Hemorrhagic Transformation Mediates Stress Hyperglycemia Effect on Functional Outcome in Minor Ischemic Stroke Authors: Francesco Bax, Mariarosaria Valente, Federico Mazzacane, Matteo Foschi, Raffaele Ornello, Alessandro Marè, Francesco Toraldo, Caterina Del Regno, Roberto Sperotto, Antonio Taglialatela, Beatrice Del Bello, Laura Spigariol, Lucia Pavolucci, Yan Tereshko, Carolina Gentile, Francesco Janes, Simone Lorenzut, Roberto Marinig, Simona Sacco, Anna Cavallini, Gian Luigi Gigli, Giovanni Merlino, Lucio D’Anna Published: 2026 ABSTRACT AbstractBackgroundStress hyperglycemia (SHG) is associated with worse clinical outcomes in minor ischemic stroke. Hemorrhagic transformation (HT) represents a potential mechanism mediating this association, but clear characterization of this relationship is lacking. We therefore aimed to investigate whether HT mediates the effect of SHG on functional outcomes in patients with minor stroke.MethodsThis was a prospective international cohort study (5 comprehensive stroke centers, Italy/United Kingdom). Individuals presenting with National Institutes of Health Stroke Scale ≤5, prestroke modified Rankin Scale ≤1, and without large vessel occlusion between January 2022 and December 2023 were included. SHG was assessed ≤24 hours using the glucose‐to‐glycated hemoglobin ratio. HT ≤7 days was classified in hemorrhagic infarction and parenchymal hematoma. Causal mediation was used to estimate SHG effect on modified Rankin Scale ≥2 at 3 months mediated by HT. Effect modification of age, diabetes, C‐reactive protein, and intravenous thrombolysis was explored.ResultsThere were 1316 patients included (mean age 69.6±12.7 years, 837 men [63.6%]). Higher glucose‐to‐glycated hemoglobin ratio quartiles were associated with modified Rankin Scale ≥2 (adjusted odds ratio [OR], 1.72 [95% CI, 1.24–2.40]; P=0.001) and HT severity (adjusted OR, 2.32 [95% CI, 1.43–3.76]; P=0.001). A significant proportion (14.8%) of the SHG effect on mRS was mediated by HT (average causal mediation effect, 0.014 [95% CI, 0.003–0.030]; P=0.006). Effect modification was found for younger age (≈60 years old), higher C‐reactive protein (≈7 mg/L), and absence of diabetes, but not for intravenous thrombolysis use.ConclusionsIn patients with minor stroke, HT significantly mediated the effect of SHG on unfavorable outcome. The mediating effect of HT was greater in patients either ≈60 years old, without diabetes, or with high baseline C‐reactive protein levels. These initial results might inform patient selection for future interventional studies.Keywords: hemorrhagic transformation, minor stroke, stress hyperglycemiaSubject Categories: Cerebrovascular Disease/Stroke, Intracranial Hemorrhage, Ischemic Stroke Nonstandard Abbreviations and Acronyms glucose‐to‐glycated hemoglobin ratio hemorrhagic infarction hemorrhagic transformation intravenous thrombolysis National Institutes of Health Stroke Scale parenchymal hematoma stress hyperglycemia Trial of ORG 10172 in Acute Stroke Treatment Research Perspective What Is New? In patients with minor stroke (National Institutes of Health Stroke Scale ≤5) recruited across 5 comprehensive stroke centers (Italy and United Kingdom), hemorrhagic transformation mediated the effect of stress hyperglycemia on unfavorable clinical outcome at 3 months. In patients with minor stroke (National Institutes of Health Stroke Scale ≤5) recruited across 5 comprehensive stroke centers (Italy and United Kingdom), hemorrhagic transformation mediated the effect of stress hyperglycemia on unfavorable clinical outcome at 3 months. The mediating effect of hemorrhagic transformation was greater in patients ≈60 years old, without diabetes, or with high baseline C‐reactive protein levels, suggesting that these factors might allow risk stratification for patient selection in clinical research. The mediating effect of hemorrhagic transformation was greater in patients ≈60 years old, without diabetes, or with high baseline C‐reactive protein levels, suggesting that these factors might allow risk stratification for patient selection in clinical research. What Question Should Be Addressed Next? Future studies should address whether early treatment of stress hyperglycemia is associated with lower incidence of hemorrhagic transformation and unfavorable outcome in patients identified at high risk in our study. Moreover, prolonged glucose monitoring after stroke will be required to understand whether an acute or a more sustained glucose lowering is needed to reduce hemorrhagic transformation occurrence. Future studies should address whether early treatment of stress hyperglycemia is associated with lower incidence of hemorrhagic transformation and unfavorable outcome in patients identified at high risk in our study. Moreover, prolonged glucose monitoring after stroke will be required to understand whether an acute or a more sustained glucose lowering is needed to reduce hemorrhagic transformation occurrence. Stress hyperglycemia (SHG) is a common event in acute stroke and refers to the transient glucose elevation that is driven by the complex hormonal counter‐regulatory response occurring during the acute illness phase. 1 Several studies have explored the effect of SHG on clinical outcomes in ischemic stroke, showing that this stress response might be detrimental at different levels. First, it might hamper reperfusion therapy performance, increasing the risk of symptomatic intracranial hemorrhage after intravenous thrombolysis (iVT) 2 and the rate of futile recanalization after mechanical thrombectomy. 3 Second, it increases the risk of recurrence after minor stroke or transient ischemic attack, 4 while also being associated with higher risk of in‐hospital and 30‐day mortality, especially in nondiabetic patients. 