001

Abbreviations: IPTW, inverse probability of treatment weighting.

Subgroup Analysis

Subgroup analyses of neurological functional outcomes were conducted after IPTW adjustment, stratified by age, baseline NIHSS score, endovascular therapy, and cerebral artery occlusion. Subgroup analyses of outcomes at 90 days are presented in Figure 3. For the outcome of mRS 0-1 at 90 days, early rivaroxaban initiation was associated with higher odds of an excellent functional outcome among patients with NIHSS >7 (aOR 3.21, 95% CI 1.06-9.67; p = .040). However, the interaction between initiation timing and baseline NIHSS was not statistically significant (p for interaction = .147), indicating no clear evidence of effect modification.

Figure 3.

Subgroup analyses of neurological functional outcomes at 90 days in AIS patients with NVAF after IPTW, including excellent functional outcome (mRS 0-1) and good functional outcome (mRS 0-2). Abbreviations: IPTW, inverse probability of treatment weighting; aOR, adjusted odds ratio; CI, confidence interval; mRS, modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale.

Subgroup analyses of neurological functional outcomes at 12 months are shown in Figure 4. For mRS 0-2 at 12 months, a trend toward a higher odd of a favorable outcome with early initiation was observed among patients with NIHSS >7 (aOR 2.68, 95% CI 0.86-8.39; p = .092), whereas no significance was observed in patients with NIHSS ≤7. The interaction test suggested a possible trend toward effect modification by baseline NIHSS (p for interaction = .068), which should be interpreted as exploratory.

Figure 4.

Subgroup analyses of neurological functional outcomes at 12 months in AIS patients with NVAF after IPTW, including excellent functional outcome (mRS 0-1) and good functional outcome (mRS 0-2). Abbreviations: IPTW, inverse probability of treatment weighting; aOR, adjusted odds ratio; CI, confidence interval; mRS, modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale.

Sensitivity Analysis

In the complete-case sensitivity analysis, excluding the five patients with missing data, IPTW-adjusted results were consistent with those of the primary analyses. Early initiation of rivaroxaban remained significantly associated with achieving mRS 0-1 at both 90 days and 12 months, and reduced LOS, whereas no significant associations were observed for mRS 0-2 or for other secondary outcomes (Supplementary Tables 2 and 4).

After excluding the 41 patients with hemorrhagic transformation, the results of the IPTW-adjusted sensitivity analysis were largely consistent with the primary analyses. Early initiation of rivaroxaban remained significantly associated with achieving mRS 0-1 at both 90 days (aOR 1.88, 95% CI 1.07-3.28, p = .028) and 12 months (aOR 2.21, 95% CI 1.30-3.77, p = .004). Achievement of mRS 0-2 showed a similar trend, reaching significance at 12 months (aOR 1.86, 95% CI 1.08-3.23, p = .027), while mRS 0-2 at 90 days did not reach statistical significance (aOR 1.80, 95% CI 0.97-3.36, p = .064). Consistent with the primary analyses, early initiation of rivaroxaban was associated with a shorter LOS, but no significant differences were observed for other secondary outcomes at 90 days or 12 months (Supplementary Tables 3 and 4).

The findings from both the complete-case and hemorrhagic transformation-excluded sensitivity analyses consistently support the benefit of early rivaroxaban initiation.

Exploratory Analysis of Rivaroxaban Dose

Among the 401 patients with AIS and NVAF included in the analysis, the initial rivaroxaban dose was ≤10 mg/d in 133 patients (33.2%), including 2 patients treated with 5 mg/d; 15 mg/d in 212 patients (52.9%); and ≥20 mg/d in 56 patients (14.0%), including 1 patient receiving 22.5 mg/d and 4 patients receiving 30 mg/d (Table 1).

An exploratory dose-stratified analysis was conducted to examine clinical outcomes according to early versus delayed rivaroxaban initiation within each prescribed dose category. Across dose strata, early initiation was generally associated with a higher proportion of favorable functional outcomes, although the strength of the associations varied and statistical significance was not consistently observed within individual dose groups. No consistent differences were observed across dose strata in recurrent IS, CRMB, all-cause mortality, or composite outcomes. Detailed results are presented in Supplementary Table 5.

