Acute ischemic stroke (AIS) continues to be one of the leading causes of death and long-term disability worldwide, and the burden is projected to grow further as populations age (1). Advances in acute reperfusion care and secondary prevention have changed outcomes for many patients, yet individuals receiving broadly comparable treatment can still recover along very different trajectories (2,3). Being able to flag, early on, which patients are likely to improve would be genuinely useful for counselling families and for planning rehabilitation resources.

Early neurological improvement (ENI) is typically judged from the change in NIHSS score over the first 24 hours after stroke onset. Several cohorts have linked this early trajectory to how patients fare months later, although the exact definitions and timing windows used to capture ENI differ from study to study (4,5). Among clinical measures, the admission NIHSS itself remains one of the more consistently reproducible indicators of stroke severity and continues to shape expectations about eventual disability (6). Age and pre-existing disability add further prognostic weight on top of stroke severity alone (7).

Metabolic status has also drawn growing interest in stroke prognosis research. The triglyceride-glucose (TyG) index, calculated from two routinely available laboratory values, offers a practical surrogate for insulin resistance (8). Several pooled analyses now link a higher TyG index with poorer vascular and cerebrovascular outcomes, although the strength of this association is not uniform across populations (9).

Inflammatory and coagulation pathways are just as tightly woven into the biology of acute cerebral ischemia. Elevated CRP is a marker of systemic inflammatory activation that has repeatedly tracked with worse early stroke trajectories (10), while D-dimer reflects ongoing fibrin turnover and has shown a similar prognostic association (11). Platelets, for their part, participate directly in arterial thrombosis and vascular inflammation, making platelet indices a biologically plausible prognostic signal in AIS (12).

Because these variables are biologically intertwined rather than independent, evaluating them properly requires more than a list of P values. Strong pairwise correlations, multicollinearity, and in some datasets complete separation between predictor and outcome can make regression estimates look far more precise than they really are. This study therefore set out to describe the clinical and laboratory correlates of ENI and 90-day mRS in a cohort of AIS patients, while explicitly checking for endpoint concordance, collinearity, and whether the resulting regression models could be reliably estimated at all.

This retrospective analytical study enrolled 200 adult patients with a confirmed diagnosis of AIS. The study was conducted at Department of Neurology Hayatabad Medical Complex Peshawar, during January to June 2026. Eligible patients were 18 years or older, had been admitted within 24 hours of symptom onset, and had complete clinical, laboratory and 90-day follow-up records. Patients with hemorrhagic stroke, transient ischemic attack, severe pre-stroke disability, active malignancy, major chronic inflammatory or autoimmune disease, end-stage renal or hepatic disease, or incomplete 90-day outcome data were excluded. The study was conducted in accordance with the Declaration of Helsinki, with institutional ethical approval and an appropriate waiver of informed consent for retrospective, anonymized data. Age, sex, hypertension, diabetes mellitus, atrial fibrillation and current smoking status were recorded at admission. Baseline NIHSS was assessed on arrival, and admission glucose, triglycerides, CRP, D-dimer, WBC count and platelet count were obtained from routine laboratory testing. The TyG index was calculated as ln[(triglycerides in mg/dL × glucose in mg/dL)/2]. ENI was defined as a reduction of at least 4 NIHSS points from baseline to 24 hours, or a 24-hour NIHSS score of 2 or less. Functional outcome was assessed with the 90-day mRS, with scores of 0-2 classified as favorable. Baseline and 24-hour NIHSS scores were additionally compared as paired observations. Continuous variables are summarized as mean ± standard deviation or median (interquartile range) depending on distribution, and categorical variables as number (percentage). Normality was checked with the Shapiro-Wilk test. Continuous comparisons used the Mann-Whitney U or Welch t-test as appropriate, and categorical variables were compared with the Pearson chi-square or Fisher exact test. Cohen's d was used to quantify standardized differences, and Pearson and Spearman correlations were calculated against 90-day mRS. Multivariable regression was explored and variance inflation factors (VIF) were calculated to assess collinearity, with VIF greater than 10 considered severe. Logistic and ordinal models were examined for complete separation, and an exploratory penalized logistic model excluding ENI was fitted to avoid directly duplicating the endpoint. ROC-based performance was treated as apparent in-sample discrimination rather than external validation. A two-sided P value below 0.05 was considered statistically significant. All analyses were performed in Python using pandas, SciPy, statsmodels and scikit-learn.