5 Despite numerous proposed mechanisms linking hyperglycemia to unfavorable clinical outcomes after ischemic stroke, 6 conclusive evidence on the underlying pathways remains elusive. Additionally, it is uncertain whether SHG's detrimental effects vary among distinct stroke populations, highlighting the need for further targeted research. Hemorrhagic transformation (HT) of ischemic stroke due to reperfusion injury has been identified as a potential mechanism, 6 but clear characterization of this relationship in patients with minor stroke is lacking. Previous observational studies have described the association between admission hyperglycemia or elevated glycated hemoglobin and increased risk of HT 7 , 8 in large ischemic stroke cohorts, but a detailed dissection of the mediating effect of HT on clinical outcome and on minor stroke is currently lacking. A more precise understanding of the mechanisms through which SHG exerts its detrimental role in patients with minor stroke might therefore hold implications for risk stratification, thus allowing a better patient selection for interventional trials targeting SHG. Moreover, a more refined characterization of patient subgroups at higher risk of worse outcomes in the context of SHG, might be informative for clinical practice. To address this question and to help clarify this complex pathway, we used a causal mediation framework to investigate the role of HT as a mediator for the association between SHG and functional outcomes at 3 months in a cohort of patients with minor ischemic stroke. METHODS Study Design and Population This international multicenter prospective cohort study included patients diagnosed with minor stroke from 5 comprehensive stroke centers registries: Udine University Hospital, Udine (Italy); Avezzano Hospital, L'Aquila (Italy); Pavia University Hospital, Pavia (Italy); S. Maria delle Croci Hospital (Azienda Unità Sanataria Locale Romagna), Ravenna, (Italy); and Charing Cross Hospital, Imperial College Healthcare National Health Service Trust, London (United Kingdom). Consecutive patients admitted between January 2022 and December 2023 were included based on a diagnosis of minor ischemic stroke defined by a National Institutes of Health Stroke Scale (NIHSS) score at presentation of ≤5 and a prestroke modified Rankin Scale (mRS) score of ≤1. Patients presenting with large vessel occlusion undergoing mechanical thrombectomy were excluded. Stroke severity was quantified by trained vascular neurologists at admission and discharge using the NIHSS score, whereas functional disability was assessed using the mRS score based on prestroke disability. Cause was classified according to the TOAST (Trial of ORG 10172 in Acute Stroke Treatment) criteria. 9 Further details on clinical variable ascertainment can be found in Data S1. This study was approved by the local institutional review board of Udine University Hospital (coordinating center) (number CEUR‐2020‐O‐173). The study was conducted using data collected as part of routine care, and informed consent was not a legal requirement. Anonymized data are available to other investigators upon reasonable request. Stress Hyperglycemia Assessment SHG was quantified using the glucose‐to‐glycated hemoglobin ratio (GAR) index, expressed as fasting plasma glucose (mg/dL)/hemoglobin A1c (%) as previously reported. 3 Briefly, the GAR index was retrospectively calculated by 2 investigators of the coordinating center (Udine) blinded to the clinical outcome using the raw values obtained within 24 hours after hospitalization from fasting (at least 12 hours) venous blood samples drawn between 6 am and 8 am. Patients were therefore stratified according to GAR quartiles (Q1–Q4), 4 with higher values indicating worse SHG. Mediator HT evaluated within 7 days from the index event was selected as mediator and classified in 4 categories according to the ECASS (European Cooperative Acute Stroke Study‐2) definition: small petechiae along the infarct margins (HI1), confluent petechiae within the infarct area (HI2), blood clots within ≤30% of the infarct area with slight mass effect (PH1), and blood clots within >30% of the infarct area with significant mass effect (PH2). 10 Selected examples of HT are reported in Figure S1. HT was adjudicated by local investigators on head computed tomography scans and validated in each center by an expert neuroradiologist not involved in the care of the patients. To adjudicate HT, we used the head computed tomography scans obtained as per clinical practice at 24/48 hours and 7 days after admission, or between 48 hours and 7 days if the patient was discharged earlier. We defined as symptomatic any intracranial hemorrhage that was associated with neurological deterioration causing an NIHSS increase ≥4 points and when the intracranial hemorrhage itself was deemed the likely cause of clinical worsening. 10 Primary and Secondary Outcomes The primary outcome was poor functional disability at 3 months after stroke defined as mRS ≥2, evaluated by a vascular neurologist trained on mRS scoring in each center, either at a follow‐up outpatient visit or via telemedicine/telephone interviews with patients or their proxies. The secondary outcome consisted of mRS ≥4 at 3 months, to explore the potential mediation effect restricted to worse outcomes. Statistical Analysis We adopted causal mediation using the counterfactual (or potential outcomes) framework 11 to analyze the mediating effect of HT on the association between SHG (expressed as GAR index) and mRS at 3 months. Briefly, in this formulation we defined: E i as the binary exposure for the subject i, which is equal to 1 for high GAR index (Q4) and to 0 for low GAR index (Q1–Q3); Y i (e) as the potential outcome (mRS ≥2 at 3 months) resulting under the exposure