The present single-center, retrospective, exposure-stratified observational cohort was performed at Heyou Hospital (Foshan, China) and included hospitalized patients with AIS who received HBOT within 72 h of symptom onset. All treatment decisions, HBOT delivery, imaging, TCD assessments, neurological evaluations, safety monitoring and follow-up were derived from routine clinical records. The investigators neither assigned treatment nor modified clinical care. Consecutive eligible patients admitted between January 2022 and April 2025 were screened. The present study examined associations between the actual HBOT initiation window and early neurohemodynamic response, early neurological response, 90-day functional outcome and HBOT-related safety outcomes among patients treated with HBOT. No non-HBOT comparison group was included. The present study was conducted in accordance with the Declaration of Helsinki and approved by the Heyou Hospital Institutional Review Board (approval no. 20250721-EXP1). As the analysis used de-identified retrospective data and involved no investigator-assigned intervention, the requirement for written informed consent was waived by the Heyou Hospital Institutional Review Board.

Eligible patients were 18-85 years of age, had AIS confirmed by cranial computed tomography (CT) or magnetic resonance (MR) imaging (17), were admitted within 72 h of symptom onset, had an anterior-circulation ischemic lesion and a baseline NIHSS score of 4-20, and received standard stroke care plus HBOT during hospitalization. Standard stroke care comprised routine guideline-based acute stroke management, including intravenous thrombolysis and/or endovascular thrombectomy when clinically indicated, as documented in the medical record. Eligibility also required a documented first HBOT initiation time within 72 h of symptom onset, baseline TCD before HBOT, repeat TCD within 72 h after the first HBOT session, baseline imaging sufficient to determine infarct side, Alberta Stroke Program Early CT Score (ASPECTS) (18), large vessel occlusion (LVO), and available 90-day mRS data (19). Exclusion criteria were pre-stroke mRS ≥3, posterior-circulation stroke, transient ischemic attack (TIA), intracranial hemorrhage at admission, a neurological disorder that could interfere with functional assessment, baseline NIHSS <4 or >20, HBOT initiated >72 h after symptom onset, unavailable first-HBOT timing, missing baseline or early post-HBOT TCD data, missing 90-day mRS data, terminal disease with an expected survival of <3 months or a 90-day functional outcome primarily determined by a non-stroke condition. Of the 1,627 hospitalized patients with AIS screened, 486 received HBOT within 72 h of symptom onset. A total of 429 patients met all eligibility criteria, comprising 136 patients in the <24 h window, 161 in the 24 to <48 h window and 132 in the 48 to 72 h window.

The primary exposure was the HBOT initiation window, defined by the interval from symptom onset to the start of compression for the first HBOT session. Symptom onset was determined from emergency records, stroke-pathway records, medical notes or family-provided history. When a definite symptom-onset time was documented, that time was used; when onset was unwitnessed or unknown, the last-known-well time, defined as the last time the patient was known to be free of symptoms attributable to the index stroke, was used. The first HBOT time was obtained from the HBOT treatment system or the first documented compression start time in the patients' medical records. Patients were classified into three initiation windows: <24 h, 24 to <48 h and 48-72 h. The 48-72 h window served as the reference in association models because it represented the latest initiation interval within the prespecified 72-h exposure range, allowing the two earlier windows to be compared with the latest-treated group. Onset-to-HBOT time was used to verify exposure separation, and onset-to-admission time was recorded as a timing-related clinical variable. The number of HBOT sessions completed within 10 days was used to characterize early treatment delivery across initiation windows. HBOT followed a standardized routine clinical protocol for hospitalized patients with stroke: 2.0 atmospheres absolute (ATA), 100% oxygen for 60 min at therapeutic pressure once daily and 10-15 min each for compression and decompression, giving a total chamber time of 80-90 min per session. The planned course was 10 sessions over 10 days. Fewer sessions could be completed because of treatment interruption, clinical scheduling, discharge or other routine-care considerations; the actual number completed was analyzed as recorded. HBOT records were reviewed for chamber type, treatment pressure, compression, oxygen-breathing period, air-break arrangement, decompression, total session duration, daily frequency, number of sessions within 10 days, premature interruption and tolerance-related discontinuation. Pre-HBOT safety screening assessed cardiopulmonary stability, consciousness and cooperation, ear and sinus tolerance, and documented contraindications. Treatment interruption or cancellation was recorded when HBOT was stopped or withheld because of intolerance, ear or sinus symptoms, claustrophobia, suspected oxygen toxicity or other safety concerns. The present study compared initiation windows only among patients who received HBOT and did not compare HBOT with non-HBOT care.

