Association between stress hyperglycemia ratio and early neurological deterioration after acute ischemic stroke: a retrospective cohort study

Abstract

Background

Stress hyperglycemia ratio (SHR) has emerged as a more accurate indicator of stress-related hyperglycemia than absolute glucose levels. However, its relationship with early neurological deterioration (END) after acute ischemic stroke (AIS) remains unclear.

Methods

We retrospectively analyzed 1,479 AIS patients admitted within 24 h of symptom onset. END was defined as an increase of ≥ 2 points in the NIHSS total or motor score within 72 h. SHR was calculated as the ratio of fasting plasma glucose to estimated average glucose derived from HbA1c and categorized into quartiles. Logistic regression, generalized additive models (GAM), two-piecewise logistic regression, and statistical mediation analyses were performed.

Results

Among 1,479 patients, 270 (18.3%) developed END. Higher SHR was independently associated with increased END risk (fully adjusted OR = 6.19, 95% CI: 2.68–14.28, P < 0.0001), showing a clear dose-response relationship across quartiles (P for trend = 0.0015). GAM revealed a non-linear relationship, and two-piecewise regression identified a potential inflection point at SHR ≈ 1.06. Subgroup analysis showed a stronger association in non-diabetic patients (interaction P = 0.0033), with no significant interactions for other variables. Sensitivity analysis adjusting for C-reactive protein (CRP) and white blood cell (WBC) count remained robust after adjustment. Mediation analysis indicated that CRP and WBC were statistically associated with the SHR-END relationship, with mediation proportions of 12.89% and 8.03%, respectively.

Conclusions

Elevated SHR was associated with an increased risk of early neurological deterioration in patients with acute ischemic stroke, following a non-linear pattern without a defined threshold. The association was more pronounced in non-diabetic patients, suggesting heterogeneity across metabolic subgroups. These findings support the potential value of SHR for early risk stratification and warrant further prospective validation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-026-04654-6.

Introduction

Acute ischemic stroke (AIS) remains a major cause of death and long-term disability worldwide, and up to one-fifth of patients experience early neurological deterioration (END) within the first days of admission [1, 2]. END, typically defined as a measurable worsening of neurological status shortly after onset, is strongly associated with poor functional outcomes and increased mortality [3, 4]. Timely identification of patients at high risk for END is therefore essential for guiding early treatment and preventing secondary brain injury [5, 6].

Acute hyperglycemia is a frequent metabolic response to cerebral ischemia, even in patients without pre-existing diabetes [7–9]. This stress-related hyperglycemia results from neuroendocrine activation, systemic inflammation, and impaired glucose utilization [10, 11]. Traditional markers such as absolute blood glucose are strongly influenced by chronic glycemic status and may underestimate the acute stress response [12, 13]. To overcome this limitation, the stress hyperglycemia ratio (SHR)—calculated as the ratio of fasting plasma glucose to estimated average glucose derived from HbA1c—has been proposed as a more robust index by accounting for background glycemia [14, 15]. Previous studies have linked elevated SHR to adverse outcomes in critical illness and AIS, but its specific role in predicting END remains insufficiently explored [16, 17].

The biological link between SHR and END may involve inflammatory activation. Hyperglycemia can amplify ischemia-induced neuroinflammation, promote leukocyte infiltration, and disrupt the blood–brain barrier, accelerating neuronal injury [18–20]. Circulating inflammatory markers such as C-reactive protein (CRP) and WBC count are easily measurable, and both have been independently associated with hyperglycemia and poor stroke outcomes [21–23]. Whether CRP and WBC mediate the relationship between SHR and END has not been clearly established.

In this study, we investigated the association between SHR and END in a well-characterized cohort of AIS patients admitted within 24 h of onset. We further examined potential non-linear and threshold effects, evaluated subgroup differences, and conducted mediation analyses to clarify the contribution of CRP and WBC. Our findings aim to establish the prognostic value of SHR in the acute phase of stroke and provide new insight into metabolic–inflammatory interactions in risk stratification.

Materials and methods

Study design and participants

This retrospective cohort study included consecutive patients with AIS admitted to Zhejiang Provincial People’s Hospital (Affiliated People’s Hospital, Hangzhou Medical College) between October 2018 and September 2021. The study was approved by the Ethics Committee of Zhejiang Provincial People’s Hospital (No. 2021QT391) and conducted in accordance with the Declaration of Helsinki. The requirement for informed consent was waived due to the use of anonymized data.

Eligible patients met all of the following criteria: (1) AIS diagnosed and admitted within 24 h of symptom onset; (2) brain MRI within 48 h confirming acute infarction on standard sequences; (3) fasting plasma glucose and HbA1c measured within 48 h; and (4) sufficient data to calculate the SHR.

Exclusion criteria were: (1) severe hepatic dysfunction, defined as alanine aminotransferase (ALT) > 10 × upper limit of normal (ULN) or aspartate aminotransferase (AST) > 3 × ULN; severe renal dysfunction (creatinine > 443 µmol/L); active malignancy; or hematologic disease; (2) significant cardiopulmonary insufficiency, including New York Heart Association (NYHA) class III–IV heart failure, left ventricular ejection fraction (LVEF) < 40%, chronic obstructive pulmonary disease (COPD), or respiratory infection on admission; (3) active infection at admission; (4) pregnancy or lactation; (5) inability to assess stroke severity using the NIHSS due to coma, profound aphasia, or other neurological limitations (exclusion based on feasibility, not severity); (6) multiple AIS admissions during the study period (only the first included); or (7) missing key laboratory data. The process of patient inclusion and exclusion is summarized in a flow diagram (Supplementary Fig. 1).

Baseline data collection

Baseline demographic and clinical data were obtained from electronic medical records, including age, sex, body temperature, systolic blood pressure (SBP), diastolic blood pressure (DBP), smoking status, hypertension, type 2 diabetes mellitus, and atrial fibrillation (AF). Stroke severity was evaluated using the NIHSS, swallowing function by the Kubota Water Drinking Test (KWDT), and consciousness level by the Glasgow Coma Scale (GCS). NIHSS assessments were performed at admission, and repeated at 24 and 72 h, by the same experienced neurologist to ensure consistency and reduce inter-observer variability.