e, such that Y i (0) is the outcome for low GAR index and Y i (1) for high GAR index; M i as the occurrence of HT in the presence of high or low GAR index (M i [1] and M i [0], respectively) 11 , 12 (Figure S2). Calculation of causal effects also required key sequential ignorability assumptions to be met, 12 namely that all possible confounding covariates influencing the exposure‐mediator, mediator‐outcome, or exposure‐outcome associations are accounted for. The specific aim of our analysis was therefore to estimate at the population level if the occurrence of HT within 7 days after a minor stroke mediates, at least partly, SHG (exposure) association with an unfavorable functional outcome at 3 months. The total effect of baseline GAR index level on mRS at 3 months was dissected in 2 quantities: (1) an average direct effect, representing the direct effect of GAR index on functional outcome independent of the mediator; and (2) an average causal mediation effect, representing the effect of GAR on mRS at 3 months mediated entirely by the mediator (HT, treated as a binary variable). To calculate indirect and direct effects, as a first step we estimated the effect of the exposure (GAR index, dichotomized as Q4 versus Q1–Q3) on the mediator (HT) using a multivariable binomial and ordinal logistic regression with HT expressed as a dichotomous (HT versus no HT) or categorical ordinal (no HT, HI1, HI2, PH1, and PH2) dependent variable, respectively. To address confounding in the association between exposure, mediator, and outcome we adjusted for a set of covariates potentially influencing GAR levels at admission, HT occurrence, and stroke outcome either singularly or concomitantly (eg, diabetes). These covariates included general demographics (age and sex), common vascular risk factors (diabetes, hypertension, hypercholesterolemia, hypertriglyceridemia, severe carotid stenosis contralateral to the stroke side, history of atrial fibrillation, current smoking, previous stroke, and myocardial infarction), antithrombotic medication use pre‐event, admission systolic blood pressure, stroke characteristics, and treatment (ie, baseline severity as defined by NIHSS, prestroke disability as defined by mRS, ivT), stroke cause (cardioembolic, large artery, or small vessel disease), and inflammation markers (CRP [C‐reactive protein] , log transformed). Overall, these variables were deemed sufficient to capture all possible confounding in this context. We also allowed an interaction term between the exposure and the mediator. Next, we estimated the effect of the mediator (HT) on the primary outcome (mRS ≥2 at 3 months), correcting for the same set of covariates. Mediation estimates (ie, total effect, average direct effect, and average causal mediation effect) were obtained via a quasi‐Bayesian Monte Carlo method with 2000 simulations, and the mediated percentage of the total effect was then calculated. 11 , 13 To provide further insights into the mediating role of the HT components, we estimated mediation effects treating HT as an ordinal variable (parenchymal hematoma [PH] versus hemorrhagic infarction [HI] versus no HT). We then calculated the percentage difference in single HT categories between GAR Q1 to Q3 and Q4 (ie, δQ4:Q1‐3) to analyze categorical shifts. The Benjamini‐Hochberg method was used to account for multiple comparisons across mediation models. Statistical analysis was conducted with R version 4.4 (R Foundation for Statistical Computing) using the mediation package, 13 with P<0.05. Further details about the statistical analysis can be found in Data S1. Subgroup Analysis A stratified analysis was conducted to explore the effect modification of age, diabetes status, iVT use, and baseline CRP on exposure–outcome, exposure–mediator, and mediator–outcome associations, using the 25th and 75th percentiles for continuous outcomes providing a reproducible and distribution independent framework to define low and high strata. These variables were chosen exploratorily based on clinical relevance and possible implication for outcome prognostication. We postulated that older age might modify the effect of SHG on functional outcome and HT occurrence after a minor stroke, although there is no conclusive evidence on the risk of HT, especially after iVT. 14 On the other hand, there is evidence for a lower susceptibility of patients with diabetes to the effects of SHG, 6 whereas higher baseline CRP levels have been associated with worse outcomes in minor stroke. 15 We therefore introduced effect modification by these covariates (Figure 1), allowing both a relaxation of the causal mediation assumptions and exploration of the mediation effect across different clinical scenarios. 16 Causal mediation framework. Black lines represent covariates adjustment. Green lines represent the average causal mediation effect. The red line represents the average direct effect. Dashed lines represent effect modification by age, diabetes, iVT, and CRP. CRP indicates C‐reactive protein; GAR, glucose‐to‐glycated hemoglobin ratio; HT, hemorrhagic transformation; iVT, intravenous thrombolysis; mRS, modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale; SBP, systolic blood pressure; and TOAST, Trial of ORG 10172 in Acute Stroke Treatment classification. Sensitivity Analysis To further test the robustness of our findings, we performed a sensitivity analysis using the glycemic gap, calculated as venous admission blood glucose, hemoglobin A1c average derived glucose (calculated using the formula [28.7 × hemoglobin A1c] − 46.7). 