Baseline data were extracted from electronic medical records, stroke evaluation records, imaging reports, laboratory systems, TCD reports, HBOT records, discharge summaries and follow-up documentation. Demographic and baseline functional variables included age, sex, body mass index (BMI) and pre-stroke mRS. Vascular risk factors and prior treatments included hypertension, diabetes mellitus, atrial fibrillation, dyslipidemia, coronary heart disease, previous stroke or TIA, current smoking and pre-stroke antithrombotic use. Acute stroke variables included baseline NIHSS score, systolic and diastolic blood pressure, Trial of Org 10172 in Acute Stroke Treatment (TOAST) subtype (20), infarct side, cortical involvement, baseline ASPECTS, LVO, baseline MCA stenosis or occlusion, intravenous thrombolysis and endovascular thrombectomy. MCA stenosis or occlusion was determined from available CT angiography, MR angiography and TCD records. Admission laboratory measures included white blood cell count, neutrophil-to-lymphocyte ratio (NLR), platelet count, glucose, D-dimer, fibrinogen, low-density lipoprotein cholesterol, serum creatinine, serum albumin and homocysteine. NLR was calculated from absolute neutrophil and lymphocyte counts measured in the same blood sample. Baseline TCD measures included affected side MCA-MFV, MCA-PI, MCA resistance index and the MCA-MFV asymmetry ratio (21), calculated as affected-side divided by contralateral MCA-MFV. Early post-HBOT TCD was performed within 72 h after the first HBOT session. Early MCA-MFV change was defined as the percentage change from baseline, whereas early MCA-PI change was defined as the absolute post-HBOT minus baseline difference.

The primary outcome was favorable functional status at 90 days, defined as an mRS score of 0-2. The 90-day mRS was also analyzed as an ordinal outcome ranging from 0 to 6. Early neurological response was defined as a decrease of ≥4 NIHSS points from baseline to day 7 or a day-7 NIHSS score of 0-1. TCD hemodynamic response was defined as an increase of ≥20% in affected-side MCA-MFV after HBOT without a concomitant increase in MCA-PI of ≥0.10. The integrated early neurohemodynamic response phenotype required both TCD hemodynamic response and early neurological response. This phenotype was analyzed solely as an exploratory, post-treatment, treatment-emergent response marker; it was neither a baseline predictive factor nor a validated surrogate endpoint and was therefore not included as a covariate in the primary HBOT timing models. Early neurological deterioration was defined as an increase of ≥4 NIHSS points within 72 h after admission. Safety outcomes included symptomatic intracranial hemorrhage, 90-day mortality, middle ear barotrauma, sinus barotrauma or pressure symptoms, claustrophobia-related treatment interruption, oxygen-toxicity symptoms, severe HBOT-related adverse events and a composite safety endpoint. Symptomatic intracranial hemorrhage was defined as imaging-confirmed intracranial hemorrhage within 7 days accompanied by an increase of ≥4 NIHSS points. Middle ear barotrauma comprised documented ear pain or fullness, hearing decline, tympanic membrane congestion or retraction, bleeding or perforation during or after HBOT. Sinus barotrauma or pressure symptoms comprised documented sinus barotrauma, sinus pain or pressure symptoms during or after HBOT. Claustrophobia-related treatment interruption was defined as premature termination or cancellation of HBOT because of claustrophobia, marked anxiety or inability to tolerate the chamber. Oxygen-toxicity symptoms included suspected treatment-emergent manifestations such as seizure-like symptoms, visual disturbance, marked nausea or dizziness. A severe HBOT-related adverse event was defined as an event leading to premature termination of HBOT, specialist management, prolonged hospitalization or definite organ injury. The composite safety endpoint comprised symptomatic intracranial hemorrhage, 90-day mortality, 90-day recurrent stroke, middle ear barotrauma or a severe HBOT-related adverse event.

Data were extracted using a predefined case-report framework. Core exposure variables, HBOT delivery variables, TCD parameters, reperfusion therapy, NIHSS scores, mRS outcomes and safety events were cross-checked against source records. Imaging and TCD variables were reviewed using the prespecified operational definitions above and, where applicable, established published criteria for ASPECTS and TCD-based cerebrovascular assessment (11,12,18); discrepancies were resolved by re-examining the original records. Range and logical-consistency checks were applied to timing variables, clinical scores, laboratory values, TCD parameters, treatment records, safety records and follow-up outcomes. Values inconsistent with the expected sequence of symptom onset, admission, HBOT initiation, TCD assessment and outcome evaluation were verified against source records. Clinically plausible extreme values were retained after verification; unverifiable values were treated as missing. As the present study included all eligible patients during the present study period, no prospective sample-size calculation was performed. The final cohort comprised 429 patients. Admission glucose was missing in 4 patients; models including this variable therefore used complete cases (n=425). Analyses not involving admission glucose retained the full cohort when applicable. No imputation was performed.

