Basilar artery occlusion (BAO), a severe form of ischemic stroke, significantly contributes to the overall disease burden and mortality [1]. The 90-day modified Rankin Scale (mRS) score is the gold standard tool for assessing functional outcomes in BAO; however, its prolonged assessment period limits its clinical utility [2]. Accurately predicting a patient’s recovery trajectory is important. The concept of a “golden plastic period” following a stroke—during which time judicious and targeted interventions, including rehabilitation, can substantially improve disability status [3]— further underscores the critical need for precise early assessment tools.

The National Institutes of Health Stroke Scale (NIHSS)—an accessible, standardized tool for evaluating neurological deficits— is widely used in early stroke assessment [4,5]. However, research on the prognostic value of NIHSS scores in patients with BAO remains limited, impeding optimized prognostic evaluation. Furthermore, different treatment strategies—endovascular treatment (EVT) versus best medical management (BMM)—may impact the predictive performance and optimal cut-off values of NIHSS scores. Identifying the optimal timing for NIHSS assessment, determining treatment-specific cut-off values, and elucidating the role of early neurological function in mediating treatment effects are critical for guiding rehabilitation strategies, enhancing patient stratification, and improving prognostic assessment.

Herein, data from the Acute Basilar Artery Occlusion: Endovascular Thrombectomy versus Standard Medical Treatment (ATTENTION) trial [6] were used to systematically evaluate and compare the predictive performance of baseline, 24-hour, and 72-hour NIHSS scores for 90-day functional outcomes in patients with BAO. The objectives were (1) to determine the optimal NIHSS assessment time point for predicting long-term prognosis in BAO and (2) to explore and validate treatment-specific (EVT vs. BMM) optimized NIHSS cut-off values. We aimed to provide evidence to support early risk stratification and guide clinical post-treatment communication in patients with BAO.

In this study, based on ATTENTION trial data, we identified the 72-hour NIHSS score as a reliable and superior early predictor of long-term outcomes in patients with BAO. Our findings show that neurological status at 72 hours more accurately predicts 90-day functional outcomes compared with that at earlier time points and better accounts for the clinical benefits of EVT. Furthermore, we established for the first time treatment-specific 72-hour NIHSS score thresholds with high predictive value for EVT and BMM. These findings provide novel evidence-based guidance for early prognostic communication and risk stratification of patients with BAO.

Compared with the baseline score, the 24-hour NIHSS score is a stronger predictor of long-term functional prognosis [4,12-14]. High-quality evidence from the Basilar Artery Occlusion Endovascular Intervention versus Standard Medical Treatment trial revealed direct comparisons of the predictive performance of percent change in NIHSS, 24-hour NIHSS, and ΔNIHSS. The absolute 24-hour NIHSS score was the strongest predictor [13]. In this study, we confirmed that the 24- and 72-hour NIHSS scores effectively predicted long-term functional outcomes (mRS score of 0-3), with the 72-hour assessment exhibiting superior predictive accuracy. This is consistent with later assessments that reached a higher agreement [15].

The advantage of the 72-hour NIHSS score was not coincidental. The evolution of cerebral edema is a time-dependent process, with its peak typically occurring 3-5 days post-onset [16]; therefore, an assessment performed at 24 hours may underestimate its impact, as the peak has not yet been reached. The full clinical impact of periprocedural complications (such as symptomatic intracranial hemorrhage) has a delayed manifestation. Perihematomal edema secondary to hemorrhage continues to develop after the initial 24 hours, with the average thickness of the edema reaching only 60% of its peak at 24 hours [17,18]. Neu-rological deficits can also be more accurately assessed after the effects of anesthesia (particularly general anesthesia) have stabilized. Reperfusion dynamics may also play crucial roles in this process. The “no-reflow” phenomenon, which is the failure of microcirculatory-level perfusion despite successful largevessel recanalization, can persist within the first 24 hours after vessel recanalization [17] and results in progressive, delayed tissue death [19,20].

The magnitude of the mediation effect observed at 72 hours is consistent with findings from existing literature. Prior studies reported that the 24-hour NIHSS score mediated between 54% and 68% of the treatment effect of EVT [8,9]. Kniep et al. [9] found that the mediation effect increased to 75% when the NIHSS score at hospital discharge was used as the mediator. Our finding that the 72-hour NIHSS score mediates 62.1% of the treatment effect aligns with this temporal trend. This finding suggests that the 72-hour assessment captures a larger proportion of the ultimate treatment benefit than the 24-hour assessment, while also being a more clinically timely and practical endpoint than discharge assessments.

