Cerebral venous sinus thrombosis (CVST) is a distinct cerebrovascular disorder characterized by thrombotic occlusion of the cerebral veins and/or dural sinuses, accounting for approximately 0.5–1% of all strokes (Alghamdi et al., 2022; Nguyen et al., 2024). Unlike arterial strokes, CVST predominantly affects younger adults and presents with a highly variable clinical spectrum, ranging from isolated headache to severe neurological deficits, coma, and death (Javed et al., 2018; Alet et al., 2020). Over the past decades, advancements in diagnostic imaging and the widespread application of anticoagulant therapy have significantly reduced the acute-phase mortality of CVST. Consequently, research and clinical focus have progressively shifted towards understanding the long-term functional outcomes and quality of life of survivors.

A critical challenge in the long-term management of CVST is the profound heterogeneity in neurological recovery. While a majority of patients achieve functional independence [modified Rankin Scale (mRS) score of 0–2], a substantial proportion experiences persistent cognitive impairment, fatigue, or residual disability, profoundly impacting their personal and professional lives (Saroja et al., 2020; Kajbaf et al., 2021). The underlying mechanisms driving these divergent recovery trajectories remain poorly elucidated. Current prognostic models primarily rely on baseline clinical and radiological factors, such as coma, intracranial hemorrhage, or malignancy, to predict a dichotomized outcome at a single time point (Ferro and de Aguiar Sousa, 2019; Pizzi et al., 2016). However, this static, group-averaged approach fails to capture the dynamic and multidimensional nature of post-CVST neurological evolution. It obscures the possibility that distinct subpopulations of patients may follow characteristically different functional trajectories over time, each governed by a unique set of biological and pathological mechanisms.

To unravel this complexity, a methodological shift from variable-centered to person-centered analytical approaches is required. Latent class mixed models (LCMM) represent a powerful statistical framework that addresses this need (Hoekstra and Twisk, 2015; Leiby et al., 2014; Zhang and Ip, 2014). Unlike conventional mixed models that estimate an average trajectory for the entire population, LCMM can identify unobserved (latent) subpopulations of individuals who share similar patterns of change in a longitudinal outcome (Pei et al., 2024; Badiya et al., 2020), such as the mRS. This approach has proven invaluable in characterizing recovery heterogeneity in other neurological conditions, such as traumatic brain injury and ischemic stroke (Wickwire et al., 2022; Gardner et al., 2019), but its application to CVST remains largely unexplored.

Elucidating these distinct trajectories is not an endpoint in itself; it is a crucial first step toward a “precision prognosis” paradigm. The subsequent, and equally important, step is to uncover the mechanisms—the specific clinical and neuroimaging factors—that predispose patients to a particular recovery path. Key postulated mechanisms include the brain’s capacity for compensation, reflected in collateral circulation status, the extent of initial parenchymal injury (e.g., hemorrhage, infarction), and the systemic thrombotic burden, measured by biomarkers like D-dimer (Domitrz et al., 2020; Sun et al., 2023). An integrative model that links trajectory subgroups with their underlying mechanisms is urgently needed to move beyond descriptive prognosis and towards predictive, actionable insights.

Therefore, this longitudinal study aims to: (1) identify and characterize distinct long-term (≥1 year) neurological functional trajectories in a cohort of CVST patients using LCMM; and (2) construct an integrated dynamic prognostic model by investigating the associations between these trajectory subgroups and a comprehensive set of baseline clinical, radiological, and laboratory factors. We hypothesize that at least two discrete mRS trajectory classes exist—one favoring favorable recovery and another associated with poor outcome—and that these classes are differentially predicted by a combination of age, collateral circulation status, and patterns of brain parenchymal injury.

This longitudinal study provides novel evidence for the existence of distinct long-term functional trajectories in patients with cerebral venous sinus thrombosis, thereby advancing our understanding of the heterogeneous nature of neurological recovery in this condition. By applying latent class mixed modeling (LCMM) to repeated mRS assessments over an extended period, we identified two discrete recovery patterns: a “Favorable-Recovery” trajectory (class 2) and a “Poor-Recovery” trajectory (class 1). Crucially, we further demonstrated that these trajectories are not random but are robustly predicted by a combination of age and the integrity of the venous collateral circulation, offering key insights into the underlying mechanisms.

Our most salient finding is the identification of two latent classes with divergent prognostic paths. The “Poor-Recovery” class, comprising over half of our cohort, was characterized by an older age at onset, a higher burden of initial neurological deficits and parenchymal injury, and a sluggish recovery course. In contrast, the “Favorable-Recovery” class was typified by younger patients with less severe initial imaging findings and a rapid, near-complete functional restoration. This delineation moves beyond traditional, static outcome measures and captures the dynamic process of recovery, aligning with the growing recognition of outcome heterogeneity in cerebrovascular diseases (Phan et al., 2017). The excellent classification quality of our model underscores that these are not arbitrary groupings but represent biologically and clinically distinct patient phenotypes.