17 , 18 This SHG index uses an absolute difference rather than a ratio, providing complementary information to the GAR index, and showed a comparable performance in predicting HT occurrence in our cohort (data not shown). Finally, to test whether the trajectory of plasma glucose differed over time in individuals with versus without HT, we modeled the interaction between HT occurrence and sampling time points on glucose levels. A subset of 338 subjects (mean age 67.8±12.6 years, 223 men [66%]) from the study coordinating center (Udine University Hospital, Italy) with available follow‐up for glucose control after 24 hours (ie, 48–72 hours and before hospital discharge) and information on insulin therapy were used for this analysis. Further details are available in Data S1. RESULTS Study Population A total of 1472 patients were screened, and after exclusion of 156 patients with either incomplete or duplicate data, and anterior/posterior circulation large vessel occlusions undergoing mechanical thrombectomy, 1316 patients with minor stroke (mean age 69.6±12.7 years, 837 men [63.6%]) were included in the study (Figure 2). A comparison of the main clinical variables among the excluded and included patients is available in Table S1, and baseline characteristics of the study population according to GAR quartiles are reported in Table 1. Higher GAR quartiles were generally associated with a higher burden of vascular risk factors including hypertension, type 2 diabetes, severe carotid stenosis, and atrial fibrillation newly detected during the admission (P from 0.07 to <0.001). On the other hand, smoking was less prevalent in higher GAR quartiles (P<0.001). Study population flowchart. AIS indicates acute ischemic stroke; HbA1c, hemoglobin A1c; mRS, modified Rankin Scale; and NIHSS, National Institutes of Health Stroke Scale. Table 1Baseline Characteristics of the Study Population Overall and Their Comparison Across GAR QuartilesVariables/GAR quartilesQ1 (<15.5)Q2 (≥15.5 and <17.4)Q3 (≥17.4 and <20.2)Q4 (≥20.2)Total P valueTotal, n (%)309 (23.5)350 (26.6)329 (25.0)328 (24.9)1316 (100)0.7Sex (men), n (%)180 (58.3)230 (65.7)218 (66.3)209 (63.7)837 (63.6)0.2Age (y), mean±SD69.9±12.669.6±12.869.3±12.769.5±12.969.6±12.70.6Admission systolic blood pressure (mm Hg), mean±SD159.4±25.5157.2±25.2160.1±25.0158.5±27.7158.8±25.90.9CRP (mg/L), median (IQR)2.7 (1.1; 6.2)1.9 (0.9; 5.4)2.5 (1.1; 6.6)2.8 (0.9; 7.5)2.4 (1.0; 6.7)0.4Total cholesterol (mg/dL), mean±SD181.1±42.3181.7±43.3182.4±43.0178.3±40.6180.9±42.30.5LDL107.6±37.7>109.2±38.9108.6±35.3106.2±35.0108.0±36.70.6HDL51.6±15.851.1±13.751.5±13.650.4±12.451.1±13.90.4Triglycerides (mg/dL)113.7±56.0114.0±52.9115.1±50.0119.5±66.9115.6±56.70.2Hypertension, n (%)195 (63.1)225 (64.3)225 (68.4)240 (73.2)885 (67.2)0.003Type 2 diabetes, n (%)46 (14.9)49 (14.0)82 (24.9)125 (38.1)302 (22.9)<0.001Atrial fibrillation, n (%)66 (21.4)56 (16.0)70 (21.3)96 (29.3)288 (21.9)0.004Newly detected32 (10.4)35 (10.0)44 (13.4)50 (15.2)161 (12.2)0.03Chronic41 (13.3)21 (6.0)28 (8.5)48 (14.6)138 (10.5)0.3Hypercholesterolemia, n (%)120 (38.8)122 (34.9)122 (37.1)133 (40.5)497 (37.8)0.5Smoking (current), n (%)92 (29.8)99 (28.3)74 (22.5)58 (17.7)323 (24.5)<0.001† Severe carotid stenosis, n (%)* 2 (0.6)4 (1.1)4 (1.2)8 (2.4)18 (1.4)0.07Previous stroke, n (%)34 (11.0)43 (12.3)39 (11.9)39 (11.9)155 (11.8)0.8Previous myocardial infarction, n (%)42 (13.6)34 (9.7)34 (10.3)40 (12.2)150 (11.4)0.7Pre‐event antithrombotic use, n (%)120 (38.8)108 (7.9)121 (9.2)119 (9.0)464 (35.3)0.9Antiplatelet89 (32.0)84 (25.5)101 (32.7)93 (30.8)367 (30.1)0.7Anticoagulant31 (14.1)20 (7.5)20 (8.8)26 (11.1)97 (10.2)0.4 P values are unadjusted for trend. CRP indicates C‐reactive protein; GAR, glucose‐to‐glycated hemoglobin ratio; HDL, high‐density lipoprotein; and LDL, low‐density lipoprotein.*Stenosis contralateral to the stroke symptoms' side. † Negative trend. Baseline Characteristics of the Study Population Overall and Their Comparison Across GAR Quartiles P values are unadjusted for trend. CRP indicates C‐reactive protein; GAR, glucose‐to‐glycated hemoglobin ratio; HDL, high‐density lipoprotein; and LDL, low‐density lipoprotein. Stenosis contralateral to the stroke symptoms' side. Negative trend. Stroke Characteristics, Acute Therapies, and HT NIHSS scores at baseline and discharge showed a positive trend with increasing GAR quartiles (P=0.01 and P<0.001, respectively), pointing to a more severe clinical presentation and outcome associated with SHG burden (Table 2). In regard to cause, cardioembolic strokes were more represented in higher GAR quartiles (P=0.04), possibly due to the high proportion of atrial fibrillation in this subgroup. Conversely, the prevalence of large artery and small vessel disease strokes remained consistent across quartiles. Similarly, there was no trend in acute therapy use, with almost 30% of patients receiving iVT, whereas only 1.1% were treated with oral anticoagulation. HT occurrence was positively correlated with SHG severity (P<0.001), and this association was likely driven by HI1 and PH2 categories. The lack of differences in acute therapies between HI and PH (Table S2) suggests that factors other than therapy alone concur to the severity of HT. Table 2Stroke Characteristics, Acute Therapies, and Overall Clinical Outcomes and Their Comparison Across GAR Quantiles (n=1316)Variable/GAR quartilesQ1Q2Q3Q4Total P valueAcute reperfusion therapies, n (%)Intravenous thrombolysis95 (30.7)112 (32.0)92 (28.0)119 (36.3)418 (31.8)0.3Single antiplatelet112 (36.2)117 (33.4)106 (32.2)111 (33.8)446 (33.9)0.5Double antiplatelet46 (14.9)44 (12.6)62 (18.8)51 (15.5)203 (15.4)0.3Oral anticoagulant3 (1.0)3 (0.9)6 (1.8)3 (0.9)15 (1.1)0.8None53 (17.2)74 (21.1)63 (19.1)44 (13.4)234 (17.8)0.1TOAST classification, n (%)Large artery disease52 (16.8)52 (14.9)46 (14.0)47 (14.3)197 (15.0)0.3Cardioembolic87 (28.2)81 (23.1)86 (26.1)113 (34.5)367 (27.9)0.04Small vessel disease63 (20.4)74 (21.1)66 (20.1)71 (21.6)274 (20.8)0.8Other10 (3.2)12 (3.4)17 (5.2)23 (7.0)62 (4.7)0.01Undetermined97 (31.4)132 (37.4)114 (34.7)74 (22.6)416 (31.6)0.01† Baseline NIHSS, median (IQR)2 (1; 3)2 (1; 3)2 (1; 4)3 (2; 4)2 (1; 4)0.01Discharge NIHSS, median (IQR)* 0 (0; 1)0 (0; 1)1 (0; 1)1 (0; 2)1 (0; 2)<0.001mRS at 3 mo, n (%)≥251 (16.5)55 (15.7)76 (23.1)104 (31.7)286 (21.8)<0.001≥332 (10.3)26 (7.4)43 (13.1)67 (20.4)168 (12.8)<0.001≥414 (4.5)11 (3.1)27 (8.2)30 (9.1)82 (6.2)0.002Hemorrhagic transformation, n (%)7 (2.3)17 (4.9)22 (6.7)38 (11.6)84 (6.4)<0.001HI12 (0.6)7 (2.0)8 (2.4)21 (6.4)38 (2.9)<0.001HI20 (0)3 (0.9)2 (0.6)5 (1.5)10 (0.8)0.06PH15 (1.6)3 (0.9)9 (2.7)5 (1.5)22 (1.7)0.6PH20 (0)4 (1.1)3 (0.9)7 (2.1)14 (1.1)0.02Spontaneous ICH within 7 d, n (%)2 (0.6)2 (0.6)3 (0.9)10 (3.0)17 (1.3)0.01 P values are unadjusted for trend. GAR indicates glucose‐to‐glycated hemoglobin ratio;HI1, small petechiae along the infarct margin; HI2, confluent petechiae within the infarct area; ICH, intracerebral hemorrhage; IQR, interquartile range; mRS, modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; PH2, blood clots within >30% of the infarct area with significant mass effect; and TOAST, Trial of ORG 10172 in Acute Stroke Treatment.*n=11 (0.8%) missing. † Negative trend. Stroke Characteristics, Acute Therapies, and Overall Clinical Outcomes and Their Comparison Across GAR Quantiles (n=1316) P values are unadjusted for trend. GAR indicates glucose‐to‐glycated hemoglobin ratio; HI1, small petechiae along the infarct margin; HI2, confluent petechiae within the infarct area; ICH, intracerebral hemorrhage; IQR, interquartile range; mRS, modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; PH2, blood clots within >30% of the infarct area with significant mass effect; and TOAST, Trial of ORG 10172 in Acute Stroke Treatment. n=11 (0.8%) missing. Negative trend. Association Analysis Overall, higher GAR quartiles were associated with higher risk of mRS ≥2 at 3 months with a stepwise gradient (odds ratio [OR]Q3vsQ1, 1.47 [95% CI, 0.99–2.19]; P=0.06; ORQ4vsQ1, 2.22 [95% CI, 1.52–3.29]; P<0.001) (Table 3). Similarly, a GAR index increase was associated with higher HT categories, again showing a severity gradient (ORQ3vsQ1, 3.10 [95% CI, 1.37–7.94]; P=0.01; ORQ4vsQ1, 5.56 [95% CI, 2.60–13.8]; P<0.001). Adjustment by predefined covariates overall confirmed these results, with some new insights (Table 4). Specifically, in the outcome model, PH2 emerged as the major HT component driving worse outcome, whereas age, NIHSS at baseline, prestroke mRS, baseline CRP, and diabetes history emerged as relevant associations. In regard to the mediator model, the likelihood of HT being in a higher category was significantly associated with a higher GAR index (P=0.001) and cardioembolic stroke cause (P=0.02), while being inversely associated with small vessel disease‐related strokes (P=0.001). Overall, these results confirmed the association between exposure–outcome, exposure–mediator, and mediator–outcome. In regard to glucose trajectories over time, serum glycemia was higher at admission and at 24 hours in individuals with HT compared with individuals without HT (P<0.001), but no clear difference was found thereafter, regardless of insulin therapy use (Table S3; Figure S3). These results further suggest that early glucose elevation, independent of subsequent insulin treatment and glycemic normalization, is associated with HT and argue against a spurious association. Table 3Unadjusted Logistic Regression Models for the Mediator (HT) and Primary Outcome (mRS ≥2)PredictorOutcome model (mRS≥2)* Mediator model (HT)† Odds ratio95% CI P valueOdds ratio95% CI P valueGAR quartilesQ2 vs Q1 0.910.60–1.390.72.200.94–5.760.08Q3 vs Q1 1.470.99–2.190.063.101.37–7.940.01Q4 vs Q1 2.221.52–3.29<0.0015.562.60–13.8<0.001Observations/R 2 1316/0.0411316/0.039GAR indicates glucose‐to‐glycated hemoglobin ratio index; HI1, small petechiae along the infarct margins, HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect.*The outcome model includes adjusting for the mediator HT categorized as no HT, HI1, HI2, PH1, and PH2. † Ordinal logistic regression models with HT categorized as no HT, HI1, HI2, PH1, and PH2. Unadjusted Logistic Regression Models for the Mediator (HT) and Primary Outcome (mRS ≥2) GAR indicates glucose‐to‐glycated hemoglobin ratio index; HI1, small petechiae along the infarct margins, HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect. The outcome model includes adjusting for the mediator HT categorized as no HT, HI1, HI2, PH1, and PH2. Ordinal logistic regression models with HT categorized as no HT, HI1, HI2, PH1, and PH2. Table 4Regression Models for the Mediator (HT) and Primary Outcome (mRS ≥2) Adjusted for Variables Potentially Confounding the Association Between Exposure, Mediator, and OutcomePredictorsOutcome model (mRS ≥2)Mediator model (HT)† Odds ratio95% CI P valueOdds ratio95% CI P valueGAR (Q4 vs Q1‐3)1.721.24–2.400.0012.321.43–3.760.001Hemorrhagic transformationHI11.350.59–2.920.4………HI20.650.09–3.010.6PH12.060.73–5.470.15PH211.63.4–46.7<0.001Age, y1.031.02–1.05<0.0011.010.99–1.030.4Sex (women)0.870.64–1. 180.41.280.79–2.120.3Baseline NIHSS1.431.28–1.60<0.0011.080.91–1.280.4Prestroke mRS2.811.95–4.05<0.0010.530.24–1.080.1Acute reperfusion therapy (iVT)0.840.60–1.170.31.190.72–1.980.4Admission systolic blood pressure (per mm Hg increase)1.000.99–1.000.61.000.99–1.010.8Baseline CRP (log of unitary increase)1.251.08–1.440.0021.230.98–1.520.06Hypertension1.320.94–1.890.11.270.73–2.270.4Diabetes history1.551.10–2.180.011.000.56–1.730.9Atrial fibrillation history1.010.59–1.710.90.770.34–1.610.5Hypercholesteremia history1.140.83–1.560.41.150.70–1.890.5Severe carotid stenosis0.490.12–1.610.30.610.03–3.660.6Smoking (current)1.290.89–1.840.20.660.33–1.220.2Pre‐event antithrombotic useAntiplatelet (vs none)0.800.55–1.160.21.040.57–1.850.9Anticoagulant (vs none)0.940.50–1.710.80.600.18–1.620.3Previous ischemic event* 1.260.86–1.830.21.190.65–2.130.6CauseCardioembolic1.340.90–2.010.11.871.11–3.170.02Large artery disease1.340.84–2.110.20.520.20–1.160.1Small vessel disease1.190.79–1.800.40.140.03–0.390.001Observations/R 2 1316/0.1801316/0.108CRP indicates C‐reactive protein; GAR, glucose‐to‐glycated hemoglobin ratio index; HI1, small petechiae along the infarct margins; HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; iVT, intravenous thrombolysis; NIHSS, National Institutes of Health Stroke Scale; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect.*Composite of transient ischemic attack, stroke, or myocardial infarction. † Ordinal logistic regression model with HT categorized as no HT, HI1, HI2, PH1, and PH2. Regression Models for the Mediator (HT) and Primary Outcome (mRS ≥2) Adjusted for Variables Potentially Confounding the Association Between Exposure, Mediator, and Outcome CRP indicates C‐reactive protein; GAR, glucose‐to‐glycated hemoglobin ratio index; HI1, small petechiae along the infarct margins; HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; iVT, intravenous thrombolysis; NIHSS, National Institutes of Health Stroke Scale; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect. Composite of transient ischemic attack, stroke, or myocardial infarction. Ordinal logistic regression model with HT categorized as no HT, HI1, HI2, PH1, and PH2. Mediation Analysis The mediation analysis showed that a significant proportion (14.8%) of the SHG effect on the primary outcome was mediated by HT (average causal mediation effect, 0.014 [95% CI, 0.003–0.030]; P=0.006) (Table 5). Similarly, there was a significant effect of SHG on the outcome in the absence of HT mediation (average direct effect, 0.087 [95% CI, 0.033–0.140]; P<0.001), pointing toward the presence of additional non‐HT‐related mechanisms driving this association. These results were similar overall when accounting for an ordinal mediator (PH versus HI versus no HT), with a proportion mediated of 14.9%. Although further mediation decomposition was not possible due to the low number of observations in single HT categories, categorical shifts analysis provided further insights. In patients with mRS ≥2, HI1 and PH2 had the greater proportional change (δQ4:Q1‐3=9.1% and 3.6%, respectively). On the other hand, in patients with mRS ≥4, PH2 showed the greatest change (δQ4:Q1‐3=10.9%) (Figure 3). The effect proportion of SHG on mRS≥4 at 3 months mediated by HT was higher (23.0%), suggesting that this pathway might be more relevant in mediating worse levels of disability. Table 5Mediation Analysis of the Association Between GAR Index and mRS at 3 Months via HT for Primary (mRS ≥2) and Secondary (mRS ≥4) OutcomesEstimates (95% CI) P value P value* Primary outcome (mRS≥2)Binary mediator (HT vs no HT)Total effect0.090 (0.037 to 0.140)<0.001<0.001Average direct effect0.087 (0.033 to 0.140)<0.001<0.001Average causal mediation effect0.014 (0.003 to 0.030)0.0030.006Proportion mediated14.8%……Ordinal mediator (PH1‐PH2 vs HI1‐HI2 vs no HT)Total effect0.098 (0.029 to 0.190)0.0030.006Average direct effect0.094 (0.028 to 0.180)0.0030.006Average causal mediation effect0.017 (0.001 to 0.060)0.020.03Proportion mediated14.9%……Secondary outcome (mRS ≥4)Binary mediator (HT vs no HT) modelTotal effect0.025 (−0.006 to 0.060)0.10.1Average direct effect0.021 (−0.013 to 0.060)0.20.2Average causal mediation effect0.007 (0.000 to 0.020)0.020.03Proportion mediated23.0%……Each mediation model included n=1316 observations.GAR indicates glucose‐to‐glycated hemoglobin ratio index; HI1, small petechiae along the infarct margins; HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect.* P values adjusted for multiple comparisons using the Benjamini‐Hochberg correction. Mediation Analysis of the Association Between GAR Index and mRS at 3 Months via HT for Primary (mRS ≥2) and Secondary (mRS ≥4) Outcomes Each mediation model included n=1316 observations. GAR indicates glucose‐to‐glycated hemoglobin ratio index; HI1, small petechiae along the infarct margins; HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect. P values adjusted for multiple comparisons using the Benjamini‐Hochberg correction. Bar chart showing categorical shifts in HT proportion across GAR quartiles for primary (A) and secondary (B) outcomes. GAR indicates glucose‐to‐glycated hemoglobin ratio; HI1, small petechiae along the infarct margins; HI2, confluent petechiae within the infarct area; HT, hemorrhagic transformation; mRS, modified Rankin Scale; PH1, blood clots within ≤30% of the infarct area with slight mass effect; and PH2, blood clots within >30% of the infarct area with significant mass effect. The stratified analysis (Table S4) showed that age modifies the effect of SHG on HT (≤60 years old: OR, 5.82 [95% CI, 1.64–22.36]; P=0.007 versus ≥80 years old: OR, 2.27 [95% CI, 0.91–5.56]; P=0.07) and of HT on the outcome (≤60 years old: OR, 6.19 [95% CI, 1.48–24.80]; P=0.01 versus ≥80 years old: OR, 2.15 [95% CI, 0.86–5.40}; P=0.1). Moreover, baseline CRP was also a modifier of HT on the outcome (≤1 mg/L: OR, 4.23 [95% CI, 1.30–13.32]; P=0.01 versus ≥7 mg/L: OR, 2.18 [95% CI, 0.86–5.45]; P=0.09) and more mildly of SHG on outcome, and diabetes was an effect modifier of HT on the outcome. A borderline effect of iVT on SHG–HT association was also found. Based on these results, we explored mediation effects conditioning on varying levels of these effect modifiers (Table S5). The proportion mediated was higher when conditioning to younger age (15.1% versus 4.0%), higher baseline CRP levels (9.2% versus 7.8%), and absence of diabetes (9.1% versus 6.9%), suggesting that this mechanistic pathway might be more relevant