Abstract
Background
Chronic psychosocial stress induces cumulative physiological dysregulation that accelerates biological aging and contributes to the development of cerebrovascular and neurocognitive disorders, including stroke and cognitive impairment. The Allostatic Load Index (ALI) is a composite measure reflecting multisystem physiological wear and tear resulting from chronic stress and has been associated with stroke mortality and cognitive decline. However, evidence regarding the relationship between ALI and cognitive impairment in populations at high risk for stroke remains limited. This study aimed to investigate the association between ALI and cognitive impairment in a stroke high-risk population, and to evaluate the utility of ALI in stroke-related cognition risk stratification.
Methods
This single-center retrospective observational cohort study was conducted between October 2023 to October 2024. A total of 200 stroke high-risk individuals aged 60 to 85 years, each meeting at least three established stroke risk factors, were included. Thirty-two clinical and laboratory variables were analyzed. Missing data were imputed using the k-nearest neighbors (KNN) algorithm, and multicollinearity was assessed using variance inflation factors (VIF).
ALI was calculated based on 10 physiological indicators spanning cardiovascular, metabolic, and inflammatory systems: waist-to-height ratio, systolic blood pressure, diastolic blood pressure, heart rate, neutrophil count, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycated hemoglobin, and serum creatinine. One point was assigned for each indicator exceeding clinically validated thresholds (total score range: 0–10). Logistic regression was used to assess the association between ALI and cognitive impairment. Sensitivity analyses categorized participants into quartiles of ALI. Feature selection was performed using the Boruta algorithm. Predictive models were developed using logistic regression, decision trees, random forests, and other machine learning algorithms, with model performance evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA).
Results
Cognitive impairment was identified in 101 participants (50.5%). Individuals with cognitive impairment were older and exhibited significantly higher waist-to-height ratios, systolic blood pressure, and white blood cell counts. Restricted cubic splines analysis identified a threshold waist-to-height ratio of 0.55, above which the risk of cognitive impairment increased markedly. ALI followed a normal distribution, and was positively associated with cognitive impairment (OR = 1.27). The association remained robust after multivariable adjustment (Model II: OR = 1.33). The Boruta algorithm confirmed ALI as an important predictive feature. Among all models, random forest model demonstrated the best discrimination, achieving an AUC of 0.8247.
Conclusion
ALI may provide a quantitative measure of cumulative physiological stress burden and appears to be independently associated with cognitive impairment in populations at high risk for stroke. The waist-to-height ratio was identified as a particularly informative component of ALI. These findings suggest the potential utility of integrating ALI into risk assessment frameworks for early identification and prevention of stroke-related cognitive impairment.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12877-026-07154-x.
Keywords: Stroke high-risk populations, Cognitive impairment, Allostatic load index, Machine learning, Risk prediction models
Introduction
Chronic social and psychological stress, together with the physiological imbalances it induces, is increasingly recognized as a key contributor to accelerated aging and the development of multiple diseases. Prolonged activation of stress-response pathways leads to sustained hormone release, causing normal homeostatic regulation to shift toward maladaptive states. When such stress persists, it results in adverse adaptations across interconnected biological systems, ultimately giving rise to cumulative physiological dysregulation [1].
The Allostatic Load Index (ALI) has been proposed as a comprehensive measure of this cumulative biological burden. Previous studies have demonstrated that elevated ALI is associated with increased risks of all-cause mortality, cardiovascular mortality, stroke-related mortality, and a higher incidence of breast cancer in women [2–5]. ALI is typically derived from a range of biomarkers reflecting cardiovascular, inflammatory, metabolic, and other physiological domains [6]. However, variability exists across studies in terms of biomarker selection and calculation methods [6, 7], largely stemming from differences in study populations (e.g., general versus clinical cohorts), research objectives, and the availability of specific biomarkers. Furthermore, given the limitations of Body Mass Index (BMI) in older adults, this study incorporates the waist-to-height ratio to improve capture of abdominal adiposity and cardiometabolic risk. Despite these methodological differences, a growing body of evidence indicates that ALI outperforms traditional single-system biomedical indicators in predicting disease incidence and mortality, largely because it integrates functional information from neuroendocrine, immune, metabolic, and cardiovascular systems [1].