Continuous variables were summarized as mean ± SD or median (interquartile range), as appropriate, and categorical variables as n (%). Baseline characteristics and treatment-delivery variables were compared across HBOT initiation windows using one-way analysis of variance or the Kruskal-Wallis test for continuous variables and the χ² test or Fisher's exact test for categorical variables. No post hoc pairwise comparisons were performed for baseline or treatment-delivery variables; the reported P-values represent overall three-group tests. The primary association between HBOT initiation window and 90-day favorable functional outcome was estimated using multivariable logistic regression with the 48 to 72 h window as the reference. The main model adjusted for age, sex, pre-stroke mRS, baseline NIHSS score, hypertension, diabetes mellitus, atrial fibrillation, baseline ASPECTS, LVO, intravenous thrombolysis, endovascular thrombectomy, admission glucose, baseline affected-side MCA-MFV and baseline affected-side MCA-PI. The 90-day mRS shift was analyzed using ordinal logistic regression with the same covariates; common odds ratios (ORs) >1 indicate greater odds of a lower, more favorable mRS category. TCD hemodynamic response and early neurological response were analyzed using adjusted logistic models incorporating baseline stroke severity, reperfusion therapy, admission glucose and baseline TCD status. Due to the limited number of events, early neurological deterioration was analyzed using an exploratory parsimonious logistic model adjusted for baseline NIHSS score and ASPECTS and intravenous thrombolysis. The integrated early neurohemodynamic response phenotype was examined in an exploratory multivariable logistic model for 90-day favorable functional outcome adjusted for HBOT initiation window and baseline clinical and TCD covariates. Safety outcomes were compared across initiation windows using the χ² test or Fisher's exact test, as appropriate. Robustness analyses for the primary outcome included inverse probability of treatment weighting (IPTW) with stabilized multinomial propensity-score weights trimmed at the 1st and 99th percentiles. Additional analyses adjusted for onset-to-admission time or MCA-MFV asymmetry ratio excluded patients who underwent endovascular thrombectomy or received intravenous thrombolysis, and used mRS 0-1 as an alternative functional outcome. To address potential survivor/immortal-time-related selection, a 72-h landmark sensitivity analysis restricted the cohort to patients who were alive and clinically evaluable at 72 h after symptom onset and reassessed the primary association using the same covariate adjustment set. The IPTW model included the covariates used in the main adjusted model, and post-weighting balance was assessed using standardized mean differences (SMDs). Baseline covariate balance before and after weighting is summarized in Table SI. Prespecified exploratory subgroup analyses assessed the association between initiation at <24 h and 90-day favorable functional outcome, with heterogeneity tested using interaction terms. Additional post hoc subgroup analyses evaluated hypertension, diabetes mellitus and atrial fibrillation using the same framework. Incremental clinical utility was explored by comparing the baseline model with a model that additionally included HBOT timing using the area under the receiver operating characteristic curve (AUC), integrated discrimination improvement (IDI) and decision curve analysis (DCA) (22). This analysis was not intended to develop a prediction model. Treatment-emergent variables, including TCD hemodynamic response, early neurological response and the integrated early neurohemodynamic response phenotype, were not included as covariates in the main HBOT timing models. All timing-model estimates were interpreted as associations rather than causal treatment effects. All tests were two-sided. Analyses were performed in R version 4.4.2 (Posit Software, PBC). P<0.05 was considered a statistically significant difference.

In the overall three-group comparison across the <24 h, 24-<48 h and 48 to 72 h initiation windows, baseline NIHSS score (P=0.039), admission glucose (P=0.004) and baseline affected-side MCA-PI (P=0.029) were higher, whereas baseline ASPECTS (P=0.028), intravenous thrombolysis use (P<0.001), baseline affected-side MCA-MFV (P=0.044) and MCA-MFV asymmetry ratio (P=0.034) were lower. NLR indicated a borderline between-group difference (P=0.050). No statistically significant differences were observed for age, BMI, sex, pre-stroke mRS, vascular risk factors, TOAST subtype, infarct side, cortical involvement, large vessel occlusion, baseline MCA stenosis or occlusion, endovascular thrombectomy or most laboratory measures (Table I).

