This cohort investigation harnessed data spanning the temporal expanse from January 2010 to December 2016, which were meticulously attained from an exclusive prospective registry system centered in South Korea (11). The primary independent variable was the BUN/Cr ratio, while the dependent variable was the 3-month outcomes in cases with AIS, categorized as either AOs or favorable outcomes.

Information was obtained from the research conducted by Kang et al., titled “Geriatric nutritional risk index predicts AOs in AIS cases - automated undernutrition screen tool” (11). It is imperative to underscore the open-access nature of this article, governed by the Creative Commons Attribution License, thereby conferring unhindered permissions for utilization, dissemination, and reproduction across all media, contingent upon the stipulation of due credit to the original author and source (11).

Using data from a single-center prospective registry in South Korea that began enrollment in October 2002, we screened 2084 patients diagnosed with AIS who were admitted within 7 days of symptom onset from January 2010 to December 2016. The blood sample of each patient was presumed to be obtained from the first blood drawing on admission according to the hospital procedure. Strict exclusion criteria for the study included the following: (1) cases lacking dysphagia testing or relevant laboratory information within the first 24 h after admission; and (2) cases lacking of records of the modified Rankin scale (mRS) score at 3 months post-stroke. A comprehensive schematic detailing the intricate process of patient selection is presented in Figure 1. Given the antecedent approval by the Institutional Review Board of Seoul National University Hospital (Approval No. 1009–062-332) for the original study (11), no additional approval was required for the ensuing secondary analysis.

Derivation of the BUN/Cr ratio involved the division of the serum BUN concentration (mg/dl) by the serum Cr concentration (mg/dL), yielding a continuous variable. The subsequent categorization of the BUN/Cr ratio into quartiles was performed as follows: Q1: <13.846, Q2: 13.861–17.000, Q3: 17.007–21.311, and Q4: ≥21.333.

Assessment of the 3-month outcomes following the onset of AIS was executed through the utilization of the mRS score (12). The meticulous acquisition of data was performed via outpatient visits or methodically conducted structured telephone interviews (11). Participants underwent categorization into two distinct groups based on their outcomes: favorable outcomes, denoted by an mRS score ≤ 2, and AOs, characterized by an mRS score ≥ 3 (12).

In the paradigm of our study, the covariates were judiciously selected, employing the Empower Stats’ Covariates module in tandem with our cumulative clinical acumen. This comprehensive selection encompassed a multitude of both continuous and categorical factors. Among the continuum were a diverse set of variables, including but not limited to the white blood cell (WBC) count, red blood cell (RBC) count, mean corpuscular hemoglobin concentration (MCHC), red blood cell distribution width (RDW), platelet (PLT) count, hemoglobin (HGB) level, hematocrit (HCT), mean corpuscular volume (MCV), total cholesterol (TC), serum triglyceride (TG) levels, low-density lipoproteins cholesterol (LDL-C) levels, high-density lipoproteins cholesterol (HDL-C) levels, alanine aminotransferase (ALT), aspartate aminotransferase (AST), BUN, Cr, glomerular filtration rate (GFR), serum albumin (ALB), total protein (TP), fasting blood glucose (FBG), activated partial thromboplastin time (APTT), body mass index (BMI), and fibrinogen (FIB). Concurrently, the categorical variables covered a broad spectrum, including age, sex, diabetes mellitus (DM), smoking habits, coronary heart disease (CHD), atrial fibrillation (AF), history of previous stroke/transient ischemic attack (TIA), hypertension, stroke etiology, and the National Institute of Health Stroke Scale (NIHSS) score (13). The meticulous retrieval of laboratory data within the initial 24 h of admission was obtained from the electronic medical records (11). The computation of BMI involved the division of weight in kilograms by the square of the height in meters (kg/m²).

In the current investigation, the instances of missing data for key variables, namely TC, TG, HDL-C, LDL-C, FIB, and FBG, were 1 (0.00%), 107 (5.61%), 99 (5.19%), 75 (3.93%), 22 (1.15%), and 139 (7.29%), respectively. To mitigate the potential ramifications of these missing covariates on statistical efficiency during the modeling process, the study employed mean imputations.