Fasting venous blood samples were drawn on the second morning after admission (06:00) by trained nurses, kept at 4 °C, and analyzed within 2 h by certified laboratory staff. Laboratory tests included WBC, CRP, fasting blood glucose (FBG), AST, ALT, glycated hemoglobin (HbA1c), homocysteine (HCY), triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-c), and low-density lipoprotein cholesterol (LDL-c), as well as free thyroxine (FT4), free triiodothyronine (FT3), and thyroid-stimulating hormone (TSH). Estimated glomerular filtration rate (eGFR) was calculated using the CKD-EPI equation.

Normal reference ranges were: TSH 0.27–4.2 mIU/L, FT4 12–22 pmol/L, and FT3 3.1–6.8 pmol/L.

Neurological severity at admission was assessed using the National Institutes of Health Stroke Scale (NIHSS) [24]. Level of consciousness was evaluated with the Glasgow Coma Scale (GCS) [25]. Baseline functional status was assessed on admission (baseline) using the modified Rankin Scale (mRS) [26], a widely validated measure of post-stroke disability. In addition, the A2DS2 score was calculated on admission (baseline) using routinely available clinical variables, including age, sex, atrial fibrillation, dysphagia, and NIHSS score [27].

Stroke etiology was classified according to the Trial of Org 10,172 in Acute Stroke Treatment (TOAST) criteria [28]. Based on clinical features, neuroimaging findings, vascular imaging, and cardiac evaluations, patients were categorized into four subtypes: large-artery atherosclerosis (LAA), small-artery occlusion (SAO), cardioembolism (CE), and other determined or undetermined etiologies (Other). TOAST classification was determined by experienced neurologists based on all available clinical and imaging data.

Definitions

END was defined as an increase of ≥ 2 points in the total NIHSS score or ≥ 2 points in the NIHSS motor subscore, occurring within the first 72 h after admission [2]. This definition has been adopted in several recent original studies of acute ischemic stroke, although definitions of END vary across the literature. Classification as END-positive was based solely on the presence of initial neurological worsening, irrespective of subsequent improvement during the observation period.

The SHR was defined as the fasting plasma glucose (FPG) level divided by the estimated average glucose (both expressed in mg/dL) [29].

The estimated average glucose was calculated from HbA1c (%) using the validated formula: .

Statistical analysis

All statistical analyses were performed using R software (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria) and EmpowerStats (version 2.0; X&Y Solutions, Inc., Boston, MA, USA). A two-tailed P-value < 0.05 was considered statistically significant. The Shapiro–Wilk test was used to assess normality. Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range, IQR) and compared using the independent-samples t-test or Mann–Whitney U test, as appropriate. Categorical variables are expressed as counts (percentages) and compared using the chi-square test.

Multivariable logistic regression was used to examine the association between SHR and END, with SHR analyzed both as a continuous variable and by quartiles (P for trend calculated). Non-linear associations were explored using generalized additive models (GAM) with a binomial distribution and logit link, and threshold effects were evaluated using two-piecewise logistic regression, with inflection points determined by likelihood-based search. Sensitivity analyses additionally adjusted for C-reactive protein and white blood cell count. Subgroup analyses were conducted by age, sex, smoking status, hypertension, diabetes, atrial fibrillation, chronic obstructive pulmonary disease, and TOAST subtype, with interactions tested by adding multiplicative terms. In subgroup analyses with a limited number of outcome events, prespecified parsimonious core-adjusted models were applied to minimize the risk of overfitting and unstable estimates.

Causal mediation analyses were performed to assess whether C-reactive protein and white blood cell count mediated the association between SHR and END, using generalized linear models with probit link and nonparametric bootstrapping (1,000 resamples) to estimate indirect, direct, and total effects, as well as the proportion mediated.

Covariates for all models are listed in the corresponding table footnotes.

Results

Baseline characteristics

A total of 1,479 patients with acute ischemic stroke were included, of whom 270 (18.3%) developed END occurring within 72 h of admission. Patients in the END group were older, had lower body temperature, and had a lower proportion of males and current smokers (all P < 0.05) (Table 1). Atrial fibrillation was more frequent, whereas diabetes was less prevalent. Laboratory tests showed higher fasting plasma glucose, LDL-C, and homocysteine, but lower triglycerides and FT3 in the END group. Inflammatory markers (CRP and WBC) were elevated, and SHR was significantly higher. Clinically, patients with END had lower GCS and higher mRS, with slightly lower NIHSS scores at admission (all P < 0.05).

Table 1.

Baseline characteristics of acute ischemic stroke patients stratified by early neurological deterioration (END) status