Discussion

In recent years, evidence from large RCTs has clarified that earlier initiation of NOACs after AIS in patients with NVAF is generally safe and does not substantially increase the risk of major bleeding. However, whether earlier initiation translates into meaningful improvement in neurological recovery under routine clinical conditions remains uncertain. In this real-world cohort, we defined early rivaroxaban initiation as treatment started within 7 days after AIS. This time window was selected based on prior RCTs,^16,18 and was further supported by RCS analyses which suggested that functional recovery was most sensitive to variations in initiation timing during the first week after stroke onset. Using this definition, initiation of rivaroxaban within 7 days after AIS was associated with a significantly increased likelihood of achieving an excellent functional outcome (mRS 0-1) at both 90 days and 12 months, without an accompanying increase in the risk of CRMB. In contrast, no significant benefit was observed for the more conventional endpoint of good functional outcome (mRS 0-2) after multivariable adjustment or IPTW. Rates of recurrent IS, all-cause mortality, and composite outcomes were comparable between early and delayed initiation groups. Notably, early initiation was also associated with a shorter LOS, suggesting more rapid clinical stabilization. Our findings provide complementary real-world evidence that helps contextualize recent trial data and informs anticoagulation timing decisions in patients with AIS and NVAF.

Existing experimental and clinical observations provide a biological context for the association observed in this study. In a rat model of brain ischemia/reperfusion injury, pretreatment with rivaroxaban was associated with smaller infarct volumes and more favorable functional outcomes, without an apparent increase in intracranial hemorrhage despite effective anticoagulation.^33,34 These effects were accompanied by reduced thrombin-mediated intracerebral thrombus formation and attenuated post-ischemic inflammatory responses. Although derived from preclinical models, these findings support the biological plausibility that early rivaroxaban initiation may improve neurological outcomes, potentially through modulation of thrombo-inflammatory pathways. Taken together, these observations suggest that the association between early rivaroxaban initiation and improved functional outcome may reflect an influence on early post-stroke pathophysiological processes, rather than a reduction in overt recurrent clinical events. This interpretation is directionally consistent with real-world studies reporting lower stroke severity among patients with prior NOACs exposure.^35,36 In our study, the association between early rivaroxaban initiation (≤7 days) and excellent functional outcome (mRS 0-1) remained robust after rigorous IPTW adjustment for baseline characteristics, suggesting that the observed benefit is unlikely to be fully explained by differences in initial stroke severity. By contrast, our study found no significant difference in the rate of good functional outcome (mRS 0-2) between early and delayed rivaroxaban initiation, with an overall rate of 64.4%. This finding is consistent with the primary results of the ELAN randomized trial, which also reported no significant increase in functional independence at 90 days with early versus delayed initiation (mRS 0-2: 66.6% vs 65.8%; OR 0.93, 95% CI, 0.79-1.09), despite employing a severity-based early initiation protocol ranging from ≤ 48 hours to 6-7 days post-stroke based on infarct size. ¹⁷ In comparison, a retrospective cohort of 395 patients found that only 40.2% achieved mRS 0-2 at 3 months, with initiation within 0-4 days associated with higher odds than initiation at 5–14 days. ³⁷ This discrepancy may reflect differences in baseline functional outcome distributions across study populations, which could influence the sensitivity of dichotomized mRS endpoints to detect treatment-related differences. When the proportion of patients achieving mRS 0-2 is already high, as observed in both our cohort and the ELAN study, the mRS 0-2 endpoint may be susceptible to a ceiling effect, potentially limiting its sensitivity to capture more subtle but clinically relevant improvements in neurological recovery. Accordingly, future studies evaluating anticoagulation timing may benefit from incorporating stricter functional thresholds, such as excellent functional outcome (mRS 0-1) or ordinal shift analyses across the full mRS distribution. Nevertheless, given the observational nature of the present analysis, residual confounding by unmeasured factors cannot be entirely excluded, and causal inference should therefore be interpreted with caution.