A prior meta-analysis reported that higher ALI levels are significantly associated with increased all-cause and cardiovascular mortality. This relationship is thought to reflect the cumulative impact of chronic stress exposure, which promotes stress-induced neuroendocrine activation, vascular injury, chronic inflammation, and metabolic dysregulation, thereby increasing vulnerability to cardiovascular damage [8]. Consistent with these findings, analyses based on the National Health and Nutrition Examination Survey (NHANES) have shown that elevated ALI is associated with a higher risk of all-cause and cardiovascular mortality among stroke survivors [7]. Beyond cardiovascular outcomes, elevated ALI has also been linked to mental health disorders, cognitive impairment, and a range of chronic somatic diseases [9].
Collectively, these findings support ALI as a robust indicator of chronic stress–related physiological burden and suggest a potential link between cumulative stress exposure and cognitive dysfunction. However, evidence specifically addressing the relationship between ALI and cognitive impairment in stroke high-risk populations remains limited. The present study therefore aims to investigate whether ALI can serve as a predictive marker of cognitive impairment in individuals at high risk of stroke, extending the application of ALI beyond its established role as a general measure of cumulative health deterioration.
Materials and methods
Data source and study design
This retrospective observational cohort study was conducted at a single tertiary medical center, Shenzhen Longhua District Central Hospital. A consecutive enrollment strategy was adopted to minimize selection bias. All hospitalized patients who met the eligibility criteria and were admitted to the Department of Geriatrics between October 2023 and October 2024 were consecutively included.
Sample size estimation
The sample size was estimated based on results from a prior pilot study. Assuming a two-sided significance level of 0.05, a statistical power of 0.80, and an expected effect size corresponding to an odds ratio of 1.8 for cognitive impairment in high and low ALI groups, the minimum required sample size was calculated to be 178 participants. To account for potential exclusions and incomplete data, a final target sample size of 200 participants was selected, ensuring sufficient power for both logistic regression analyses and machine learning model development.
Cognitive impairment assessment
Cognitive impairment was assessed using the Chinese version of the Mini-Mental State Examination (MMSE) score, standardized by Zhang Mingyuan (1990). For consistency with the clinical criteria utilized in this study, an MMSE score of 26 or lower was defined as cognitive impairment. All assessments were conducted by trained neurologists using standardized protocols to ensure consistency and reliability.
Study population
Participants were classified as belonging to a high-risk stroke population according to the Chinese Guidelines for Stroke Prevention and Treatment [10, 11]. To reduce clinical heterogeneity, participants were categorized into primary and secondary prevention groups. The primary prevention group consisted of individuals aged 60 to 85 years with 3 or more modifiable stroke risk factors, weighted by clinical significance as specified in the guidelines [11]. These factors included hypertension, diabetes mellitus, atrial fibrillation/valvular heart disease, smoking history, overweight/obesity, physical inactivity and a family history of stroke. A weighted risk score ≥ 5 was considered indicative of high stroke risk. Secondary prevention group included individuals aged 60 to 85 years with a documented history of transient ischemic attack (TIA) or stroke confirmed by cranial CT/MRI [11]. All participants were required to provide written informed consent.
Abnormal lipid levels were defined as low-density lipoprotein cholesterol (LDL-C) > 3.4 mmol/L or high-density lipoprotein cholesterol (HDL-C) < 1.0 mmol/L. Physical inactivity was defined as engaging in less than 150 min per week of moderate-intensity physical activity based on medical records and self-reported information. Overweight or obesity was defined as a body mass index of 28 kg/m² or higher.
Patients were excluded if they had experienced an acute stroke within the previous six months. Individuals with severe life-threatening conditions involving the heart, liver, lungs, or kidneys were also excluded, as these conditions significantly alter key ALI biomarkers (e.g., serum creatinine, inflammatory indices) and independently impact cognitive function, potentially introducing confounding bias. Participants were also excluded if they were unable to cooperate with clinical or cognitive assessment. Individuals with established dementia that significantly impaired daily functioning and self-care capabilities were also excluded, defined by an MMSE score ≤ 17 or inability to independently perform basic activities of daily living.