Demographic characteristic	ALL(N = 1479)	non-END (N = 1209)	END (N = 270)	P-value*
Age (years)	69.01 ± 12.57	68.29 ± 12.57	72.21 ± 12.08	< 0.001
SBP (mmHg)	153.03 ± 21.62	152.65 ± 21.56	154.70 ± 21.85	0.205
DBP (mmHg)	83.03 ± 12.98	83.11 ± 12.96	82.65 ± 13.09	0.559
Body temperature (℃)	36.47 ± 0.47	36.49 ± 0.48	36.41 ± 0.39	< 0.001
Gender, n (%)				0.031
Female	588 (39.76%)	465 (38.46%)	123 (45.56%)
Male	891 (60.24%)	744 (61.54%)	147 (54.44%)
Current smoking, n (%)				0.045
No	912 (61.66%)	731 (60.46%)	181 (67.04%)
Yes	567 (38.34%)	478 (39.54%)	89 (32.96%)
Hypertension, n (%)				0.899
No	333 (22.52%)	273 (22.58%)	60 (22.22%)
Yes	1146 (77.48%)	936 (77.42%)	210 (77.78%)
Diabetes, n (%)				0.324
No	937 (63.35%)	773 (63.94%)	164 (60.74%)
Yes	542 (36.65%)	436 (36.06%)	106 (39.26%)
Atrial fibrillation, n (%)				< 0.001
No	1224 (82.76%)	1025 (84.78%)	199 (73.70%)
Yes	255 (17.24%)	184 (15.22%)	71 (26.30%)
COPD, n (%)				0.436
No	1387 (93.78%)	1131 (93.55%)	256 (94.81%)
Yes	92 (6.22%)	78 (6.45%)	14 (5.19%)
TOAST, n (%)				0.052
LAA	724 (57.37%)	605 (56.97%)	119 (59.50%)
SAO	304 (24.09%)	255 (24.01%)	49 (24.50%)
CE	183 (14.50%)	152 (14.31%)	31 (15.50%)
Other	51 (4.04%)	50 (4.71%)	1 (0.50%)
Laboratory parameters
HbA1c, (%)	6.00 (5.50–7.20)	6.00 (5.60–7.20)	6.05 (5.50–7.20)	0.786
FPG (mg/dL)	102.06 (90.00-130.41)	100.98 (89.82-128.52)	106.20 (93.02-140.53)	0.008
TG (mg/dL)	113.37 (80.60-157.65)	114.26 (81.48-160.31)	104.96 (73.73-146.14)	0.029
TC (mmol/L)	4.29 (3.65–5.02)	4.28 (3.62–5.01)	4.34 (3.77–5.16)	0.162
HDL-C (mmol/L)	1.15 ± 0.31	1.15 ± 0.31	1.18 ± 0.33	0.128
LDL-C (mmol/L)	2.88 ± 1.03	2.85 ± 1.01	3.01 ± 1.10	0.033
Homocysteine (mmol/L)	15.30 (12.07–19.60)	14.92 (11.70–19.40)	17.00 (13.60–20.00)	< 0.001
FT3 (pmol/L)	3.85 (3.32–4.38)	3.88 (3.35–4.39)	3.66 (3.16–4.28)	0.012
FT4 (pmol/L)	13.06 (11.34–14.68)	13.06 (11.37–14.78)	13.07 (11.02–14.43)	0.195
FT3/FT4	0.29 (0.24–0.34)	0.29 (0.24–0.34)	0.28 (0.23–0.33)	0.139
TSH (µIU/mL)	1.62 (0.97–2.63)	1.62 (0.98–2.59)	1.55 (0.89–2.79)	0.716
AST (U/L)	19.00 (15.40–24.00)	19.00 (15.30–24.00)	20.00 (16.00–25.00)	0.161
ALT (U/L)	17.00 (12.00–25.00)	17.00 (12.00–25.00)	16.00 (12.00–26.00)	0.522
CRP (mg/L)	3.00 (1.30-7.00)	2.82 (1.23–6.26)	4.33 (2.00–15.00)	< 0.001
WBC (×109/L)	7.07 (5.72–8.80)	6.96 (5.60–8.57)	7.81 (6.16–9.96)	< 0.001
EGFR (ml/min/1.73m2)	99.29 (79.50-118.94)	99.21 (80.93-118.72)	100.10 (74.66-120.68)	0.474
UA (umol/L)	314.00 (255.80-383.10)	313.00 (256.80-379.10)	318.80 (250.65-397.85)	0.430
SHR	0.81 (0.73–0.91)	0.81 (0.72–0.90)	0.84 (0.77–0.96)	< 0.001
Clinical characteristics
KWDT	1.00 (1.00–1.00)	1.00 (1.00–1.00)	1.00 (1.00–1.00)	0.603
A2DS2	3.00 (2.00–4.00)	2.00 (2.00–4.00)	3.00 (2.00–4.00)	0.224
GCS	15.00 (14.00–15.00)	15.00 (14.00–15.00)	15.00 (13.00–15.00)	< 0.001
mRS	1.00 (0.00–3.00)	1.00 (0.00–2.00)	4.00 (2.00–5.00)	< 0.001
Baseline NIHSS	2.00 (1.00–5.00)	3.00 (1.00–6.00)	2.00 (1.00–4.00)	0.001

Data are presented as mean ± standard deviation (SD), median (interquartile range, IQR), or number (percentage), as appropriate

P-values were calculated using independent t-test, Mann–Whitney U test, or chi-square test, depending on variable distribution. Bold P-values indicate statistical significance (P < 0.05)

Abbreviations: END early neurological deterioration, COPD chronic obstructive pulmonary disease, TOAST Trial of Org 10,172 in Acute Stroke Treatment, LAA large-artery atherosclerosis, SAO small-artery occlusion, CE cardioembolism, Other other determined or undetermined etiology, FPG fasting plasma glucose, SHR stress hyperglycemia ratio, TG triglyceride, FT3 free triiodothyronine, FT4 free thyroxine, TSH thyroid-stimulating hormone, AST aspartate aminotransferase, CRP C-reactive protein, WBC white blood cell count, UA uric acid, KWDT Kubota water drinking test, A2DS2 age, atrial fibrillation, dysphagia, sex, and stroke severity score, GCS Glasgow Coma Scale, mRS modified Rankin Scale, NIHSS National Institutes of Health Stroke Scale

Across SHR quartiles, significant differences were observed in metabolic, inflammatory, and clinical characteristics (Table 2). Higher SHR was associated with a higher prevalence of atrial fibrillation, as well as higher HbA1c and fasting glucose levels. LDL-C and total cholesterol showed modest upward trends. FT3 decreased, FT4 increased, and the FT3/FT4 ratio declined. Inflammatory markers (CRP, WBC) and uric acid were elevated in the highest quartile. Clinical severity scores, including KWDT, A2DS2, mRS, and NIHSS, also increased with higher SHR.

Table 2.