The cohort included 200 patients with a mean age of 64.0 ± 11.0 years; 100 (50.0%) were male. Hypertension was present in 146 (73.0%), diabetes mellitus in 69 (34.5%), atrial fibrillation in 53 (26.5%), and current smoking in 75 (37.5%). ENI occurred in 108 patients (54.0%). Baseline NIHSS had a median of 13 (IQR 8-19), and 90-day mRS had a median of 2 (IQR 1-4). Table 1 presents the overall and ENI-stratified baseline characteristics.

Table 1. Baseline clinical and laboratory characteristics by ENI status.

Patients with ENI were significantly younger and had less severe neurological impairment at admission. Glucose, triglycerides, the TyG index, CRP, D-dimer and WBC count were all lower in the ENI group, while platelet count was higher. Sex and diabetes mellitus did not differ significantly between groups, whereas hypertension, atrial fibrillation and smoking status did.

The magnitude of separation between groups was substantial across most variables. Cohen's d reached -3.39 for age, -3.40 for baseline NIHSS, -3.06 for CRP, -3.28 for D-dimer, -3.12 for WBC count and +3.06 for platelet count. The TyG index and glucose showed more moderate standardized differences, and triglycerides showed only a small one. Table 2 summarizes these effect sizes alongside Spearman correlations with 90-day mRS.

Table 2. Effect sizes for ENI group differences and correlations with 90-day mRS.

The strongest rank correlations were seen for 24-hour NIHSS (ρ=0.988), baseline NIHSS (ρ=0.986), CRP (ρ=0.972), D-dimer (ρ=0.973), age (ρ=0.971), platelet count (ρ=-0.971) and WBC count (ρ=0.966). These relationships were interpreted cautiously given that they co-occurred with complete endpoint separation, discussed below.

Mean baseline NIHSS was 13.57 ± 5.75 and mean 24-hour NIHSS was 11.80 ± 9.56, for a mean paired reduction of 1.77 points. The paired t-test confirmed a significant overall change (t=6.00, P<0.001), as did the Wilcoxon signed-rank test (W=4326, P<0.001). Table 3 summarizes this paired analysis.

Table 3. Paired analysis of baseline and 24-hour NIHSS.

This overall change, however, masked opposite trajectories in the two groups: ENI-positive patients improved from 9.01 ± 2.54 to 3.44 ± 2.07, while ENI-negative patients worsened slightly, from 18.93 ± 3.31 to 21.62 ± 3.81. Because the ENI definition itself incorporates 24-hour NIHSS, this variable was not treated as an independent predictor of ENI.

ENI and favorable 90-day outcome proved to be perfectly concordant: all 108 ENI-positive patients had a favorable outcome and all 92 ENI-negative patients had an unfavorable one (κ=1.00, zero discordant pairs). Every ENI-positive patient also had mRS 0-2, and every ENI-negative patient had mRS 3-6. Table 4 summarizes these diagnostic findings.

Table 4. Endpoint concordance and model-diagnostic findings.

Conventional logistic and ordinal regression models exhibited complete separation, and an exploratory penalized model that excluded ENI still produced an apparent AUC of 1.000, with 100% sensitivity and specificity at the selected threshold. This points in one direction: the outcome structure in this dataset contains insufficient overlap for conventional independent prognostic estimation, and the very large odds ratios generated during the failed ordinal modelling were not treated as genuine clinical effect estimates.

This study found marked differences in several clinical and laboratory measures between patients with and without ENI. The clearest clinical signal was baseline stroke severity (6). NIHSS is a well-validated index of neurological deficit (6), and international stroke management guidance continues to anchor acute-phase decision-making around it because of its consistent link with subsequent disability (13). Patients who went on to show ENI had, on average, considerably lower admission NIHSS scores than those who did not, which fits comfortably with the broader literature (6,13). The size of the standardized difference observed here was nonetheless unusually large, and this magnitude should not be extrapolated to other cohorts without independent replication.

Age followed a similar pattern. Older patients are repeatedly shown to recover less completely after stroke (14), a pattern attributed to comorbidity, reduced physiological reserve and slower neurological plasticity (7). Pre-existing functional burden adds further prognostic weight on top of age alone (7,14). The present results point in the same direction, although age cannot be treated as an independently adjusted predictor here given the extreme correlation structure described below.