The mechanistic insights from our study are twofold, centering on the two independent predictors identified in the multivariable model. First, increasing age emerged as a powerful determinant of an unfavorable trajectory. This finding is consistent with the broader stroke literature, where advanced age is consistently associated with worse outcomes, potentially due to reduced neuronal plasticity, increased comorbidities, and lower brain reserve capacity (Costigan et al., 2009; Wang et al., 2022; Hu et al., 2024; Anatürk et al., 2021). In the specific context of CVST, an older brain may possess a diminished ability to compensate for the acute insult of venous hypertension and ischemia, leading to more extensive injury and a hampered recovery process.

Second, and perhaps more specific to CVST pathophysiology, we found that a lower venous collateral circulation score was independently associated with the “Poor-Recovery” trajectory. The cerebral venous system possesses a rich collateral network, and its efficacy is critical in mitigating the hemodynamic consequences of sinus occlusion. A robust collateral circulation (score of 2) likely facilitates the redistribution of venous outflow, thereby limiting venous congestion, intracranial pressure elevation, and subsequent parenchymal damage (hemorrhage/infarction). Our results provide empirical support for this, showing that a robust collateral circulation (score of 2) was far more prevalent in the “Favorable-Recovery” class (43%) compared to the “Poor-Recovery” class (8%, p < 0.001). The strong independent effect of collateral status after adjusting for age and parenchymal injury suggests that it is a primary mechanism influencing the recovery trajectory, rather than merely a secondary phenomenon.

It is noteworthy that while univariate analysis showed significant associations for focal deficits, brain parenchymal changes, and D-dimer levels with class membership, these factors were not independent in the multivariable model. This indicates that their influence is likely mediated through or confounded by the core mechanisms of age and collateral capacity. For instance, the extent of parenchymal injury may be a downstream consequence of poor collaterals, and its prognostic information is thus captured by the collateral score itself.

Our findings must be contextualized within the landscape of existing prognostic tools for CVST. Clinical prediction scores, such as those by Ferro et al. (2004), Bushnaq et al. (2018), and more recently Li et al. (2023), have been instrumental in stratifying patients based on baseline clinical and radiological factors. Notably, age has emerged as a consistent and strong predictor across these models, a finding robustly corroborated by our LCMM analysis. However, a critical examination reveals that these established scores predominantly incorporate factors such as coma, intracranial hemorrhage, malignancy, or infection, which often represent the consequences of the thrombotic event or severe systemic states. In contrast, our model highlights venous collateral circulation—a factor reflecting the brain’s intrinsic compensatory capacity at the onset of venous occlusion. While thrombus burden scores, exemplified by the work of Wang et al. (2023), provide a valuable quantification of the extent of sinus involvement, they primarily describe the anatomical extent of the insult. Our results suggest that the functional status of venous collaterals may act as a key moderator between thrombus burden and parenchymal injury, thereby influencing the ultimate recovery trajectory. The fact that collateral status remained an independent predictor after adjusting for age and parenchymal changes underscores its role as a primary, rather than secondary, pathophysiological determinant. Therefore, our study extends the existing paradigm by proposing that integrating a measure of physiological reserve (collateral circulation) with a core demographic factor (age) may offer a more mechanism-oriented framework for “precision prognosis,” complementing the established clinical-anatomical models.

Several scales exist to grade venous collateral circulation. We selected the Sheth score for this analysis based on three principal considerations (Sheth and Liebeskind, 2014; Sheth et al., 2018). First, its simplicity and reproducibility—a straightforward three-tier ordinal scale (0–2)—make it highly feasible for consistent application in routine clinical radiology practice, a key factor for potential future translation. Second, it has established validity and prognostic correlation specifically within the CVST population, directly linking collateral status to clinical outcomes (Sheth et al., 2018). Third, compared to more complex quantitative measures, this scale efficiently captures the functional capacity of the venous collateral network (absent, present but non-connected, present and connected), which aligns with our aim to assess a patient’s intrinsic physiological compensatory reserve at admission.

In conclusion, this study demonstrates that CVST patients follow one of two distinct long-term functional trajectories, which are predominantly determined by the interplay between patient-specific reserve (age) and disease-specific compensatory capacity (venous collateral circulation). This “precision prognosis” framework enhances our ability to predict recovery patterns and provides a mechanistic basis for future research aimed at improving outcomes in this patient population.