Variables collected
A total of 32 variables were included in the analysis, encompassing demographic characteristics, anthropometric measurements, vital signs, laboratory indicators, and comorbid conditions. Demographic and anthropometric variables included age, sex, waist-to-height ratio, and BMI. Vital signs included systolic and diastolic blood pressure (SBP/DBP), heart rate (HR), and respiratory rate (RR). Anthropometric measurements, including height and weight, were directly measured by trained nursing staff using calibrated equipment at admission. BMI was calculated as weight (kg) divided by height squared (m²). Venous blood samples were collected from all participants after an overnight fast of 8–12 h. Blood samples were drawn in the morning using standardized procedures. Serum separation tubes were used for biochemical analyses, EDTA-anticoagulated tubes for hematological measurements, and sodium citrate tubes for coagulation-related assays. All samples were processed according to standard laboratory protocols at the hospital’s central laboratory. Laboratory parameters were categorized into hematological, biochemical, and coagulation indicators. Hematological parameters, including white blood cell count (WBC), red blood cell count (RBC), hemoglobin (Hb), platelet count (PLT), neutrophil count, lymphocyte ratio, and monocyte ratio, were analyzed using the Sysmex XN-9000 automated hematology analyzer. Biochemical parameters, comprising triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), fasting blood glucose, serum potassium, serum sodium, uric acid, serum creatinine, serum albumin, total bilirubin, and creatine kinase MB isoenzyme (CKMB), were measured using the Beckman AU5800 automatic biochemistry analyzer. Coagulation function, specifically D-dimer, was assessed using standard coagulation assays. All laboratory tests were conducted in the hospital’s central laboratory following strict quality control procedures in accordance with national laboratory standards (ISO 15189). Comorbid conditions included hypertension, type 2 diabetes mellitus, and coronary heart disease.
Statistical analysis
Variables with more than 20% missing value were excluded from analysis. For variables with 20% or less missingness, missing values were imputed using the K-Nearest Neighbor (KNN) algorithm (k = 5). This approach was selected due to its suitability for retrospective clinical data with moderate missingness and its ability to preserve the original distribution of variables. Multicollinearity among variables was assessed using the Variance Inflation Factor (VIF), and variables with a VIF greater than 5 were excluded.
The Allostatic Load Index was constructed using ten physiological indicators representing cardiovascular, metabolic, and inflammatory systems, including waist-to-height ratio, systolic blood pressure, diastolic blood pressure, heart rate, neutrophil count, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycated hemoglobin, and serum creatinine. Clinically validated cut-off values were applied, including systolic blood pressure of at least 140 mmHg, diastolic blood pressure of at least 90 mmHg, heart rate greater than 100 beats per minute, neutrophil count of at least 1.0 × 10⁹/L, triglycerides of at least 1.7 mmol/L, high-density lipoprotein cholesterol of 1.0 mmol/L or less, low-density lipoprotein cholesterol greater than 3.4 mmol/L, glycated hemoglobin greater than 6.0%, serum creatinine greater than 97 µmol/L, and a waist-to-height ratio greater than 0.55. Each indicator exceeding its threshold contributed one point, yielding an ALI score ranging from 0 to 10, with higher scores indicating greater cumulative physiological stress.
Restricted cubic spline analyses were used to explore nonlinear associations and identify potential threshold effects of waist-to-height ratio on cognitive impairment. Continuous variables with normal distributions were summarized as mean with standard deviation and compared using Student’s t test, while non-normally distributed variables compared using the Mann-Whitney U test or Kruskal-Wallis test. Categorical variables were expressed as counts and percentages, and analyzed using the chi-square test or Fisher’s exact test as appropriate.
Logistic regression analyses were performed to estimate odds ratios and 95% confidence interval for cognitive impairment. The unadjusted model (Model I) included only the exposure of interest, whereas the adjusted model (Model II) additionally controlled for age, respiratory rate, sex, hypertension, coronary heart disease, and type 2 diabetes mellitus.
