Subarachnoid hemorrhage (SAH) remains a severe condition associated with high mortality and unfavorable outcomes. The optimal treatment strategy (coiling vs. clipping) varies based on multiple factors: age, comorbidities, posterior aneurysm location, high mFisher grade, the presence of intracerebral hemorrhage, and worse scores on clinical scales (GCS, H&H, WFNS) play an important role according to the latest guidelines [12]. Additionally, the combination of multiple factors, as per the SAFIRE score, can show an accordance with the treatment strategy as emerged from the current results. Indeed, age (p < 0.01), aneurysm size (p < 0.01), mFisher grade (p < 0.01) and WFNS state after resuscitation (p < 0.01) differed significantly between treatment types. These factors can influence the decisional process, firstly, judging between treatment and no treatment and secondly, as elements of discussion for different teams arguing in favor of clipping or coiling.

In line with previous studies [2, 13, 14], the in-hospital mortality rate during the acute phase was 31%, increasing to 38% for patients admitted to ICU. To address this risk, several scoring systems have been developed to predict the severity of the condition, in-hospital mortality [11, 15, 16], and unfavorable long-term outcomes [9, 10, 14, 15].

The current analysis confirms the good performance of the HAIR score in predicting in-hospital death with an AUC of 0.812, consistent with the original study and previous external validation [11, 17]. Although in the latter an overestimation problem was posed, in the studied cohort a higher rate of 30-day mortality has been recognized for lower scores (i.e., 5.8, 14.5 and 17.5 among the first three classes, respectively), facing the opposite problem. However, as the HAIR score was designed to predict in-hospital mortality, its ability to predict long-term outcomes was limited.

Similarly, the web-based AI tool, Nutshell, which uses neural networks, did not demonstrate adequate discrimination for short- and long-term outcome prediction. While it performed moderately well for 30-day mortality prediction (AUC: 0.749), it was inferior to models based on multivariate logistic regression, such as HAIR score. Nonetheless, the tool showed improvement over previous external validations, which reported an AUC of 0.636.

Approximately 70% of patients experienced complications during hospitalization. The most common neurological complications were vasospasm (clinically and radiologically confirmed), chronic hydrocephalus, and delayed cerebral ischemia (DCI), with frequencies of 20.4%, 19.7%, and 15.0%, respectively. These rates are lower than those reported in the literature for vasospasm and DCI, typically 33% and 20–40% [18, 19], while the appearance of chronic hydrocephalus was in line with previous reports [20]. A statistical significance was observed between the onset of vasospasm and the BNI grade [21].

Although an early treatment (< 24 h) by securing the aneurysm has been associated with reduced mortality and better neurological outcome [12, 22], the subsequent clinical setting can change according to the neurological status and the place where the patient is treated, often depending on bed availability.

Indeed, only 25–30% of patients in “good” neurological status are admitted to non-intensive but highly specialized wards such as Stroke Unit or non-intensive neurosurgical units, decreasing for the former the possibility of offering specialized care for more frequently encountered patients (i.e., ischemic stroke) and for the latter the possibility to offer elective surgery. The same problem becomes exponentially big in the setting of the Intensive Care Unit (ICU) where the cures are highly demanding [23]. On the one hand, admission in a specialized Intensive Care Unit has been demonstrated to offer an advantage on the outcome, by recognizing both neurological and systemic complications that can affect aSAH patients, but, on the other hand, ICU hospitalization may result in additional complications such as ventilator-associated pneumonia, systemic infection or acute kidney injury, which increase the duration and the cost of hospitalization. These factors have been related to a higher risk of delayed medical complications (even in patients with low-grade aSAH), vasospasm, mortality, and 30-day readmissions [24]. Additionally, the duration of ICU stay and mechanical ventilation have been identified as significant contributors to mortality [25].

Similarly to other reports [26], 74.1% of patients were admitted to the ICU, with a mean length of stay of 17.4±19.6 days and a mortality rate of 38.4%. These patients passing through the ICU had a significant (p < 0.001) worse outcome (3.9±2.18 for mRS and 2.54±1.54 for GOS) than those who were admitted directly in the ward (1.07±1.48 for mRS and 4.51±0.95 for GOS). Six months has been recognized as the critical period for significant neurological recovery, after which stabilization typically occurs [27]. Indeed, the current population passed from an aggregate mRS of 3.16±2.37 and GOS of 3.05±1.65 at discharge (considering the entire study population) to 1.64±2.17 (mRS) and 4.11±1.30 (GOS) in those with a minimum follow-up of at least 6 months.

Evaluating the performance of the SAFIRE and HATCH scores, both demonstrated excellent discriminatory ability for unfavorable outcomes, with AUCs of 0.866 and 0.886 at discharge and 0.825 and 0.83 at one year, respectively. These results confirm the reliability of these scores, with SAFIRE showing improved AUC performance compared to previous validations (AUC: 0.64–0.75) [8, 28, 29].

Among patients with a SAFIRE score > 15 (Group 5) who were treated, an in-hospital mortality rate of 63% was observed, rising to 89.5% if extended to the entire follow-up period (7.2±5.89 months). Overall, a mortality rate of 90% and a poor-outcome rate of 100% were observed (mRS 4 and 5 were reported for the two patients who survived). This supports the evidence that while early mortality may not always be exceedingly high, the risk of poor outcomes increases exponentially in the medium to long term for poor-grade SAH [30]. In this context, prognostic scores such as SAFIRE and HATCH could be particularly valuable.

Indeed, these scores can help identify patients who would require more intensive monitoring despite a relatively stable clinical condition, as well as those who, due to the severity of the SAH, would not benefit from aggressive management in the short and long term. In this regard, these scores also improve communication between physicians and families regarding difficult information. Furthermore, in contexts where the resources are limited, evidence-based data could sustain physicians in choosing the optimal solution to reduce the duration of hospitalization, hospital costs and the suffering of both the patient and the family.

Although the HATCH score demonstrated greater sensitivity and specificity for outcome prediction, it has limitations compared to SAFIRE. For instance, it includes treatment type as a variable, which prevents its use for prognosticating patients in critical condition at diagnosis—precisely when the benefit of treatment is most uncertain, given the high risks of complications, poor outcomes, and extended ICU stays. Moreover, HATCH considers complications such as hydrocephalus, which negatively impact prognosis but may arise long after the initial hemorrhage. Consequently, not all its variables are confined to the admission phase.

The SAFIRE score more than the HATCH score with its simple combination of usually evaluated factors can be readily available at the moment of SAH diagnosis sustaining the decisional process that may take place in the emergency department.

This study presents several limitations that should be acknowledged. Although it was an external validation, the retrospective, single-center design may limit the generalizability of the findings. Indeed, the study’s reliance on historical clinical records and imaging data carries inherent risks of information bias, particularly in evaluating treatment decisions and complications. Additionally, while the study rigorously evaluated the SAFIRE and HATCH scores, other predictive tools and external validations were not comprehensively included, potentially overlooking alternative prognostic indicators. Specifically, the lack of data concerning the SAHIT score is an important limitation but was forced by the impossibility of finding it. Another limitation stems from missing or incomplete follow-up data for some patients, which might have introduced bias in assessing long-term outcomes.