Baseline characteristics of acute ischemic stroke patients across quartiles of stress hyperglycemia ratio (SHR)

SHR quartile	Q1 (n = 370)	Q2 (n = 369)	Q3 (n = 370)	Q4 (n = 370)	P-value*
Demographic characteristic
Age (years)	69.71 ± 11.42	68.30 ± 13.18	68.19 ± 12.89	69.83 ± 12.67	0.201
SBP (mmHg)	150.95 ± 21.64	153.98 ± 22.08	152.76 ± 21.50	154.43 ± 21.18	0.127
DBP (mmHg)	81.66 ± 12.42	83.28 ± 13.86	83.01 ± 12.63	84.15 ± 12.88	0.059
Body temperature (℃)	36.51 ± 0.41	36.46 ± 0.37	36.48 ± 0.35	36.45 ± 0.67	0.338
Gender, n (%)					0.059
Female	144 (38.92%)	127 (34.42%)	159 (42.97%)	158 (42.70%)
Male	226 (61.08%)	242 (65.58%)	211 (57.03%)	212 (57.30%)
Current smoking, n (%)					0.059
No	215 (58.11%)	216 (58.54%)	245 (66.22%)	236 (63.78%)
Yes	155 (41.89%)	153 (41.46%)	125 (33.78%)	134 (36.22%)
Hypertension, n (%)					0.362
No	81 (21.89%)	95 (25.75%)	81 (21.89%)	76 (20.54%)
Yes	289 (78.11%)	274 (74.25%)	289 (78.11%)	294 (79.46%)
Diabetes, n (%)					< 0.001
No	185 (50.00%)	268 (72.63%)	263 (71.08%)	221 (59.73%)
Yes	185 (50.00%)	101 (27.37%)	107 (28.92%)	149 (40.27%)
Atrial fibrillation, n (%)					< 0.001
No	304 (82.16%)	319 (86.45%)	318 (85.95%)	283 (76.49%)
Yes	66 (17.84%)	50 (13.55%)	52 (14.05%)	87 (23.51%)
COPD, n (%)					0.247
No	345 (93.24%)	353 (95.66%)	348 (94.05%)	341 (92.16%)
Yes	25 (6.76%)	16 (4.34%)	22 (5.95%)	29 (7.84%)
TOAST, n (%)					0.180
LAA	171 (53.44%)	184 (56.27%)	203 (60.96%)	166 (58.87%)
SAO	80 (25.00%)	91 (27.83%)	67 (20.12%)	66 (23.40%)
CE	55 (17.19%)	43 (13.15%)	44 (13.21%)	41 (14.54%)
Other	14 (4.38%)	9 (2.75%)	19 (5.71%)	9 (3.19%)
Laboratory parameters
HbA1c, (%)	6.60 (6.00-8.40)	6.00 (5.60–6.70)	5.80 (5.40–6.60)	5.90 (5.20–7.18)	< 0.001
FPG (mg/dL)	94.05 (82.98-118.71)	96.84 (88.38–113.40)	101.70 (93.60-121.81)	124.02 (102.60-168.21)	< 0.001
TG (mg/dL)	116.03 (78.83-150.57)	117.80 (82.37-163.85)	109.38 (81.71-156.99)	110.27 (77.06–159.20)	0.452
TC (mmol/L)	4.16 (3.49–4.84)	4.27 (3.68–5.09)	4.41 (3.71–5.02)	4.31 (3.74–5.15)	0.045
LDL-C (mmol/L)	2.76 ± 1.09	2.91 ± 0.97	2.94 ± 1.03	2.91 ± 1.03	0.018
HDL-C (mmol/L)	1.14 ± 0.28	1.15 ± 0.31	1.14 ± 0.28	1.18 ± 0.37	0.382
Homocysteine (mmol/L)	14.70 (11.71–18.80)	15.80 (12.08–20.10)	14.90 (12.10–19.40)	15.98 (12.40–19.70)	0.075
FT3 (pmol/L)	3.88 (3.31–4.39)	3.88 (3.40–4.33)	3.88 (3.37–4.49)	3.71 (3.18–4.32)	0.021
FT4 (pmol/L)	12.82 (10.95–14.63)	12.89 (11.34–14.46)	12.69 (11.10-14.34)	13.54 (11.95–15.33)	< 0.001
FT3/FT4	0.29 (0.25–0.34)	0.29 (0.25–0.34)	0.29 (0.26–0.34)	0.26 (0.22–0.32)	< 0.001
TSH (mIU/L)	1.69 (1.10–2.65)	1.63 (0.93–2.57)	1.68 (0.98–2.69)	1.41 (0.81–2.61)	0.083
AST (U/L)	19.00 (15.33–24.62)	19.00 (16.00–24.00)	19.00 (15.00–23.00)	20.00 (15.20–25.30)	0.223
ALT (U/L)	17.45 (12.00–24.00)	17.50 (12.00-26.10)	16.00 (12.00–24.00)	16.00 (12.00–25.00)	0.543
CRP (mg/L)	2.58 (1.16–5.91)	2.45 (1.20–6.21)	3.00 (1.29–5.97)	4.13 (2.00-12.79)	< 0.001
WBC (×109/L)	6.97 (5.59–8.72)	6.90 (5.59–8.54)	6.79 (5.54–8.34)	7.71 (6.20–9.75)	< 0.001
EGFR (ml/min/1.73m2)	97.42 (77.52-118.14)	99.48 (81.05-119.18)	102.62 (83.16-121.68)	97.70 (76.46-116.05)	0.069
UA (umol/L)	318.85 (252.53-387.77)	323.30 (270.20-392.60)	302.60 (250.70-368.97)	310.80 (245.53–385.40)	0.035
SHR	0.67 (0.61–0.70)	0.78 (0.76–0.79)	0.86 (0.84–0.88)	1.00 (0.95–1.10)	< 0.001
Clinical characteristics
KWDT	1.00 (1.00–1.00)	1.00 (1.00–1.00)	1.00 (1.00–1.00)	1.00 (1.00-2.75)	< 0.001
A2DS2	2.00 (2.00–4.00)	2.00 (2.00–4.00)	2.00 (2.00–4.00)	3.00 (2.00–5.00)	< 0.001
GCS	15.00 (15.00–15.00)	15.00 (14.00–15.00)	15.00 (13.00–15.00)	15.00 (13.00–15.00)	< 0.001
mRS	1.00 (1.00–2.00)	1.00 (1.00–3.00)	1.00 (1.00–3.00)	2.00 (1.00–4.00)	< 0.001
Baseline NIHSS	2.00 (1.00–5.00)	2.00 (1.00–5.00)	2.00 (1.00–5.00)	3.00 (1.00–8.00)	< 0.001