Sensitivity analysis
Sensitivity analyses were conducted by categorizing participants into quartiles based on ALI scores. Logistic regression analyses were repeated using these ALI quartile groups, with both unadjusted (Model III) and adjusted models (Model IV) applied using the same covariates as in the primary analysis.
Predictive model development and validation
Feature selection was performed using the Boruta algorithm applied to both the continuous ALI dataset and the ALI quartile dataset to identify relevant predictors. The dataset was randomly divided into training and validation sets using a 3:7 ratio. Predictive models were developed using logistic regression (LR), decision tree (DT), K-nearest neighbors (KNN), random forest (RF), extreme gradient boosting (XGBoost), elastic net (ENET), support vector machine (SVM), and multilayer perceptron (MLP) algorithms. Five-fold cross-validation was employed to enhance model stability and reduce overfitting.
Model discrimination was evaluated using the area under the receiver operating characteristic curve (AUC). Decision curve analysis (DCA) was used to assess the clinical utility of the models, and calibration curves were generated to evaluate the agreement between predicted and observed risks. This study was conducted in accordance with the TRIPOD statement and TRIPOD-AI extension to ensure transparent reporting of prediction model development and validation[11].
Results
Baseline characteristics
Baseline characteristics of the overall cohort and stratified by cognitive impairment status are presented in Table 1. Compared with participants without cognitive impairment, individuals with cognitive impairment were significantly older, with a mean age of 72.4 ± 5.8 years versus 67.9 ± 4.5 years (P < 0.001). Individuals with cognitive impairment also exhibited higher waist-to-height ratio (0.57 ± 0.04 vs. 0.52 ± 0.03, P < 0.001), higher systolic blood pressure (146.3 ± 11.9 vs. 131.7 ± 10.2 mmHg, P < 0.001), and elevated glycated hemoglobin levels (6.7%±0.8% vs. 5.9%±0.6%, P < 0.001). In addition, the prevalence of hypertension (82.2% vs. 61.4%, P < 0.001) and type 2 diabetes (67.3% vs. 43.6%, P < 0.001) was significantly higher in the cognitive impairment group.
Table 1.
Baseline characteristics
| Characteristic | Cognitive impairment | P value | ||
|---|---|---|---|---|
| Overall | No | Yes | ||
| Age (year) | 72±8 | 70±8 | 74±8 | <0.001 |
| Waist height ratio | 0.55±0.06 | 0.53±0.05 | 0.56±0.06 | <0.001 |
| SBP (mmHg) | 137±21 | 134±22 | 140±21 | 0.025 |
| DBP (mmHg) | 80±12 | 80±12 | 81±13 | 0.789 |
| Heart rate (beats/min) | 75±14 | 74±13 | 76±15 | 0.280 |
| Respiratory rate (beats/min) | 19.28±1.32 | 19.20±0.99 | 19.36±1.58 | 0.663 |
| WBC (109 /L) | 7.08±5.52 | 6.32±2.30 | 7.83±7.38 | 0.003 |
| RBC (1012/L) | 4.46±0.69 | 4.41±0.70 | 4.50±0.68 | 0.665 |
| Hb (g/L) | 132±20 | 132±22 | 132±18 | 0.822 |
| Platelet count (109 /L) | 224±71 | 213±55 | 234±82 | 0.023 |
| Neutrophile granulocyte (109 /L) | 4.37±2.36 | 4.13±2.12 | 4.60±2.56 | 0.071 |
| Lymphocytes (109 /L) | 1.79±1.91 | 1.89±2.62 | 1.70±0.73 | 0.961 |
| Monocyte (109 /L) | 0.50±0.21 | 0.47±0.18 | 0.53±0.23 | 0.014 |