in younger, nondiabetic patients with higher baseline CRP. On the other hand, when conditioning to iVT treatment status, no clear differences emerged from the mediation analysis. We note that the effect modification found for age might also reflect the lower proportion of diabetes (≤60 years old, 13.9% versus ≥80 years old, 24.9%) and the relatively higher absolute GAR index values (median GAR index ≤60 years old, 21.4 versus ≥80 years old, 19.4) observed in younger patients with HT. Finally, the sensitivity analysis using the glycemic gap was overall concordant with the main results, showing similar proportions mediated both for the primary (13.3%) and secondary (19.6%) outcomes, further supporting the robustness of our findings (Table S6). DISCUSSION This international multicenter cohort study provides supporting evidence on the possible causal role of HT in mediating the SHG effect on unfavorable clinical outcome at 3 months in patients with minor stroke. Our findings suggest that, especially in patients who are younger (≈60 years old) and nondiabetic, and with high baseline levels of CRP, HT of any degree might represent a relevant factor driving the effect of SHG, therefore providing an initial mechanistic framework in this stroke population. The proportion mediated by HT found in our cohort (≈15%) also suggests that a relevant quota of SHG effect on functional outcome is mediated by alternative mechanisms including futile recanalization, 3 vascular inflammation, 19 , 20 increased brain lactate production, 21 and oxidative stress. 22 , 23 However, we note that part of these pathways can also contribute to HT development, 14 , 19 , 20 further complicating the picture. In this context, HT might therefore represent an early and clinically detectable effect of SHG, potentially allowing the identification of a subpopulation at higher risk of unfavorable outcome. It is plausible that part of the mediation effect found could be driven by the high proportion of patients with cardioembolic strokes in our cohort, because this causative subgroup usually presents a higher incidence of HT after ivT. 14 However, our analysis was adjusted both for stroke cause and reperfusion therapy, therefore suggesting an effect of SHG on HT, at least partly, independent of it. Moreover, the lack of clear effect modification according to iVT treatment argues against a strong role of thrombolysis in this setting. Supporting these findings, evidence from an embolic middle cerebral artery occlusion mouse model shows that preexisting hyperglycemia is associated with worse HT severity at 24 hours, independent of iVT treatment. 19 The causal framework proposed (Figure 1) and the findings of a causal link between SHG and HT occurrence driving worse outcome after a minor stroke are supported by clinical data from large cohort studies showing that hyperglycemia at admission (either acute or chronic as defined by high levels of hemoglobin A1c) is associated with increased risk of HT in patients with stroke. 7 , 8 Furthermore, evidence from experimental mouse models indicates that hyperglycemia induced before embolic or transient middle cerebral artery occlusion is associated with increased HT, alongside increased markers of blood–brain barrier dysfunction and inflammation. 19 , 20 Hyperglycemia can act by inducing a proinflammatory response with secondary upregulation of metalloproteinases, causing extracellular matrix degradation and potential triggering of HT. 19 , 20 In this context, our findings on CRP are consistent with previous studies 24 , 25 showing that risk of subsequent HT after ischemic stroke was associated, although only in univariate analysis, with higher admission levels of CRP, and that its concentration increased with the hemorrhage severity. Although part of this association might be explained by the extension of the initial ischemic lesion and the potential concomitant infectious or inflammatory comorbidities in the acute phase, 26 it is unlikely that these factors have played a major role in our cohort of patients with minor stroke. Nevertheless, it is also possible that CRP acts as a hallmark of the acute phase response accompanying SHG, and therefore simply represents an epiphenomenon of a wider stress reaction. We hereby confirmed previous findings on the lower impact of SHG in patients with diabetes, a result that could be explained by 2 main occurrences. First, patients with history of diabetes are often already treated with glycemic lowering medications at admission, and if not, they are more likely to receive glucose‐lowering agents during their hospital stay based on their previous diagnosis. Second, if insulin therapy is administered, it might exert anticoagulant effects via inhibition of thromboxane production and reduction of plasminogen activator inhibitor‐1, therefore limiting the extension of the ischemic lesion. 27 In this regard, the increased mediating effect of HT found in lower age strata might be partly attributed to the lower proportion of diabetes observed among younger study participants. Our results also raise the question of whether targeting SHG with an intensive glucose lowering regimen, specifically in patients at higher risk of HT, might provide a clinical benefit not found in previous interventional trials considering broader ischemic stroke populations. 