In our cohort, the risks of recurrent IS, all-cause mortality, and composite outcomes were comparable between early and delayed initiation groups. The observed incidence of CRMB at 90 days was numerically higher in the early initiation group (5.7% vs 4.5%), although this difference was not statistically significant. Importantly, no signal of excess major hemorrhagic complications was observed. Taken together, these findings align with accumulating evidence indicating that initiating NOACs within the early post-stroke period does not lead to a net increase in adverse clinical events. Using the same 7-day threshold, a prospective observational cohort of 155 patients reported no intracerebral hemorrhage (ICH) events among patients treated with NOACs and no significant difference in recurrent IS between early (≤7 days) and delayed (>7 days) initiation groups. ³⁸ Comparable results have been reported across larger pooled observational datasets. A pooled analysis of eight European and Japanese cohorts, including 2550 patients with AIS related to NVAF, found that early (≤5 days) versus delayed (>5 days) NOACs initiation was not associated with risk of recurrent IS (aHR 1.2, 95% CI 0.5-2.9, p = .69) or ICH (aHR 6.0, 95% CI 0.6-56.3, p = .12). Notably, the absolute risk of recurrent IS was approximately sevenfold higher than that of ICH, further supporting early NOACs use, highlighting the asymmetric balance of ischemic versus hemorrhagic risk during the early post-stroke period. ³⁹ Evidence from even earlier initiation thresholds further reinforces the safety of early anticoagulation. The SAMURAI study (n = 499) with NVAF after AIS or transient ischemic attack (TIA) found no significant differences in stroke/systemic embolism, major bleeding, or death between early (≤3 days) and late (≥4 days) initiation groups, at both 3 months and 2 years. ⁴⁰ Similarly, a pooled analysis of registry data from 1289 patients found no significant difference in 90 days reduced ischemic and hemorrhagic outcomes among patients starting anticoagulation at 0–3 days, 4–14 days, or >14 days. ⁴¹ Most recently, high-level evidence from the CATALYST individual patient data meta-analysis has further strengthened this evidence base. Pooling data from four RCTs including 5441 patients with AIS and NVAF, CATALYST demonstrated that NOACs initiation within 4 days was associated with a lower 30-day risk of the composite outcomes with later initiation (2.1% vs 3.0%; OR 0.70, 95% CI 0.50-0.98, p = .039), as well as a reduced risk of recurrent IS alone was also lower (1.7% vs 2.6%; OR 0.66, 95% CI 0.45-0.96, p = .029), without an increase in symptomatic ICH (0.4% vs 0.4%; OR 1.02, 95% CI 0.43-2.46; p = .96). ⁴² Together, these findings support the efficacy and safety of NOACs initiation within the first days following AIS.

In exploratory subgroup analyses, early rivaroxaban initiation was associated with a higher likelihood of achieving an excellent functional outcome (mRS 0-1) at 90 days among patients with more severe strokes (NIHSS >7); however, the formal test for interaction was not statistically significant. At 12 months, the effect on good functional outcome (mRS 0-2) appeared to vary by baseline stroke severity, with a borderline interaction. Specifically, a numerically favorable trend was observed in patients with NIHSS >7, whereas no clear association was evident among those with milder strokes (NIHSS ≤7). These findings should be interpreted as exploratory. One potential explanation is that patients with more severe strokes, often reflecting a larger infarct burden or major vessel occlusion, are at a higher baseline risk of early recurrent ischemic events, in whom earlier anticoagulation could confer a greater benefit. In support of this hypothesis, a large Chinese cohort study has shown that patients with persistently moderate-to severe NIHSS scores experience significantly higher risks of recurrent stroke and major cardiovascular events compared with those with mild NIHSS scores. ⁴³ This direction of this exploratory signal is also consistent with emerging evidence suggesting that early initiation of NOACs after AIS may be feasible even in patients with more severe strokes, without an excess risk of hemorrhagic complications. ⁴⁴ However, RCTs that specifically evaluated stroke severity using NIHSS-based subgroup analyses, including OPTIMAS ¹⁸ and TIMING, ¹⁶ did not demonstrate clear heterogeneity of treatment effect across subgroups defined by baseline stroke severity. Although directionally favorable signals were observed in certain NIHSS-defined subgroups at specific time points in ELAN, ⁴⁵ these effects were not uniform across follow-up periods and did not establish stroke severity as a robust effect modifier. Therefore, the subgroup trends observed in the present study should be interpreted with caution and require confirmation in adequately powered prospective studies.

