This prospective observational study was conducted at Harborview Medical Center (University of Washington, Seattle, WA, USA) between December 2020 and December 2022. Eligible participants were required to enroll within 4 days of aSAH and be capable of undergoing standard assessments for CV. Patients with intracranial hemorrhage not attributed to aSAH or those known to be taking antiviral medications within 2 weeks of arrival were excluded. Written informed consent was obtained from all participants or their legal proxy. All procedures were approved by the University of Washington Human Subjects Division (approval no. STUDY00010645; Seattle, WA, USA).

Participants were monitored for up to 14 days post-aSAH using daily neurological exams and transcranial Doppler ultrasonography (TCD). CV was diagnosed based on new neurological deficits not explainable by other causes, cross-sectional and/or catheter angiographic imaging, and evidence of elevated blood flow velocities on TCD. According to established clinical guidelines (2), a Lindegaard ratio (LR) >3, derived from TCD examinations comparing mean blood flow velocity in the middle cerebral artery to that of the ipsilateral extracranial internal carotid artery, was indicative of CV (2,34,35).

Saliva samples were collected on days 4, 7, 10 and 14 (±1 day) post-aSAH using sterile Dacron swabs (Puritan Medical Products). Samples were stored in DNA preservation medium prior to analysis. HSV-1 DNA was extracted from saliva samples using the QIAamp DNA Mini Kit (cat. no. 51304; Qiagen GmbH), according to the manufacturer's protocol. To ensure specificity for HSV-1, targeted primers of the glycoprotein G gene (36) were used: Forward, 5'-CGCGAACAACAGTGTTAGCG-3' and reverse, 5'-ACGGTCGTCGCATCTGTCTT-3'. The PCR thermocycling conditions included an initial denaturation step at 95˚C for 10 min; followed by 40 cycles of denaturation at 95˚C for 15 sec, annealing at 60˚C for 1 min and extension at 72˚C for 1 min; and a final extension step at 72˚C for 10 min. HSV-1 DNA was quantified in saliva samples using a quantitative fluorescent probe-based PCR assay (TaqMan^® Gene Expression Assay; cat. No. 4331182; Applied Biosystems; Thermo Fisher Scientific, Inc.), according to the manufacturer's instructions. Per standard, a positive threshold for viral shedding was defined as 150 copies of HSV-1 DNA per ml of saliva (26). To determine HSV-1 serostatus, 7 ml blood was drawn from the peripheral IV line of the patients into a serum separator tube. The clinical laboratory at Harborview Medical Center separated the serum from the cells within 2 h. A 1-ml aliquot of the serum was then transferred to a standard transport tube, refrigerated and sent to an external laboratory for analysis within 2 weeks of collection. Specimens underwent HSV-1 Glycoprotein G-Specific Antibody, IgG testing by Chemiluminescent Immunoassay (test 0050292; ARUP Laboratories), following the laboratory standard protocol. An EU/ml index value exceeding 1.10 classified samples as positive (37,38). All personally identifiable information was removed to ensure participant anonymity and confidentiality.

Descriptive statistics were used to characterize the demographic and clinical profiles of the cohort. Categorical variables were analyzed using Pearson's χ² test or Fisher's exact test (where expected cell counts were <5), and continuous variables were analyzed using the Mann-Whitney U test. Linear regression analysis was applied to assess the relationship between HSV-1 viral shedding and CV severity across all shedding time points. While maximum viral load reduces the dataset to one value per subject, individual time point analysis preserves more data, allowing for greater statistical power. The model fit was evaluated using the coefficient of determination (R²), and effect size was assessed using Cohen's f². Traditional multiple regression models were not ideal for this analysis due to the small sample size and the longitudinal nature of the data, where repeated measures were taken from the same subjects over time. Therefore, a Random Forest (RF) multiple regression analysis was performed to assess the relationship between viral load, statistically significant demographic variables and the LR. A model with 20 decision trees (n_estimators=20) was used to balance model complexity and minimize the risk of overfitting. The dataset was randomly divided into a training set (70%) and a test set (30%) for model evaluation. The model's performance was assessed using the R² score, which reflects the proportion of variance in the LR explained by the independent variables. To further account for repeated measures within subjects, generalized estimating equations (GEE) were employed, using an exchangeable correlation structure. A linear GEE model was fitted to assess the association between log-transformed viral load and LR over time. To account for non-linear effects, a quadratic GEE model was also fitted, incorporating both the log-transformed viral count and its squared term.