The methodology for characterizing continuous variables involved the application of descriptive statistics, specifically, the mean ± standard deviation (SD) for variables exhibiting a Gaussian distribution, and the median (interquartile ranges) for those displaying a skewed distribution. Categorical variables were delineated through frequencies and percentages. The analytical framework encompassed the χ², one-way analysis of variance (ANOVA), or Kruskal–Wallis H tests to discern differences across distinct BUN/Cr ratio groups. Subsequent to a rigorous collinearity assessment [Supplementary Table S1, selection method: the variable inflation factor (VIF) was calculated for each variable, if the highest VIF value ≥5, the variable was removed. The above step was repeated until all remaining variables had a VIF <5, resulting in the elimination of HGB, HCT, and TC], four nuanced models were constructed. These encompassed both univariate and multivariate binary logistic regression analyses, probing the intricate association between the BUN/Cr ratio and AOs in patients with AIS. The fourfold model structure comprised an unadjusted model, a minimally adjusted model accounting only for sociodemographic variables (age sex), a moderately adjusted model accounting for sociodemographic variables and some clinical features (smoking, BMI and past history), and a fully adjusted model accounting for sociodemographic variables and more clinical features (WBC, RBC, MCHC, RDW, AST, ALT, FBG, FIB, TG, LDL-C, ALB, TP, GFR, BMI, DM, previous stroke or TIA, hypertension, AF, CHD, stroke etiology, smoking, and NIHSS score); when latter variables were added to this model, the odds ratio (OR) of the matches changed by at least 10%. Effect sizes, elucidated with 95% confidence intervals (95% CIs), were meticulously documented, with adjustments informed by a synthesis of clinical expertise, literature findings, insights from the Empower Stats’ Covariates module, and outcomes derived from the univariate analyses. In the pursuit of result robustness, a sensitivity analysis was performed by categorizing the BUN/Cr ratio into quartiles and assessing the p-value for trend. We opted to exclude cases with specific conditions, such as DM, abnormal FBG, hypercholesterolemia, elevated BMI, and renal failure. Potential unmeasured confounding factors were systematically explored utilizing E-values. The examination of potential nonlinearity in the relationship between the BUN/Cr ratio and AOs employed sophisticated techniques, including application of generalized additive models (GAMs) and the employment of intricate smooth curve fitting methods, such as penalized splines. The detection of nonlinearity prompted the initiation of a recursive algorithm to pinpoint the inflection point. Subsequently, this led to the establishment of a two-piece binary logistic regression model (BLRM), strategically deployed on either side of this pivotal juncture. The most appropriate model was adjudged based on a log-likelihood ratio test.

All reported findings strictly adhered to the STROBE statement, and the statistical analysis encompassed the usage of both R and Empower Stats (X&Y Solutions, Inc., Boston, MA, United States) software. Significance, in this context, was ascribed to a two-tailed p < 0.05 (14).

Following screening with strict inclusion and exclusion criteria, this study excluded cases lacking dysphagia testing or relevant laboratory information within 24 h after admission (n = 72) and cases lacking a mRS score at 3 months after stroke (n = 106). After the exclusion of 178 cases, the final analysis cohort consisted of 1906 individuals. Table 1 and Supplementary Table S2 delineate the demographic and clinical attributes of the study cohort. The analysis cohort was 61.28% male (1,168 individuals). Participants were stratified into age categories: <60 years (436 participants, 22.88%), 60 to <70 years (505 participants, 26.50%), 70 to <80 years (670 participants, 35.15%), and ≥ 80 years (295 participants, 15.48%). The breakdown of the stroke etiology indicated that 606 (31.79%), 365 (19.15%), 493 (25.87%), 171 (8.97%), and 271 (14.22%) cases were small-vessel occlusion (SVO), large-artery atherosclerosis (LAA), cardiogenic embolism (CE), other determined, and undetermined, respectively. The median (interquartile ranges) of the NIHSS score was 5.0 (1.0–11.0). Participants were further categorized into subgroups based on BUN/Cr ratio quartiles: Q1: <13.846, Q2: 13.861–17.000, Q3: 17.007–21.311, and Q4: ≥21.333. Compared with Q1, Q4 exhibited elevated values of HDL-c, BUN, and GFR, with reduced levels of TG, Cr, and FIB. Q4 featured a higher proportion of women (61.33%), individuals with hypertension (65.90%), and DM cases (34.72%). The distribution of the ratio is visually depicted in Figure 2, illustrating a skewed distribution ranging from 2.469 to 58.621, with a median of 18.163. Supplementary Table S3 shows participants’ baseline characteristics classified by functional outcomes.