Data are presented as mean ± standard deviation (SD), median (interquartile range, IQR), or number (percentage), as appropriate

P-values were calculated using one-way ANOVA, Kruskal–Wallis test, or chi-square test, depending on variable distribution. Bold P-values indicate statistical significance (P < 0.05)

Abbreviations: SBP systolic blood pressure, DBP diastolic blood pressure, AF atrial fibrillation, COPD chronic obstructive pulmonary disease, TOAST Trial of Org 10,172 in Acute Stroke Treatment, LAA large-artery atherosclerosis, SAO small-artery occlusion, CE cardioembolism, Other other determined or undetermined etiology, HbA1c glycated hemoglobin, FPG fasting plasma glucose, TG triglyceride, TC total cholesterol, LDL-C low-density lipoprotein cholesterol, HDL-C high-density lipoprotein cholesterol, UA uric acid, WBC white blood cell count, CRP C-reactive protein, eGFR estimated glomerular filtration rate, FT3 free triiodothyronine, FT4 free thyroxine, HCY homocysteine, KWDT Kubota water drinking test, A2DS2 age, atrial fibrillation, dysphagia, sex, and stroke severity score, GCS Glasgow Coma Scale, mRS modified Rankin Scale, NIHSS National Institutes of Health Stroke Scale, SHR stress hyperglycemia ratio

Multivariable logistic regression analysis

When modeled as a continuous variable, SHR was associated with higher odds of END: unadjusted OR = 3.78 (95% CI: 2.01–7.09, P < 0.0001); partially adjusted OR = 3.52 (95% CI: 1.92–6.46, P < 0.0001); and fully adjusted OR = 6.19 (95% CI: 2.68–14.28, P < 0.0001). When categorized by quartiles, compared with Q1, the fully adjusted ORs were 1.78 (95% CI: 1.09–2.89, P = 0.0211) for Q2, 2.23 (95% CI: 1.38–3.62, P = 0.0011) for Q3, and 2.14 (95% CI: 1.32–3.46, P = 0.0021), with a P for trend of 0.0015 (Table 3).

Table 3.

Association between stress hyperglycemia ratio and risk of early neurological deterioration: multivariable logistic regression analysis

Exposure	Crude Model (Model 1)		Partially Adjusted Model (Model 2)		Fully Adjusted Model (Model 3)
	OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value
SHR	3.78 (2.01, 7.09)	< 0.0001	3.52 (1.92, 6.46)	< 0.0001	6.19 (2.68, 14.28)	< 0.0001
SHR quartile
Q1	1.0		1.0		1.0
Q2	1.55 (1.04, 2.32)	0.0327	1.62 (1.08, 2.43)	0.0208	1.78 (1.09, 2.89)	0.0211
Q3	1.75 (1.17, 2.60)	0.0059	1.81 (1.21, 2.69)	0.0038	2.23 (1.38, 3.62)	0.0011
Q4	1.90 (1.28, 2.81)	0.0014	1.88 (1.27, 2.80)	0.0018	2.14 (1.32, 3.46)	0.0021
P for trend	0.0015		0.0020		0.0015

Values are expressed as odds ratios (ORs) with 95% confidence intervals (CIs). P-values < 0.05 were considered statistically significant. Bold values indicate statistical significance (P < 0.05)

Model 1: unadjusted

Model 2: adjusted for age and sex

Model 3: adjusted for age, sex, smoking status, hypertension, diabetes mellitus, atrial fibrillation (AF), chronic obstructive pulmonary disease (COPD), Trial of Org 10,172 in Acute Stroke Treatment (TOAST) classification, systolic blood pressure (SBP), diastolic blood pressure (DBP), and the following laboratory variables: triglycerides (TG, Box-Cox transformed), estimated glomerular filtration rate (eGFR), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), aspartate aminotransferase (AST, Box-Cox transformed), alanine aminotransferase (ALT, Box-Cox transformed), uric acid (UA, Box-Cox transformed), and homocysteine (HCY, Box-Cox transformed)

Abbreviations: OR odds ratio, CI confidence interval, SHR stress hyperglycemia ratio, END early neurological deterioration

Non-linear relationship between SHR and END risk modeled by GAM

In the primary generalized additive model (GAM) without adjustment for CRP and WBC, SHR showed a significant non-linear association with the risk of END (edf = 3.86, χ² = 15.56, P for non-linearity = 0.0084; deviance explained = 7.32%; adjusted R² = 0.0525) (Fig. 1; Supplementary Table 1, Model A). Other significant covariates included atrial fibrillation, homocysteine, and NIHSS score (all P < 0.05).

Fig. 1.