The TyG index was lower among patients who experienced ENI. This index has increasingly been used as a low-cost surrogate for insulin resistance (8), and has been tied to unfavorable cerebrovascular outcomes in several pooled analyses (9). Mechanistically, insulin resistance is thought to promote endothelial dysfunction, low-grade inflammation, platelet activation and impaired fibrinolysis (8,9), a plausible biological bridge between metabolic status and neurological recovery. The present data, however, cannot establish a causal or independent TyG effect.

Inflammatory markers told a comparable story. CRP and WBC count were both considerably higher among patients who did not show ENI, in keeping with reports that elevated CRP tracks with worse early stroke trajectories (10). Systemic inflammation is closely tied to endothelial activation and blood-brain barrier breakdown after cerebral ischemia (15), processes that can compound secondary tissue injury and slow recovery (10,15).

D-dimer was likewise elevated in the ENI-negative group (11). This is consistent with evidence that heightened fibrin turnover and coagulation activation track with stroke severity (11) and with poorer short-term outcomes after thrombectomy and other acute interventions (16).

Platelet count, in contrast, was higher among patients with ENI. Platelets are active participants in thrombus formation and vascular inflammatory signalling (12), and platelet-related indices have been explored repeatedly as prognostic markers in AIS (12,17). Even so, this analysis cannot distinguish a direct causal role for platelet count from its possible role as a marker of other, unmeasured clinical factors.

The most important finding in this study, however, is methodological rather than biological, and it echoes concerns raised elsewhere about deterministic outcome structures in stroke prediction research (18). ENI and favorable 90-day outcome were perfectly concordant in this cohort, and ENI status completely separated mRS categories. When a predictor separates an outcome this cleanly, maximum-likelihood logistic coefficients no longer converge in the usual sense (18), producing standard errors and confidence intervals that carry little practical meaning (18,19). The very large odds ratios generated during the attempted ordinal regression should therefore not be read as genuine clinical effect sizes.

The same caution applies to the apparent AUC of 1.000 obtained from the exploratory model. In-sample discrimination this perfect is a flag for overfitting or an unusually deterministic outcome structure, not evidence of a clinically deployable model (20). Any prediction tool built from these variables would need proper internal and, ideally, external validation before it could be trusted in practice (20).

There is also a more conceptual issue worth raising. ENI is itself derived from early NIHSS change, while long-term outcome is captured through mRS. These two measures can legitimately move together, but the perfect agreement seen here blurs the line between predictor and outcome (21). The broader surrogate-endpoint literature makes a related point: association with a surrogate marker does not automatically confer independent predictive validity for the outcome that ultimately matters to patients (21).

Taken together, these findings support continuing to study admission NIHSS (6,13), age (7,14), the TyG index (8,9), inflammatory markers (10,15), coagulation biomarkers (11,16) and platelet indices (12,17) in stroke cohorts designed to avoid this kind of endpoint overlap. They do not, however, support treating the regression coefficients generated here as independent predictors. Future work should draw on larger, independently collected cohorts with prespecified predictor sets, imaging and treatment covariates, formal missing-data handling, calibration assessment (20), internal resampling, and external validation.

Several clinical and laboratory measures, most notably admission NIHSS, age, CRP, D-dimer, WBC count, platelet count and the TyG index, showed strong associations with early neurological improvement in this 200-patient AIS cohort. However, the near-perfect overlap between ENI and 90-day functional outcome, together with complete separation across mRS categories, made it impossible to estimate independent adjusted effects with any confidence. These associations should be treated as descriptive signals rather than validated predictors until they are confirmed in independently collected cohorts that do not share this deterministic endpoint structure.