A total of 546 participants encountered AOs, yielding an overall incidence rate of 28.65% (26.65–30.71%) (Table 2). The specific incidence rates across the BUN/Cr ratio quartiles were: Q1, 29.14% (25.19–33.34%); Q2, 25.63% (21.86–29.70%); Q3, 25.64% (21.85–29.72%); and Q4, 34.10% (29.96–38.42%).

The univariate analysis elucidated that AOs in patients with AIS lacked statistically significant associations with the MCV (OR = 0.988, 95% CI: 0.969, 1.007), PLT (OR = 0.999, 95% CI: 0.998, 1.001), HDL-C (OR = 0.997, 95% CI: 0.990, 1.005), Cr (OR = 1.017, 95% CI: 0.926, 1.116), GFR (OR = 1.000, 95% CI: 0.996, 1.004), and APTT (OR = 0.988, 95% CI: 0.970, 1.007) (all p > 0.05). Conversely, a positive association was found with the WBC (OR = 1.079, 95% CI: 1.043, 1.116), RDW (OR = 1.232, 95% CI: 1.154, 1.316), BUN (OR = 1.016, 95% CI: 1.006, 1.027), ALT (OR = 1.009, 95% CI: 1.002, 1.016), FBG (OR = 1.008, 95% CI: 1.005, 1.010), FIB (OR = 1.003, 95% CI: 1.002, 1.004), and BUN/Cr ratio (OR = 1.027, 95% CI: 1.012, 1.043) (all p < 0.05). Furthermore, female participants (OR = 1.660, 95% CI: 1.357, 2.030), those aged ≥80 years (OR = 3.996, 95% CI: 2.874, 5.557), individuals with hypertension (OR = 1.343, 95% CI: 1.088, 1.658), diabetes mellitus (OR = 1.446, 95% CI: 1.174, 1.781), a history of previous stroke/TIA (OR = 1.811, 95% CI: 1.437, 2.283), an NIHSS score ≥ 14 (OR = 15.106, 95% CI: 10.827, 21.075), atrial fibrillation (AF) (OR = 2.001 [1.590, 2.517]), and other determined stroke etiologies (OR = 2.073, 95% CI: 1.458, 2.928) showed an elevated risk of AOs (all p < 0.05). In contrast, RBC (OR = 0.583, 95% CI: 0.498, 0.682), HGB (OR = 0.819, 95% CI: 0.779, 0.862), HCT (OR = 0.932, 95% CI: 0.916, 0.949), MCHC (OR = 0.868, 95% CI: 0.797, 0.947), TC (OR = 0.995, 95% CI: 0.993, 0.998), TG (OR = 0.996, 95% CI: 0.994, 0.998), LDL-C (OR = 0.997, 95% CI: 0.994, 1.000), AST (OR = 0.993, 95% CI: 0.986, 1.000), ALB (OR = 0.274, 95% CI: 0.215, 0.350), TP (OR = 0.674, 95% CI: 0.573, 0.793), and BMI (OR = 0.915, 95% CI: 0.886, 0.945) exhibited negative associations with the risk of AOs (Table 3; Supplementary Table S4).

Four distinct models were constructed utilizing the BLRM to thoroughly explore the association between the BUN/Cr ratio and the likelihood of AOs in patients with AIS. The unadjusted model revealed a 2.7% rise in the risk of AOs for every 1-unit increase in the BUN/Cr ratio (OR = 1.027, 95% CI: 1.012, 1.043, p = 0.001), signifying statistical significance. The moderately adjusted model revealed a 1.7% rise in the risk of AOs for every 1-unit increase in the BUN/Cr ratio (OR = 1.017, 95% CI: 1.001, 1.034, p = 0.043), showing statistical significance. However, both the minimally adjusted and fully adjusted models in the multivariate logistic regression analysis did not yield statistically significant results (all p > 0.05) (Table 4).