Non-linear association between stress hyperglycemia ratio (SHR) and early neurological deterioration (END) in acute ischemic stroke. The figure shows the generalized additive model (GAM) evaluating the non-linear relationship between SHR and the risk of END. In the primary model (Model A), fully adjusted for age, sex, smoking status, hypertension, diabetes mellitus, atrial fibrillation (AF), chronic obstructive pulmonary disease (COPD), Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification, systolic blood pressure (SBP), diastolic blood pressure (DBP), triglycerides (TG, Box-Cox transformed), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), aspartate aminotransferase (AST, Box-Cox transformed), alanine aminotransferase (ALT, Box-Cox transformed), uric acid (UA, Box-Cox transformed), homocysteine (HCY, Box-Cox transformed), estimated glomerular filtration rate (eGFR), and NIHSS score, SHR exhibited a significant non-linear association with END risk (edf = 3.86, χ² = 15.56, P = 0.0084). In sensitivity analysis additionally adjusting for CRP and WBC (Model B), the association remained significant (edf = 4.07, χ² = 12.97, P = 0.0243). Shaded areas represent 95% confidence intervals of the smoothed fit. END: early neurological deterioration; SHR: stress hyperglycemia ratio; GAM: generalized additive model

In a sensitivity analysis additionally adjusting for CRP and WBC, the non-linear relationship remained significant with a modest increase in deviance explained (edf = 4.07, χ² = 12.97, P = 0.0243; deviance explained = 9.04%; adjusted R² = 0.0664) (Supplementary Fig. 2; Supplementary Table 1, Model B).

Threshold analysis using two-piecewise logistic regression identified a potential inflection point at SHR = 1.06 (Supplementary Table 2). Below this threshold, the association between SHR and END was not statistically significant (OR = 3.56, 95% CI: 0.95–13.35, P = 0.0599). Above the threshold, the effect estimate was imprecise and not statistically significant (OR = 4816.92, 95% CI: 0.55–∞, P = 0.0673). The likelihood ratio test showed no significant improvement of the two-piecewise model over the linear model (P = 0.151).

Subgroup analyses of SHR and risk of END

Subgroup analyses using multivariable logistic regression demonstrated that the association between SHR and END remained significant across most clinical strata, including patients aged ≥ 70 years, both sexes, non-smokers, those with hypertension, atrial fibrillation, and large-artery atherosclerosis (Table 4). A significant interaction was found for diabetes status (P for interaction = 0.0033). The association was stronger in non-diabetic patients (OR = 11.92, 95% CI: 2.98–47.72, P = 0.0005), whereas in diabetic patients it was attenuated but remained significant (OR = 2.41, 95% CI: 1.11–5.25, P = 0.0259). No significant interactions were detected for age, sex, smoking, hypertension, atrial fibrillation, COPD, or TOAST subtype (all P for interaction > 0.05). In exploratory analyses applying a higher age threshold, SHR showed a consistent direction of association with END in patients aged ≥ 85 years; however, the estimates were less precise due to the limited number of END events in this subgroup. Similar patterns of wide or unbounded confidence intervals were observed in other small subgroups, including patients with COPD and less frequent stroke etiologies. To address this issue, additional exploratory subgroup analyses stratified by age, COPD status, and stroke etiology (LAA vs. non-LAA) were performed using parsimonious core-adjusted models, yielding more stable estimates without evidence of significant interaction effects (Supplementary Table 3).

Table 4.

Subgroup analyses of the association between stress hyperglycemia ratio and early neurological deterioration

END
SHR	N	OR (95% CI)	P-value	P for interaction
AGE				0.9630
<70	723	6.88 (1.57, 30.12)	0.0104
>=70	756	7.09 (2.44, 20.60)	0.0003
Gender				0.8932
Female	588	8.81 (2.85, 27.26)	0.0002
Male	891	6.30 (1.53, 25.97)	0.0108
Smoking status				0.3969
No	912	7.43 (2.76, 20.03)	< 0.0001
Yes	567	4.19 (0.74, 23.78)	0.1060
Hypertension				0.6387
No	333	6.08 (0.90, 41.29)	0.0648
Yes	1146	7.18 (2.75, 18.76)	< 0.0001
Diabetes				0.0308
No	937	18.23 (4.43, 75.13)	< 0.0001
Yes	542	3.39 (1.16, 9.93)	0.0259
Atrial fibrillation				0.4700
No	1224	4.67 (1.68, 13.03)	0.0032
Yes	255	13.70 (2.38, 78.78)	0.0034
COPD				0.7866
No	1387	6.64 (2.72, 16.20)	< 0.0001
Yes	92	0.00 (0.00, Inf)	0.9998
TOAST				0.9695 (0.9146 #)
LAA	724	6.60 (1.78, 24.49)	0.0048
SAO	304	9.65 (1.98, 46.97)	0.0050
CE	183	3.11 (0.25, 38.72)	0.3787
Other	51	14647.19 (0.00, Inf)	1.0000

Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using multivariable logistic regression models adjusted for age, sex, smoking status, hypertension, diabetes mellitus, atrial fibrillation (AF), chronic obstructive pulmonary disease (COPD), Trial of Org 10,172 in Acute Stroke Treatment (TOAST) classification, systolic blood pressure (SBP), diastolic blood pressure (DBP), and laboratory covariates

Interaction effects were tested by including multiplicative terms between SHR and each stratifying variable in the fully adjusted model. For TOAST subtype, the trend test (treated as ordinal variable) is shown in parentheses (#)

Extremely large ORs or infinite CIs in certain subgroups (e.g., COPD = Yes, TOAST = CE/Other) are due to sparse data and should be interpreted with caution

Abbreviations: END early neurological deterioration, SHR stress hyperglycemia ratio, AF atrial fibrillation, COPD chronic obstructive pulmonary disease, TOAST Trial of Org 10,172 in Acute Stroke Treatment, LAA large-artery atherosclerosis, SAO small-artery occlusion, CE cardioembolism. Bold P values indicate statistical significance (P < 0.05)

Mediation analysis

Mediation analysis suggested that CRP accounted for part of the association between SHR and END, with an indirect effect of 0.004687 (95% CI: 0.001851–0.008594, P < 0.0001) and a mediation proportion of 12.89% (Fig. 2). WBC also showed a smaller but significant mediation effect, with an indirect effect of 0.002874 (95% CI: 0.000653–0.005980, P = 0.008) and a mediation proportion of 8.03%. Direct effects remained significant after adjustment for potential confounders.

Fig. 2.