For patients receiving EVT, a 72-hour NIHSS score of ≤11 was identified as a robust prognostic cut-off, based on its consistent optimality across multiple metrics. Notably, this threshold is higher than the 24-hour NIHSS cut-off score of ≤9 proposed in a previous key study [21]. This difference is likely attributable to the more severe patient population in our study. Crucially, this conclusion is supported by the subgroup analysis from Kniep et al. [21], which showed that the sensitivity of their ≤9 threshold dropped significantly in patients with a baseline NIHSS score ≥20. In contrast, our proposed threshold of ≤11 represents a prognostic marker with high clinical applicability for this more severe patient population. It is the most reliable prognostic marker for predicting favorable long-term functional outcomes in patients with moderate-to-severe BAO.

The prognostic cutoff for the 72-hour NIHSS score in the BMM subgroup was determined. The conventional method of maximizing Youden’s index produces a high risk of “false optimism,” reaching clinically intolerable levels. Such false positive predictions pose significant risks. This could hinder families from preparing for long-term disability and may even lead clinicians to incorrectly withhold effective treatment [22]. Furthermore, as shown in Supplementary Figure 7A, when predicting a favorable outcome (mRS score of 0-3), multiple NIHSS score thresholds have comparable Youden’s indices, indicating that relying solely on Youden’s index lacks clinical utility. We tested three strategies for the sensitivity analysis. Notably, all three analytical paths, each satisfying different priorities, consistently indicated the same optimal threshold. This finding greatly enhances the robustness of this threshold and confirms a core principle: in BMM cohorts with a low prior probability of favorable outcomes, prioritizing predictive reliability over maximizing Youden’s index is paramount to ensuring the model’s clinical applicability and supporting responsible physician-patient communication. When evaluating the strictest outcome (mRS 0-1), determining an ideal threshold that achieves a high PPV while balancing sensitivity and specificity is extremely challenging. As a rarer outcome, mRS 0-1 inherently has a lower PPV than mRS 0-3, significantly increasing the risk of the model making “false optimistic” predictions.

Delayed recovery from “sleeping neurons” may occur beyond 72 hours [23], potentially causing early NIHSS scores to overestimate post-EVT injury. Our exploratory analysis of DNI indicated that age was a key independent predictor in patients receiving EVT. Therefore, for younger patients, the prognosis should not be determined prematurely based solely on severe early neurological deficits, as they still have considerable potential for achieving favorable functional recovery at a later stage. These findings are consistent with those of previous research [16], possibly attributable to the role of the cerebral reserve [24], with younger patients’ brains exhibiting greater postoperative plasticity and tolerance to injury [25].

This study had certain limitations. First, the 72-hour NIHSS score is only a snapshot of the dynamic disease course, and its predictive value depends on subsequent conditions. Second, our clinically driven strategy for the BMM subgroup, in which a high PPV was prioritized, resulted in a trade-off with lower sensitivity. Third, our data showed the acceptability of this trade-off for the primary outcome (the sensitivity of a 72-hour NIHSS cut-off score of ≤9 remains above 80%). Fourth, our analysis did not extend beyond 72 hours. The NIHSS score after 72 hours may hold a stronger value, and subsequent observation times should be explored in future studies. Fifth, the NIHSS may not fully capture symptoms such as ataxia or dysphagia [26]. Nonetheless, its use remains appropriate, universal, and comparable in patients with BAO within the context of current practices [21]. Sixth, we were unable to adjust for post-treatment confounders between 24 and 72 hours, such as sedation weaning, early complications (e.g., pneumonia), or decompressive surgery. These events may represent significant confounders that could influence the mediating effect of NIHSS. Seventh, although the sensitivity analysis of the imputed data showed robust results, excluding patients with missing data was a potential source of bias that may limit the generalizability of our findings. Finally, our study population was restricted to patients with moderate-to-severe BAO (baseline NIHSS score ≥10). Therefore, our proposed thresholds cannot be generalized to patients with mild stroke presentations.

The 72-hour NIHSS score outperformed the ΔNIHSS score from baseline to 72 hours, the 24-hour NIHSS score, and the ΔNIHSS score from baseline to 24 hours in predicting 90-day functional outcomes. The 72-hour NIHSS score-based approach proposed in this study provides a valuable reference for clinical prognostic stratification.