28 Our findings are in contrast with a secondary analysis of the SHINE (Stroke Hyperglycemia Insulin Network Effort) trial, 29 showing no association between SHG indexes at baseline and functional outcome at 3 months. However, the SHINE trial included patients with a higher NIHSS score at baseline (median score of 7) and used a more conservative definition of good functional outcome (mRS score of 0) for patients presenting with an NIHSS score between 3 and 7, the subgroup closer to our study population. Moreover, the trial evaluated the effect of different levels of baseline SHG indexes on intensive versus standard glycemic control, whereas we did not account for acute insulin treatment. Taken together, these results underline the importance of HT as a mechanism in non‐mechanical thrombectomy‐treated stroke populations, pointing to pathways other than reperfusion injury (typically associated with mechanical thrombectomy‐treated patients) driving HT and unfavorable outcome in the context of SHG. Although a formal subgroup analysis limited to single HT categories was not possible, the ordinal mediation model showed similar results compared with the binary model (Table 2), suggesting that there might be a comparable mediation effect across HI and PH. This hypothesis is further supported and complemented by the observation that HI1 and PH2 relative proportions showed the greater increase from GAR Q1‐3 to Q4 compared with other HT categories in patients with mRS ≥2, and PH2 in patients with mRS ≥4 (Figure 3). These findings confirm previous evidence pointing to PH2 as the category usually associated with worse functional outcome, 30 while providing new insights into the mediating effect of HI in the context of SHG, at least for intermediate levels of mRS. However, given the limited number of observations for some of the HT strata, this subanalysis should be interpreted cautiously. Overall, our study expands previous evidence on SHG in minor stroke and outcome prognostication. Former studies in minor stroke populations focused on ischemic stroke recurrence due to SHG and were based on classic associational regression analysis. 4 The mediation approach design used herein allows at least some degree of causal inference and provides evidence for HT as one of the potential causal mechanisms of SHG in determining unfavorable outcome. Moreover, we highlighted relevant baseline factors associated with higher risk of HT following SHG, providing evidence for potential risk stratification and patient selection in clinical research. In this context, future studies should address whether early treatment of SGH is associated with lower incidence of HT in patients identified at high risk in our study. Another relevant point that should be addressed with a more prolonged glucose monitoring follow‐up is whether an acute or more sustained glucose lowering would be needed to reduce HT occurrence. There are, however, several limitations. First, although we tried to account for all covariates and effect modifiers associated both with the mediator and the outcome, it is plausible that there might be residual unmeasured confounders driving some of the associations. In particular, we did not systematically collect information on pre‐event or in‐hospital acute treatment of hyperglycemia as well as any data on ischemic and hemorrhagic lesions size, severity, or location, therefore limiting any inference in this regard. Second, no interrater agreement across study centers for HT assessment was planned, raising the possibility of differential HT ascertainment across sites. However, the ECASS scale is an extensively validated tool, 31 , 32 with an overall good performance in distinguishing HT versus no‐HT and PH versus HI among different raters, 31 indirectly supporting an overall balanced assessment across sites. Third, we did not plan a systematic blood glucose collection after the 24‐hour time point across sites, limiting inference after this timeframe. Moreover, the categorization of GAR index into quartiles might have resulted in potential loss of information; however, we used this approach to uniform with previous landmark literature on stress hyperglycemia in the context of stroke. 4 Finally, although we dissected 1 causal pathway, SHG might act through other concomitant pathways that might not be captured by the available data set. CONCLUSIONS In our study, SHG was associated with unfavorable outcome at 3 months (mRS ≥2) in patients with minor stroke, and this effect was partly, but significantly, mediated through HT. The mediating effect was greater in patients either ≈60 years old, without diabetes, or with high baseline CRP levels, suggesting that these subpopulations might be at higher risk of worse outcome in the context of SHG. However, whether tighter acute, and possibly subacute, glycemic control after stroke would translate in reduced occurrence of HT and better clinical outcomes in this patient population will require further randomized evidence. In this regard, these initial results, based on a large multicentric cohort, might serve to inform patient selection for future interventional studies addressing SHG treatment in minor stroke. Sources of Funding None. Disclosures None. Supporting information Data S1. Supplemental Methods Tables S1–S6 Figures S1–S3 Acknowledgments Author contributions: Conceptualization: F.B., G.M., L.D.A., M.V. Methodology: F.B., G.M., L.D.A., M.V. Formal analysis: F.B. Data curation: G.M., L.D.A., F.M., M.F. Data collection: F.M., M.F., R.O., A.M., F.T., C.D.R., R.S., A.T., B.D.B., L.S., L.P., Y.T., C.G., F.J., S.L., R.M. Writing (original draft): F.B., G.M., L.D.A., M.V. Writing (revision and editing for intellectual content): F.B., G.M., L.D.A., M.V., F.M., M.F., R.O., A.M., F.T., C.D.R., R.S., A.T., B.D.B., L.S., L.P., Y.T., C.G., F.J., S.L., R.M., A.C., S.S., G.L.G. This article was sent to Jose R. Romero, MD, Associate Editor, for review by expert referees, editorial decision, and final disposition. Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.125.046218 For Sources of Funding and Disclosures, see page 11. References Associated Data Supplementary Materials Data S1. Supplemental Methods Tables S1–S6 Figures S1–S3