Onset-to-HBOT and onset-to-admission times were longer across progressively later HBOT initiation windows (both P<0.001). The number of HBOT sessions completed within 10 days did not differ significantly across windows (P=0.079; Fig. 1A-C; Table SII): the median numbers were 9 (IQR, 8, 10), 9 (IQR, 8, 9) and 9 (IQR, 8, 9) for the <24 h, 24 to <48 h and 48 to 72 h groups, respectively.

Earlier HBOT initiation was associated with a greater early increase in MCA-MFV and a smaller early change in MCA-PI (both P<0.001). Adjusted probabilities of TCD hemodynamic response (P<0.001) and early neurological response (P=0.014) were also higher with earlier initiation (Fig. 2).

The 90-day mRS distribution shifted toward less disability with earlier HBOT initiation (P<0.001). Relative to the 48 to 72 h window, initiation at <24 h was associated with greater odds of a lower 90-day mRS category (common OR=2.44; 95% CI, 1.53-3.88; P<0.001), as was initiation at 24 to <48 h (common OR=1.68; 95% CI, 1.09-2.57, P=0.017) (Fig. 3).

In adjusted models using the 48 to 72 h window as the reference, initiation at <24 h and at 24 to <48 h was each associated with greater odds of a favorable 90-day functional outcome (P<0.001 and P=0.033, respectively), a lower 90-day mRS category (P<0.001 and P=0.017) and TCD hemodynamic response (P<0.001 and P=0.043). Initiation at <24 h was additionally associated with greater odds of early neurological response (P=0.003) and lower odds of early neurological deterioration (P=0.025). Neither association was significant for the 24 to <48 h window (P=0.136 and P=0.162, respectively; Table II).

In the present exploratory analysis, the integrated early neurohemodynamic response phenotype was more frequent with earlier HBOT initiation (P<0.001). Favorable functional outcome at 90 days was also more frequent among phenotype-positive than phenotype-negative patients (P<0.001). After adjustment for HBOT initiation window and baseline covariates, phenotype positivity remained associated with greater odds of favorable functional outcome (OR=2.95; 95% CI, 1.76-4.92; P<0.001) (Fig. 4). This treatment-emergent association was not interpreted as evidence that the phenotype is a baseline predictor or a validated surrogate endpoint.

Safety outcomes did not differ significantly across HBOT initiation windows. Relative to the 48 to 72 h group, associations between earlier initiation and favorable 90-day functional outcome remained directionally consistent in the IPTW-weighted analysis, following additional adjustment for onset-to-admission time or MCA-MFV asymmetry ratio, after exclusion of patients who underwent endovascular thrombectomy or received intravenous thrombolysis, and when mRS 0-1 was used as the alternative outcome; these robustness analyses remained statistically significant. In the 72-h landmark analysis (n=421; 215 events), both the <24 h window (OR=2.51; 95% CI, 1.44-4.38; P=0.001) and the 24 to <48 h window (OR=1.76; 95% CI, 1.05-2.94; P=0.031) remained associated with favorable functional outcome (Table III).

After IPTW, covariate balance improved across the prespecified baseline clinical and TCD variables. All weighted covariates met the predefined balance threshold except baseline affected-side MCA-PI, for which the weighted SMD was 0.110, close to the conventional 0.10 threshold (Table SI).

In the prespecified exploratory subgroup analyses, the association between initiation at <24 h and favorable 90-day functional outcome was directionally consistent across most baseline strata. Baseline NIHSS score indicated evidence of interaction (interaction P<0.001), whereas age, sex, baseline ASPECTS, LVO, intravenous thrombolysis, endovascular thrombectomy, admission glucose, baseline affected-side MCA-MFV and baseline affected-side MCA-PI did not show evidence of interaction (Fig. S1). Additional post hoc analyses across common comorbidities demonstrated the same overall direction. For <24 h versus 48 to 72 h, adjusted ORs were 2.71 (95% CI, 1.25-5.88) without hypertension and 2.66 (95% CI, 1.43-4.94) with hypertension; 2.74 (95% CI, 1.48-5.08) without diabetes mellitus and 2.53 (95% CI, 1.12-5.71) with diabetes mellitus; and 2.79 (95% CI, 1.51-5.15) without atrial fibrillation and 2.38 (95% CI, 1.02-5.54) with atrial fibrillation. No interaction was detected for hypertension (interaction P=0.963), diabetes mellitus (interaction P=0.861) or atrial fibrillation (interaction P=0.734; Fig. S1).

Adding HBOT timing to the baseline model increased the AUC numerically, although the change was not statistically significant (P=0.107). IDI was higher after inclusion of HBOT timing (P<0.001), and DCA suggested greater net benefit across part of the evaluated threshold-probability range (Fig. S2).