| TG (mmol/L) | 1.97±1.79 | 1.85±1.38 | 2.09±2.13 | 0.710 |
| HDL (mmol/L) | 1.22±0.31 | 1.24±0.30 | 1.19±0.32 | 0.139 |
| LDL (mmol/L) | 2.40±0.97 | 2.54±0.98 | 2.25±0.94 | 0.019 |
| Blood glucose (mmol/L) | 7.18±2.93 | 6.85±2.79 | 7.51±3.04 | 0.040 |
| HbA1c (%) | 6.34±1.00 | 6.14±0.98 | 6.53±0.98 | <0.001 |
| Potassium (mmol/L) | 4.25±2.33 | 4.47±3.27 | 4.04±0.46 | 0.083 |
| Sodium (mmol/L) | 139.7±9.9 | 139.3±13.5 | 140.1±3.8 | 0.158 |
| Uric acid (umoI/L) | 358±104 | 353±96 | 363±111 | 0.500 |
| Creatinine (umoI/L) | 84±38 | 86±46 | 82±28 | 0.910 |
| Albumin (g/L) | 41.7±5.0 | 42.3±5.5 | 41.0±4.4 | 0.159 |
| Total bilirubin (umoI/L) | 12.3±5.4 | 13.1±5.2 | 11.5±5.5 | 0.010 |
| CKMB (ug/L) | 11±8 | 12±8 | 10±7 | 0.111 |
| D-dimer (ug/mL) | 0.79±1.79 | 0.63±0.98 | 0.94±2.32 | 0.071 |
| Gender (n,%) | 0.881 | |||
| Male | 106 (53.0%) | 53 (53.5%) | 53 (52.5%) | |
| Female | 94 (47.0%) | 46 (46.5%) | 48 (47.5%) | |
| Hypertension (n,%) | 0.003 | |||
| No | 67 (33.5%) | 43 (43.4%) | 24 (23.8%) | |
| Yes | 133 (66.5%) | 56 (56.6%) | 77 (76.2%) | |
| Diabetes II (n,%) | <0.001 | |||
| No | 127 (63.5%) | 83 (83.8%) | 44 (43.6%) | |
| Yes | 73 (36.5%) | 16 (16.2%) | 57 (56.4%) | |
| Heart failure (n,%) | <0.001 | |||
| No | 104 (52.0%) | 67 (67.7%) | 37 (36.6%) | |
| Yes | 96 (48.0%) | 32 (32.3%) | 64 (63.4%) | |
Association between waist-to-height ratio and cognitive impairment
Restricted cubic spline analysis was performed to examine the nonlinear association between waist-to-height ratio and cognitive impairment after adjusting for age, systolic blood pressure, heart rate, sex, hypertension, type 2 diabetes mellitus, and coronary heart disease. As shown in Fig. 1, a positive linear association was observed between waist-to-height ratio (WHtR) and the risk of cognitive impairment. An inflection point was identified at a WHtR of 0.55, above which the odds ratio exceeded 1 and the risk of cognitive impairment increased markedly. Based on this finding, 0.55 was defined as the critical threshold for WHtR in subsequent analyses.
Fig. 1.
Restricted cubic spline analysis of WHtR and cognitive impairment risk
Association between allostatic load index and cognitive impairment
Figure 2 illustrates the distribution of the ALI, calculated using ten physiological indicators, in the total cohort and stratified by cognitive impairment status. Although the theoretical ALI range was 0–10, observed values ranged from 0 to 8, with approximately 59% of participants having an ALI score of 3. The distribution of ALI approximated normality, and individuals with cognitive impairment demonstrated higher ALI score when treated as a continuous variable.
Fig. 2.
ALI score distribution stratified by cognitive impairment status
The associations between ALI and cognitive impairment are summarized in Table 2. In univariate logistic regression analysis, each one-unit increase in ALI was associated with a 27% higher risk of cognitive impairment ( Model I: OR = 1.27, 95%CI = 1.04–1.55). This association remained statistically significant after adjustment for age, respiratory rate, sex, hypertension, coronary heart disease, and type 2 diabetes mellitus (Model II, OR = 1.33, 95%CI = 1.04–1.70).
Table 2.
