Warfarin Exposure, Not Standard Monitoring, Tracks Brain Small Vessel Damage in Elderly Stroke Patients

For millions of older adults living with atrial fibrillation, warfarin has long been the workhorse of stroke prevention. Yet a new retrospective cohort study from Xuzhou, China, suggests that the standard yardstick clinicians use to judge whether the drug is doing its job may be missing a silent, accumulating form of brain injury. In a cohort of 300 patients aged 65 and older with ischemic stroke and non-valvular atrial fibrillation, researchers found that model-estimated warfarin exposure in the bloodstream was significantly associated with the burden of cerebral small vessel disease, while the conventional time-in-therapeutic-range metric showed no association at all.

The study, published in Frontiers in Neuroscience, applied population pharmacokinetic-pharmacodynamic (PopPK/PD) modeling to estimate each patient’s actual steady-state drug concentration rather than relying solely on international normalized ratio (INR) readings. The central finding was striking: each 100 ng/mL increase in steady-state average warfarin concentration raised the odds of high cerebral small vessel disease burden by roughly 32 percent (odds ratio 1.32, p < 0.001). By contrast, time in therapeutic range, the metric recommended by guidelines worldwide, was not significantly associated with that burden (p = 0.444).

Cerebral small vessel disease refers to pathological changes in the brain’s smallest blood vessels, visible on magnetic resonance imaging as white matter hyperintensities, lacunar infarcts, and cerebral microbleeds. The researchers graded white matter lesions with the modified Fazekas scale, counted microbleeds on susceptibility-weighted imaging, and combined these markers into a total burden score from 0 to 4, defining a score of 3 or higher as high burden. Of the 300 patients enrolled, 194 fell into the high-burden group and 106 into the low-burden group.

The clinical stakes are considerable. Cerebral small vessel disease is a major contributor to vascular cognitive impairment and dementia, with prior research suggesting that roughly 45 percent of dementia cases are attributable to or closely linked with the condition. In this cohort, high small vessel disease burden more than doubled the risk of cognitive impairment, defined as a Montreal Cognitive Assessment score below 26 at six months (Cox hazard ratio 2.04, p = 0.007). The incidence of cognitive impairment also climbed stepwise across tertiles of estimated warfarin exposure, rising from 42.2 percent in the lowest exposure group to 54.0 percent in the middle group and 75.4 percent in the highest (p < 0.001).

To estimate individual exposure, the team built a one-compartment model with first-order absorption for S-warfarin using nonlinear mixed-effects modeling software. The model incorporated covariates known to shape warfarin handling in the body: age, body weight, estimated glomerular filtration rate, serum albumin, carriage of CYP2C9 loss-of-function alleles, and the VKORC1-1639AA genotype. Advanced age, reduced kidney function, low albumin, and these genetic variants all significantly reduced apparent clearance and thereby increased systemic exposure. The typical population clearance was estimated at 0.263 L/h with inter-individual variability of 26.3 percent, and the model performed well in validation: 998 of 1,000 bootstrap replicates converged, parameter bias stayed below 15 percent, and the coefficient of determination between predicted and observed clearance was 0.703.

From this model, the researchers derived three exposure metrics for each patient: steady-state average concentration (Cavgss), steady-state peak concentration (Cmaxss), and degree of fluctuation (DF). When tested for their ability to discriminate patients with high small vessel disease burden, these model-derived parameters formed a clear hierarchy above the traditional measures. Cmaxss achieved an area under the receiver operating characteristic curve of 0.695, Cavgss 0.679, and DF 0.651. INR standard deviation managed only 0.557, and Rosendaal-calculated time in therapeutic range performed at 0.493, essentially no better than chance. DeLong tests confirmed that Cavgss significantly outperformed both time in therapeutic range (difference in AUC 0.186, p < 0.001) and INR variability (difference in AUC 0.122, p = 0.011).