Mediation analysis showing partial indirect effects of C-reactive protein (a) and white blood cell count (b) on the association between stress hyperglycemia ratio and early neurological deterioration in acute ischemic stroke. Panels (a) and (b) display the results of causal mediation analysis evaluating the indirect effects of CRP and WBC, respectively, on the relationship between the stress hyperglycemia ratio (SHR) and early neurological deterioration (END). For CRP (a), the indirect effect was 0.004687 (95% CI: 0.001851–0.008594, P < 0.0001), accounting for 12.89% of the total effect. For WBC (b), the indirect effect was 0.002874 (95% CI: 0.000653–0.005980, P = 0.008), accounting for 8.03% of the total effect. Direct effects of SHR on END remained statistically significant after adjustment for potential confounders, indicating that inflammation only partially mediated the observed association

Discussion

In this retrospective cohort study of patients with AIS, we found that elevated SHR was independently associated with an increased risk of END [30, 31]. This association showed a nonlinear pattern, but no clear threshold effect was observed [32, 33]. Subgroup analysis further revealed that the relationship was more pronounced in non-diabetic patients, suggesting that premorbid glycemic status may modify the prognostic value of SHR [34]. Moreover, mediation analysis suggested that inflammatory markers, particularly CRP, were statistically associated with this relationship, supporting a close link between metabolic and inflammatory stress responses in acute stroke without implying a definitive causal pathway [35–37].

Our findings are consistent with a growing body of evidence linking SHR to adverse outcomes after AIS [30, 38]. For instance, Xiao et al. demonstrated that higher SHR was significantly associated with 90-day poor functional outcomes in a large cohort of AIS patients [30]. Similarly, Huang et al. conducted a meta-analysis including more than 180,000 patients and confirmed a nonlinear dose-response relationship between SHR and adverse outcomes [38]. Moreover, Zhang et al. reported that higher SHR levels were robustly associated with increased 30-day and 90-day mortality, regardless of diabetes status [39, 40]. Most prior studies have primarily focused on long-term outcomes, such as mortality and functional disability [41–43]. In contrast, our study identified SHR as a predictor of early neurological deterioration. This finding expands the clinical relevance of SHR from long-term prognosis to early in-hospital risk stratification [44]. Definitions of END vary across studies with respect to NIHSS thresholds, motor subscores, and time windows. The definition used in the present study is consistent with that adopted in several recent original studies, while acknowledging that different cut-offs may influence reported incidence and effect estimates [45, 46]. In contrast to prior END-related studies that primarily relied on linear association models, the present study incorporated non-linear modeling and mediation analysis to more comprehensively characterize their association.

The biological mechanisms underlying the association between SHR and END remain incompletely understood. Hyperglycemia during acute stress may exacerbate ischemic injury through multiple pathways, including oxidative stress, blood–brain barrier disruption, and inflammatory activation [47]. Our mediation analysis suggests a potential mediating role of CRP at a statistical level in the association between SHR and END [48]. Importantly, early neurological deterioration in the present study was defined based on changes in neurological status assessed by NIHSS and evaluated by treating neurologists, rather than on the occurrence of specific medical complications. Although systemic infections such as pneumonia may contribute to clinical worsening during the acute phase of stroke, we did not systematically distinguish END attributable to stroke progression from that related to intercurrent medical complications. It is therefore plausible that stress hyperglycemia may increase vulnerability to both neurological deterioration and systemic complications through shared inflammatory and metabolic pathways, which often coexist in the early post-stroke period.

The timing of glucose measurement may influence the observed associations. In this study, SHR was calculated using fasting plasma glucose obtained after admission together with HbA1c, and stress hyperglycemia may be most pronounced at presentation and subsequently modified by early management or evolving neurological status.

Beyond oxidative stress and systemic inflammation, other biological mechanisms may plausibly contribute to the observed association between stress hyperglycemia and early neurological deterioration, including disturbances in energy metabolism, immune–inflammatory activation, and vascular dysfunction [49–54]. However, these mechanisms cannot be directly examined in the present observational study and should be interpreted as supportive context rather than causal explanations. Finally, differences between diabetic and non-diabetic patients deserve attention. In diabetics, long-term adaptation to chronic hyperglycemia may blunt the effects of acute metabolic fluctuations. By contrast, non-diabetics experience sudden glycemic surges that trigger stronger inflammatory and vascular responses [55, 56]. These differences together may explain why the prognostic impact of SHR is more evident in non-diabetic patients. In patients with diabetes, interpretation of SHR is more complex, as HbA1c and post-stroke glucose measurements may be influenced by glycemic variability, acute stress responses, and early treatment-related factors. This complexity may partly explain the attenuated association observed in this subgroup.

In addition to acute metabolic and inflammatory mechanisms, chronic cerebrovascular vulnerability may also contribute to early neurological deterioration. Silent white matter alterations, reflecting an underlying burden of cerebral small vessel disease, are common in elderly patients and have been associated with stroke progression and early neurological worsening, even in the absence of large infarcts [57]. Although such imaging markers were not systematically quantified in the present study, it is plausible that pre-existing white matter damage may reduce neural reserve and increase susceptibility to acute metabolic stress, thereby amplifying the impact of stress hyperglycemia on early neurological deterioration.

In patients with acute ischemic stroke, atrial fibrillation often reflects an underlying cardioembolic source rather than an isolated arrhythmic finding. This clinical context may partly contribute to the observed association between higher SHR and END, as severe embolic events are often accompanied by pronounced systemic stress responses, including stress hyperglycemia [58]. In exploratory analyses, the association between SHR and early neurological deterioration showed a consistent direction in patients aged ≥ 85 years, although the estimates were less precise, likely due to the limited number of events in this subgroup. This finding should be interpreted in the context of prior studies reporting a high-risk and clinically complex profile among very old patients with acute ischemic stroke [59]. Prior studies have similarly shown that elevated SHR is associated with a higher risk of hemorrhagic transformation [35] and worse outcomes after intravenous thrombolysis [60] or endovascular therapy [61]. Interestingly, the effect of SHR differed according to diabetes status. Several studies including Duan et al. [62] and Wang et al. [61], demonstrated that the prognostic impact of SHR is more evident in non-diabetic patients. This observation aligns with our findings and underscores the importance of considering premorbid glycemic status when interpreting the prognostic role of SHR [63].