Conducting a sensitivity analysis entailed the initial transformation of the BUN/Cr ratio from a continuous variable to a categorical variable, employing quartiles, and subsequently reintegrating the categorically transformed BUN/Cr ratio into the model. The multivariate-adjusted model disclosed an inverse association, presenting an OR of 0.609 (95% CI: 0.424, 0.876) for the Q3 versus the Q1. However, Q4, with an OR of 0.702 (95% CI: 0.484, 1.019), did not achieve statistical significance (Table 4, Model III). This observation suggests the potential existence of a nonlinear relationship between the BUN/Cr ratio and AOs in patients with AIS. Furthermore, an E-value of 1.88 surpassed the relative risk (RR; 1.54) associated with unmeasured confounders, indicating that unknown or unmeasured confounders had a minimal influence on the association between the BUN/Cr ratio and the likelihood of AOs in patients with AIS.

In an additional iteration of the sensitivity analysis, cases with TC ≥200 mg/dL, DM, FBG ≥6.1 mmol/L, BMI ≥25 kg/m², and Cr1.2 mg/dL were selectively excluded. The findings unveiled that the previously observed linear relationship between the BUN/Cr ratio and the risk of AOs remained statistically insignificant following meticulous adjustments for confounding factors (all p > 0.05) (Table 5). Consequently, the purported linear association between the BUN/Cr ratio and the risk of AOs did not hold true, both in the overall population and within the specific subgroups. The sensitivity analysis accounted for all covariates, including age, sex, WBC, RBC, MCHC, RDW, AST, ALT, FBG, FIB, TG, LDL-C, ALB, TP, GFR, BMI, DM, history of stroke or TIA, hypertension, AF, CHD, stroke etiology, smoking, and the NIHSS score. Notably, Model I for non-DM cases did not include DM as an adjusted covariate.

In addition, we performed a subgroup analysis based on the TOAST subtypes and different HDL-C levels; however, the results of the analysis were not statistically significant (Supplementary Table S5).

The outcomes derived from the multivariate BLRM did not achieve statistical significance, suggesting the potential influence of intricate nonlinear relationships. Furthermore, the investigation into multivariate-adjusted models utilizing BUN/Cr ratio quartiles as categorical variables hinted at a complex nonlinear association between the BUN/Cr ratio and AOs in patients with AIS. Through the employment of sophisticated statistical techniques, such as GAM and smooth curve fitting, while adjusting for a comprehensive array of covariates, including age, sex, WBC, RBC, MCHC, RDW, AST, ALT, FBG, FIB, TG, LDL-C, ALB, TP, GFR, BMI, DM, history of stroke or TIA, hypertension, AF, CHD, stroke etiology, smoking, and the NIHSS score, a distinctive U-shaped relationship was revealed between the BUN/Cr ratio and the risk of AOs in patients with AIS (Figure 3). To comprehensively capture the nuances of this relationship, a piecewise BLRM was deployed, accommodating two distinctive slopes, with model selection predicated on the log-likelihood ratio test in the sensitivity analysis. Importantly, the p-value for the log-likelihood ratio test in this study was <0.05. Utilizing a recursive algorithm, the inflection point was identified at 21.591; the effect sizes and CIs were meticulously computed on both sides of this inflection point using the two-piecewise BLRM. Notably, on the left side of the inflection point, each 1-unit increment in the Z score of the BUN/Cr ratio was associated with a 31.4% lower risk of AOs (OR = 0.686, 95% CI: 0.519, 0.906, p = 0.008). Conversely, on the right side of the inflection point, the effect size (OR) was 1.406 (95% CI: 1.018, 1.942, p = 0.039) (Table 6). In the analysis, a selective measure was implemented by excluding participants with a TC level ≥ 200 mg/dL, DM, FBG level ≥ 6.1 mmol/L, BMI ≥25 kg/m², and Cr level > 1.2 mg/dL. Importantly, the consistent observation of the same U-shaped curve and an inflection point near 21.591 persisted across the GAM, smooth curve fitting, and recursive algorithm (Supplementary Figure S1).