The conceptual gap between the two families of metrics is pharmacologically meaningful. Time in therapeutic range quantifies the proportion of time INR values fall within the target window, and INR variability captures how much those values swing between measurements. But INR is a clotting-time surrogate, not a direct measure of drug concentration, and it cannot fully capture inter-individual variation in actual systemic exposure. Two patients with identical time-in-therapeutic-range values may carry very different steady-state warfarin concentrations because of their genes, age, weight, and kidney function. The baseline data bore this out: time in therapeutic range was statistically indistinguishable between the low- and high-burden groups (72.48 versus 71.22 percent, p > 0.05), while every model-derived exposure parameter differed significantly between groups (all p < 0.05).

That means the model-based approach could flag vulnerable patients whose anticoagulation looked adequate by conventional standards but whose brains were carrying a heavier microvascular load. Notably, the authors point out that even among patients whose anticoagulation quality appeared acceptable on traditional metrics, discriminative ability for identifying high burden remained modest for all markers, so the exposure parameters are best viewed as complementary tools rather than replacements. The study also found a distinct role for INR variability: patients with an INR standard deviation above 0.5 experienced significantly more composite endpoints, a combination of intracranial hemorrhage, major bleeding, recurrent stroke or transient ischemic attack, and all-cause death (odds ratio 2.95, p = 0.009). INR variability, in other words, appears to signal unstable anticoagulation and downstream clinical events, while steady-state exposure levels better reflect cumulative cerebral microvascular injury.

From a mechanistic standpoint, the authors frame the findings around the distinction between exposure magnitude and exposure stability. The pathological cascade driving small vessel disease-related cognitive decline involves blood-brain barrier disruption, endothelial dysfunction, chronic cerebral hypoperfusion, and neuroinflammation. Stable, appropriately dosed anticoagulant exposure within the therapeutic range may be correlated with alleviated cerebral microvascular injury and barrier leakage, whereas excessively high systemic warfarin exposure, reflected by elevated average steady-state concentration in this dataset, was associated with greater imaging burden and higher rates of subsequent cognitive impairment. The authors are careful to describe these as associations observed within a single cohort rather than proof of causation, and they deliberately avoid labeling the pharmacokinetic parameters as anticoagulation quality indicators, since they measure a biologically distinct quantity.

The study carries important limitations that temper enthusiasm. It was a single-center retrospective analysis of 300 patients treated at Xuzhou First People’s Hospital between January and December 2025, with cognitive impairment assessed using a uniform MoCA cutoff of 26 without formal education-adjusted thresholds, though years of education were adjusted as a covariate in the regression and Cox models. Age, CHA2DS2-VASc score, and baseline cognitive function differed between burden groups and were adjusted for in multivariable analyses, but residual confounding cannot be excluded in a retrospective design. The authors call for large-sample, prospective, multicenter studies to validate whether PopPK/PD-derived exposure parameters can be practically deployed in routine anticoagulation management. Even so, the work offers a pharmacological foundation for a shift in thinking: instead of asking only how long a patient’s INR stayed in range, clinicians may eventually be able to ask how much drug that patient’s brain is actually exposed to, and use that answer to protect both vessels and cognition in one of medicine’s most vulnerable populations.

Subject of Research: Associations between warfarin pharmacokinetic exposure, cerebral small vessel disease burden, and cognitive impairment in elderly patients with ischemic stroke and atrial fibrillation

Article Title: Warfarin anticoagulation control and risks of cerebral small vessel disease and cognitive impairment among elderly patients with ischemic stroke and atrial fibrillation: a retrospective cohort study

Article References: Zhao, D., He, W., Shen, Z., Tan, C., & Zhang, M. (2026). Warfarin anticoagulation control and risks of cerebral small vessel disease and cognitive impairment among elderly patients with ischemic stroke and atrial fibrillation: a retrospective cohort study. Frontiers in Neuroscience, 20, 2015432. https://doi.org/10.3389/fnins.2026.2015432

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.2015432

Keywords: warfarin, atrial fibrillation, ischemic stroke, cerebral small vessel disease, cognitive impairment, pharmacokinetics, population pharmacokinetic-pharmacodynamic modeling, time in therapeutic range, INR variability, elderly patients, white matter hyperintensities, vascular dementia