The robustness of our primary findings was supported by multiple sensitivity analyses. In the complete-case analysis, the IPTW-adjusted results remained largely directionally consistent after excluding patients with missing data. Similarly, after excluding patients with hemorrhagic transformation, early rivaroxaban initiation continued to show a significant association with excellent functional outcome (mRS 0-1) at both 90 days and 12 months. Together, these findings suggest that the observed associations are unlikely to be driven solely by missing data or the inclusion of patients with hemorrhagic transformation. Nevertheless, as with all observational studies, residual confounding cannot be entirely excluded.

In addition, early rivaroxaban initiation was consistently associated with a shorter hospital stay, which may indicate more rapid clinical stabilization and earlier discharge readiness in patients receiving timely anticoagulation. In the dose-stratified analyses, early rivaroxaban initiation showed a generally consistent direction of association with improved functional outcomes across most dose categories. However, these associations did not uniformly reach statistical significance within individual dose strata. This pattern is not unexpected in a real-world setting, where rivaroxaban dose selection is not randomly assigned but is strongly influenced by baseline clinical risk profiles, including age, renal function, bleeding risk, and neuroimaging features. Consequently, stratification by dose results in smaller subgroup sizes and greater baseline heterogeneity, which limits statistical power and precludes definitive conclusions regarding dose-specific effects.

Several limitations should be acknowledged. First, the retrospective, single-center design may introduce selection and information bias. Although IPTW adjustment and sensitivity analyses were performed, residual confounding cannot be fully eliminated, and causal inference should be interpreted with caution. Second, the limited sample size, low rates of key clinical events, and heterogeneity in rivaroxaban dosing may have reduced statistical power to detect modest differences; thus, the findings should be considered exploratory rather than confirmatory, and non-significant findings for secondary outcomes should be interpreted with caution. Third, the observed trend toward interaction with baseline NIHSS in subgroup analyses is hypothesis-generating due to multiple testing and should be validated in larger, prospective cohorts.

Conclusion

Early initiation of rivaroxaban was independently associated with a higher likelihood of excellent functional outcome at 90 days and 12 months in patients with AIS and NVAF after IPTW adjustment. No significant associations were observed for good functional outcome, recurrent IS, CRMB, or all-cause mortality. These findings suggest that early initiation of rivaroxaban may be feasible in routine clinical practice and highlight the need for prospective studies to define optimal, individualized anticoagulation timing.

Abbreviations

AF

atrial fibrillation

APTT

activated partial thromboplastin time

aHR

adjusted hazard ratio

aOR

adjusted odds ratio

AIS

acute ischemic stroke

CHA2DS2-VASc-60

congestive heart failure, hypertension, age 65 years or older, diabetes mellitus, previous stroke/transient ischemic attack, vascular disease, age 60 to 65 years, female

CI

confidence interval

Co-infection

concurrent infection

CRMB

clinically relevant major bleeding

CRP

C-reactive protein

CT

computed tomography

eGFR

estimated Glomerular Filtration Rate

ECG

electrocardiogram

FBG

fasting blood glucose

FIB

fibrinogen

HAS-BLED

hypertension, abnormal renal or liver function, stroke, bleeding history, labile international normalized ratio, age 65 years or older, and drugs that increase bleeding tendency or alcohol use

HB

hemoglobin

HR

hazard ratio

ICH

intracerebral hemorrhage

IPTW

inverse probability of treatment weighting

IQR

interquartile range

IS

ischemic stroke

ISTH

International Society on Thrombosis and Hemostasis

LDL-C

low-density lipoprotein cholesterol

LMWH

low-molecular-weight heparin

LOS

length of stay

LYM

lymphocyte count

MCAR

Missing Completely at Random

MICE

multiple imputation by chained equations

MRI

magnetic resonance imaging

mRS

modified Rankin Scale

NIHSS

National Institute of Health Stroke Scale

NVAF

non-valvular atrial fibrillation

NOACs

non-vitamin K antagonist oral anticoagulants

OAC

oral anticoagulant

OR

odds ratio

PLT

platelet count

PT

prothrombin time

RCS

restricted cubic spline

RCTs

randomized controlled trials

SD

standard deviation

SMD

standardized mean differences

TG

triglyceride

TIA

transient ischemic attack

VIF

variance inflation factor

WBC

white blood cell