Finally, Classification and Regression Tree (CART) analysis was performed to identify viral load thresholds associated with increased CV risk. The CART model was constructed based on both average and maximum log₁₀-transformed viral counts. Decision nodes represent thresholds for viral load that categorize patients as at moderate or high risk for developing CV, with LR >3 used as the criterion for defining CV. Statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc.), R version 4.0.5 (R Foundation for Statistical Computing) and Python version 3.8 (Python Software Foundation). P<0.05 was considered to indicate a statistically significant difference. This study conforms to the STROBE (39) guidelines for reporting observational research.

A total of 36 patients with acute aSAH were enrolled in this study (Table I). Of these, 44% (16/36) developed CV and 1 patient died during the course of the study. In total, 139 saliva samples were collected. The proportion of HSV-1 seropositive patients was 88.9% (32/36), with 19.4% (27/139) of all saliva samples testing positive for HSV-1 DNA. Of the seropositive patients, 50% exhibited viral shedding at some point during the study period. Seronegative patients (11.1%; 4/36) did not develop CV or viral shedding.

In the cohort, significant differences were observed between patients who developed CV and those who did not. Specifically, patients in the CV group were significantly younger, with a mean age of 53.6 years compared with 64.0 years in the non-CV group (P=0.02). Additionally, aneurysm size was notably smaller in the CV group, with 93.75% of these patients having <7-mm aneurysms, compared with 50% in the non-CV group (P<0.01). The higher proportion of Hispanic or Latino/a patients in the CV group (31.25% vs. 5%, P=0.07) and the lower proportion of aneurysms measuring 7-10 mm in the CV group (6.25% vs. 35%, P=0.05) were also notable, although these findings were on the threshold of statistical significance.

A significant positive association was identified between HSV-1 shedding and CV severity. Each log₁₀ increase in viral DNA count corresponded with a 0.57 increase in the LR, indicating heightened CV severity. Linear regression analysis (Fig. 1) demonstrated that HSV-1 viral shedding explained 27.8% of the variability in CV severity among patients with aSAH (Cohen's f²=0.386, P=0.005), indicating a large effect size. A multiple regression analysis was conducted to assess the association between HSV-1 viral shedding and CV severity, adjusting for age. After pairwise deletion in instances where no data were collected (due to patient death), the multiple regression analysis included 33 participants. The multiple RF regression achieved an R² of 0.640, explaining 64.0% of the variance in the LR. The predicted values ranged from 1.70 to 5.62, indicating a reasonable alignment with the actual values, despite the use of a smaller number of trees. The model effectively captured the relationship between viral count and CV severity (Fig. 2).

To evaluate the temporal relationship between HSV-1 viral shedding and the development of CV post-aSAH, the Mann-Whitney U test was used to compare the median viral loads (log₁₀-transformed) between patients with and without CV on days 4, 7, 10 and 14 post-aSAH (Fig. 3). Median values were chosen for this analysis due to their robustness in small datasets, as they are less affected by outliers and skewed distributions compared with mean values. The analysis revealed that median viral loads were higher in patients with CV, particularly at Day 10 post-aSAH (P=0.032).

To account for the correlated nature of repeated measures within participants over time, a linear GEE model was fitted (Table II). The model demonstrated a positive association between viral load and LR, showing that for every unit increase in log-transformed viral count, LR increased by 0.024 (95% CI: 0.001-0.048), indicating a significant (P=0.039) but subtle effect. This effect started from an LR of 2.92 when the viral count was at the predicted baseline level.

The study further explored the non-linear dynamics between viral count and CV risk through a quadratic GEE model, incorporating both log-transformed viral count and its square. The analysis (Table II) revealed significant evidence for both the linear component (P<0.001, 95% CI: 0.084-0.259) and the quadratic deviation (P<0.001, 95% CI: 0.012-0.040), illustrating that while LR initially increased with viral count, this trend diminished at higher levels of viral shedding.

Supervised machine learning was employed to identify viral count thresholds associated with increased CV risk. CART analysis, using both average and maximum log₁₀ HSV-1 viral counts, revealed significant thresholds (Fig. 4). The primary threshold, based on the average log₁₀ viral count, identified 3.05 (1,123 count/ml, 95% CI: 2.85-3.25) as the point above which patients were moderately likely to develop CV. The secondary threshold, based on the maximum log₁₀ viral count, identified 4.82 (65,966 count/ml, 95% CI: 4.60-5.04) as the point above which patients were highly likely to develop CV. These thresholds suggested progressive stages of CV risk, emphasizing a non-linear pattern, particularly at elevated levels of viral shedding.