Logistic regression analysis of the association between Allostatic Load Index and cognitive impairment in stroke high-risk population
| Model | OR (95% CI) | P value |
|---|---|---|
| I | 1.27 (1.04-1.55) | 0.019 |
| II | 1.33 (1.04-1.70) | 0.021 |
| III | ||
| [0,2) | Reference | |
| [2,3) | 0.98 (0.39-2.50) | 0.971 |
| [3,4) | 1.88 (0.77-4.55) | 0.164 |
| [4,8] | 2.57 (1.06-6.23) | 0.036 |
| IV | ||
| [0,2) | Reference | |
| [2,3) | 1.56 (0.49-4.97) | 0.451 |
| [3,4) | 2.46 (0.85-7.18) | 0.099 |
| [4,8] | 3.83 (1.29-11.40) | 0.016 |
Model I: AL as a continuous variable; Unadjusted OR (95%CI)
Model II: AL as a continuous variable; adjusted OR (95%CI)
Model III: AL as categorical variable; Unadjusted OR (95%CI)
Model IV: AL as categorical variable; adjusted OR (95%CI)
In sensitivity analyses, participants were categorized into quartiles based on ALI scores, with quartile 1 (ALI 0–2) serving as the reference group. Individuals in the highest quartile (ALI 4–8) exhibited a significantly increased risk of cognitive impairment in both unadjusted analysis (Model III: OR = 2.57, 95%CI = 1.06–6.23) and adjusted analysis (Model IV: OR = 3.83, 95%CI = 1.29–11.40). These findings indicate a dose–response relationship between increasing ALI and the incidence of cognitive impairment.
Predictive model development and validation
Feature selection was conducted using the Boruta algorithm, which identifies important predictors by comparing original features with randomly permuted shadow features using random forest–based importance measures [12]. Variables classified as “acceptable” were retained for model development.
When ALI was included as a continuous variable, the Boruta algorithm confirmed ALI as an acceptable predictive feature (Fig. 3). ROC curves for the predictive models are shown in Fig. 4. Among all models, the random forest model demonstrated the highest discriminative ability, with an AUC value of 0.8247. The AUC values for the decision tree, k-nearest neighbors, XGBoost, logistic regression, elastic net, support vector machine, and multilayer perceptron models were 0.6704, 0.7312, 0.7871, 0.7538, 0.7559, 0.7527, and 0.7344, respectively. Calibration curves and decision curve analyses are presented in Figure S1 and Figure S2.
Fig. 3.
Boruta-based feature importance with continuous ALI
Fig. 4.
ROC curves of machine learning models with continuous ALI
When ALI quartiles were treated as categorical variables, the Boruta algorithm again confirmed ALI group as an acceptable predictor (Fig. 5). The corresponding ROC curves are presented in Fig. 6, with logistic regression achieving the highest AUC value of 0.857. Calibration analysis demonstrated that the random forest model exhibited minimal deviation from the reference line, indicating strong agreement between predicted and observed risks (Figure S3). Decision curve analysis further showed that the logistic regression model provided the greatest net clinical benefit across a wide range of threshold probabilities, supporting its clinical applicability (Figure S4).
Fig. 5.
Boruta-based feature importance with ALI quartiles
Fig. 6.
ROC curves of machine learning models with ALI quartiles
Discussion
Chronic accumulation of significant stress and health-damaging behaviors can lead to an overload of adaptive capacity when it exceeds an individual’s coping abilities [13]. The concept of adaptive load provides a theoretical framework to elucidate the relationship between stress and the development of adaptive-related diseases [14], facilitating a better understanding of the cumulative effects of stress, adaptation, and their connection to disease [15]. When the cumulative stress surpasses an individual’s capacity to adapt, an overload of adaptive load may arise [16, 17]. Without sufficient buffering factors, or when such factors are lacking, the body’s stress response system could be repeatedly activated [18], resulting in long-term wear and tear. ALI serves to illustrate the relationship between stress and disease, with this relationship being sustained by adaptive balance, whereby organisms stabilize their internal environment through internal feedback regulation [15]. In summary, the harm inflicted by stress on the body constitutes a complex biological process and behavioral response that can facilitate the onset of chronic diseases via multiple mechanisms [19], including biological wear, aging, inflammatory responses, and oxidative stress [2]. Collectively, these factors may elevate an individual’s mortality risk. The introduction of ALI offers a more accurate representation of the relationship between stress and health, serving as a cumulative measure of physiological adaptations associated with stress [14]. The factors used in calculating the ALI come from a subset of biomarkers related to the neuroendocrine, metabolic, cardiovascular, and immune systems [20], and it is calculated by aggregating these biomarkers [1, 21]. A higher ALI signifies a dysregulation in stress adaptation, which can promote the development of chronic diseases through various mechanisms. Numerous studies indicate that the total ALI is a more effective predictor of mortality outcomes than individual biomarkers within the ALI [22–24].