From a clinical perspective, our findings support incorporating SHR into risk stratification models for AIS. Unlike absolute glucose levels, SHR adjusts for background glycemic status and therefore provides a more individualized assessment [64]. Several studies have also shown that SHR outperforms conventional glycemic indices in predicting outcomes after thrombolysis [54, 60] and mechanical thrombectomy [65]. Although large randomized controlled trials such as SHINE did not show benefits from intensive glucose lowering in AIS [66], our findings suggest that SHR may help identify subgroups of patients—particularly non-diabetics—who face a higher risk of early deterioration. Such patients could benefit from closer monitoring, early intervention, and more targeted management strategies. Importantly, current stroke management guidelines mainly emphasize controlling absolute hyperglycemia, but do not consider the relative degree of stress-induced hyperglycemia. Our results indicate that SHR could complement existing approaches, offering a more dynamic and individualized risk assessment tool. In practice, SHR can be rapidly calculated from routinely available FPG and HbA1c, suggesting that it has potential for bedside application in emergency and inpatient stroke care to guide risk stratification and personalized management.

The strengths of our study include a relatively large sample size, systematic adjustment for potential confounders, and the application of mediation analysis to explore underlying mechanisms [67]. Nevertheless, several limitations should be acknowledged. First, this was a single-center retrospective study, which may restrict the generalizability of our findings despite strict inclusion and exclusion criteria. In particular, the homogeneity of the study population in terms of ethnicity, lifestyle, and access to medical resources may limit the applicability of the results to other settings. Second, blood glucose and HbA1c were measured only once at admission, precluding evaluation of their temporal dynamics, which might provide additional prognostic insights as suggested by previous studies. This limitation is particularly relevant in patients with diabetes, in whom HbA1c may not accurately reflect a stable baseline glycemic state, and fasting glucose measured after stroke onset may be influenced by acute stress responses and early in-hospital management. Consequently, misclassification of stress-induced versus chronic hyperglycemia cannot be fully excluded in this subgroup. Admission random glucose values were not systematically available, precluding sensitivity analyses based on glucose measurements obtained immediately on presentation. Third, although our multivariable models adjusted for important confounders such as stroke severity (NIHSS) and dysphagia, residual confounding cannot be fully excluded. Notably, we did not systematically collect data on in-hospital infections or other medical complications, and therefore could not differentiate early neurological deterioration attributable to stroke progression from that potentially related to intercurrent systemic events. Furthermore, we did not systematically assess the burden of silent white matter alterations or other markers of chronic cerebral small vessel disease, which may influence stroke progression and early neurological deterioration, especially in elderly patients. Unmeasured factors, such as level of consciousness, aspiration risk, use of nasogastric or jejunal tubes, prolonged bed rest, and infarct location, may still influence the results. Fourth, while we employed advanced methods including GAM, subgroup analyses, and mediation models, the possibility of multiple testing bias or model overfitting cannot be completely ruled out. Fifth, we focused on SHR alone and did not incorporate other glycemic markers (e.g., glycemic variability, glucose clearance rate, or insulin resistance indices), nor did we assess inflammatory or immune biomarkers (e.g., cytokines or immune cell function), which may provide complementary prognostic information. Finally, we mainly assessed in-hospital and short-term outcomes, without evaluating long-term endpoints such as stroke recurrence, post-discharge mortality, or quality of life.

Future prospective multicenter studies with larger and more diverse populations are needed to validate our findings and enhance external validity. In particular, dynamic assessment of SHR during the early phase of acute ischemic stroke, together with inflammatory biomarkers, may help clarify the temporal relationship between metabolic stress and early neurological deterioration. Moreover, from a research perspective, future research should focus on integrating SHR into established prognostic frameworks and on elucidating the metabolic–inflammatory pathways underlying END, rather than on therapeutic implications at this stage.

Conclusion

In this retrospective study of patients with acute ischemic stroke, elevated SHR was independently associated with an increased risk of early neurological deterioration, following a nonlinear pattern without a clearly defined threshold. The association was more pronounced among non-diabetic patients, indicating that baseline metabolic status may influence the prognostic relevance of stress hyperglycemia. While inflammatory markers were statistically associated with this relationship, SHR appears to provide complementary prognostic information and can be readily derived from routine clinical testing. These findings support the potential value of SHR in early risk stratification after acute ischemic stroke, particularly in non-diabetic populations. Further prospective, multicenter studies are warranted to validate these observations and to clarify the clinical implications of integrating metabolic and inflammatory markers in early stroke risk assessment.

Authors’ contributions

Y.L., Y.F., and T.T. contributed to study conceptualization and design. Y.L., Y.F., and T.T. performed the formal analysis and software implementation. Y.L. drafted the original manuscript. Y.F. and T.T. contributed to data validation and visualization.W.Y. and Y.S. contributed to investigation, data curation, and methodology. W.Y. and Y.S. provided supervision, resources, and project administration.All authors participated in manuscript revision and approved the final version.

Funding

This study is supported by the Medical and Health Science and Technology Project of Zhejiang Province, China (Grant No. 2023KY032 and 2024KY706); and the Zhejiang Provincial Program of Traditional Chinese Medicine Science and Technology (Grant No.2023ZL238).

Data availability

The datasets generated and/or analyzed during the current study are available in the supplementary materials of this article.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Zhejiang Provincial People's Hospital (Affiliated People's Hospital, Hangzhou Medical College) (Approval Number: 2021QT391). The requirement for informed consent was waived by the Ethics Committee because of the retrospective study design and the use of anonymized, de-identified patient data. The study was conducted in accordance with the principles of the Declaration of Helsinki. All data were anonymized prior to analysis.

Competing interests

The authors declare no competing interests.