1. Feigin VL, Brainin M, Norrving B, Martins S, Sacco RL, Hacke W, et al. World Stroke Organization (WSO): Global Stroke Fact Sheet 2022. Int J Stroke. 2022;17(1):18-29.
2. Kleindorfer DO, Towfighi A, Chaturvedi S, Cockroft KM, Gutierrez J, Lombardi-Hilliard D, et al. 2021 Guideline for the Prevention of Stroke in Patients With Stroke and Transient Ischemic Attack: A Guideline From the American Heart Association/American Stroke Association. Stroke. 2021;52(7):e364-e467.
3. Boulenoir N, Turc G, Henon H, et al. Early neurological deterioration following thrombolysis for minor stroke with isolated internal carotid artery occlusion. Eur J Neurol. 2021;28(2):479-490.
4. Liu H, Liu K, Zhang K, Zong C, Yang H, Li Y, et al. Early neurological deterioration in patients with acute ischemic stroke: a prospective multicenter cohort study. Ther Adv Neurol Disord. 2023;16:17562864221147743.
5. Heitsch L, Ibanez L, Carrera C, et al. Early neurological change after ischemic stroke is associated with 90-day outcome. Stroke. 2021;52(1):132-141.
6. Girot JB, Richard S, Gariel F, et al. Predictors of unexplained early neurological deterioration after endovascular treatment for acute ischemic stroke. Stroke. 2020;51(10):2943-2950.
7. Han TS, Fry CH, Gulli G, Affley B, Robin J, Irvin-Sellers M, et al. Prestroke disability predicts adverse poststroke outcome: a registry-based prospective cohort study of acute stroke. Stroke. 2020;51(2):594-600.
8. Yang Y, Huang X, Wang Y, Leng L, Xu J, Feng L, et al. The impact of triglyceride-glucose index on ischemic stroke: a systematic review and meta-analysis. Cardiovasc Diabetol. 2023;22:2.
9. Zhou Y, Chen Y, Wang J. Can the triglyceride-glucose index predict stroke outcomes? A systematic review and meta-analysis. Pak J Med Sci. 2025;41(10). doi:10.12669/pjms.41.10.12762.
10. Duan Z, Guo W, Tang T, Tao L, Gong K, Zhang X. Relationship between high-sensitivity C-reactive protein and early neurological deterioration in stroke patients with and without atrial fibrillation. Heart Lung. 2020;49(2):193-197.
11. Abbas NI, Sayed O, Samir S, Abeed N. D-dimer level is correlated with prognosis, infarct size, and NIHSS in acute ischemic stroke patients. Indian J Crit Care Med. 2021;25(2):193-198.
12. Sotero FD, Calçada A, de Sousa DA, Dias M, Fonseca AC, Melo TP, Canhão P. Mean platelet volume is a prognostic marker in acute ischemic stroke patients treated with intravenous thrombolysis. J Stroke Cerebrovasc Dis. 2021;30(6):105718.
13. Powers WJ, Rabinstein AA, Ackerson T, Adeoye OM, Bambakidis NC, Becker K, et al. Guidelines for the early management of patients with acute ischemic stroke: 2019 update to the 2018 guidelines for the early management of acute ischemic stroke. Stroke. 2019;50(12):e344-e418.
14. Tanaka K, Goyal M, Menon BK, Campbell BCV, Mitchell PJ, Jovin TG, et al. Significance of baseline ischemic core volume on stroke outcome after endovascular therapy in patients age ≥75 years: a pooled analysis of individual patient data from 7 trials. Stroke. 2022;53(12):3564-3571.
15. Takata F, Nakagawa S, Matsumoto J, Dohgu S. Blood-brain barrier dysfunction amplifies the development of neuroinflammation: understanding of cellular events in brain microvascular endothelial cells for prevention and treatment of BBB dysfunction. Front Cell Neurosci. 2021;15:661838.
16. Jiang S, Guo P, Cai L, Qian C, Yu J, Xu L, et al. Emergency admission plasma D-dimer and prothrombin activity: novel predictors for clinical outcomes after thrombectomy in acute ischemic stroke with large artery occlusion. CNS Neurosci Ther. 2025;31(2):e70267.
17. Yao Y, Cao X, Zou R, Wen H, Zhang S, Xu H, Guo X, Guo Y. Study on the baseline factors and platelet indices that predict outcome of acute ischemic stroke patients after thrombolytic therapy. Cerebrovasc Dis. 2022;51(3):357-364.
18. Albert A, Anderson JA. On the existence of maximum likelihood estimates in logistic regression models. Biometrika. 1984;71(1):1-10.
19. Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable in logistic regression. Am J Epidemiol. 2007;165(6):710-718.
20. Steyerberg EW, Vergouwe Y. Towards better clinical prediction models: seven steps for development and validation. Eur Heart J. 2014;35(29):1925-1931.
21. Prentice RL. Surrogate endpoints in clinical trials: definition and operational criteria. Stat Med. 1989;8(4):431-440.