This research finds an association between higher ALI and increased risk of cognitive impairment in high-risk stroke populations. Univariate analysis shows that each 1-unit increase in ALI is associated with a 27% higher risk of cognitive impairment in this group; after adjusting for age, respiratory rate, gender, hypertension, type 2 diabetes, and coronary heart disease, this association remains significant with a slight enhancement. Importantly, as this is a retrospective study, we cannot establish temporal sequence or causal relationships between ALI and cognitive impairment—only an observational association. By categorizing the ALI into quartiles, it is found that individuals in the highest quartile exhibit a 157% increase in the risk of cognitive impairment compared to those in the first quartile, and this association persists even after adjusting for covariates. Prior research has shown that in populations of stroke survivors, elevated baseline levels of ALI correlate with risks of all-cause and cardiovascular mortality [7]. This study offers new perspectives for research concerning high-risk stroke populations.
There is still no unified consensus on the best method for measuring the ALI. Currently, the calculation of ALI generally employs metrics from the cardiovascular, metabolic, and immune systems, with differences in the biomarkers included in various studies. In the NHANES and several prior studies, the metrics used to compute ALI included systolic blood pressure, diastolic blood pressure, and pulse from the cardiovascular system; C-reactive protein from the immune system; and body mass index, total cholesterol, high-density lipoprotein, glycated hemoglobin, glucose, and triglycerides from the metabolic system. The ALI derived from these metrics can effectively reflect chronic stress [4, 25, 26]. A critical methodological innovation of our study is the first-time inclusion of waist-to-height ratio (WHtR) in the ALI calculation framework. Unlike most prior ALI studies— which primarily relied on BMI to assess body composition, a metric that fails to capture abdominal fat distribution. The accumulation of abdominal fat is linked to a heightened risk of cardiovascular metabolic complications and is a more robust determinant of disease development compared to BMI [27]. Utilizing the new ALI evaluation system established in this study, a significant correlation has been found between ALI and the incidence of cognitive impairment among stroke high-risk populations, demonstrating strong predictive ability. Additionally, the same metric may not hold the same significance across various populations and ethnic groups. Thus, we suggest that future research teams consider tailoring the selection of indicators used to calculate ALI according to the characteristics of the study population.
An earlier systematic review has validated the association between ALI and specific brain structures and functions. Furthermore, brain reserve can help prevent or manage significant brain alterations resulting from aging or disease, increasing the threshold for the emergence of cognitive impairment or functional decline symptoms [28]. The study results reveal that areas particularly vulnerable to chronic physiological stress, as assessed by the ALI, differ among various populations. In individuals suffering from learning and memory impairments, along with dysregulated neuroendocrine activity, alternative and retrospective measures of chronic stress have been linked to a decrease in hippocampal volume [29]. In older adults and male participants, the hippocampus and white matter volume are influenced; in contrast, in individuals with schizophrenia spectrum disorders, the cortex, fornix, and hippocampus are affected [9]. This indicates that chronic stress can lead to hippocampal alterations in different populations. Further research has identified potential mechanisms including neurodegeneration, vascular changes, inflammation, and oxidative stress [9]. In patients with schizophrenia, elevated ALI significantly affects cortical thickness, leading to structural and cognitive deficits in the brain [30]. Among the six cortical regions significantly associated with ALI, five are located in the prefrontal cortex, which may be related to the effects of chronic stress [30].
This research, being a single-center retrospective study, has several key limitations. First, retrospective design inherently carries selection bias and unmeasured confounders, which may distort the ALI-cognitive impairment association. Second, missing data (handled via KNN imputation) is a major limitation—though KNN is suitable for moderate missingness, it cannot fully eliminate bias from unrecorded values. Third, the cohort was recruited exclusively from Shenzhen, restricting generalizability. The cross-sectional nature of data only captures a snapshot of ALI and cognitive status, precluding definitive conclusions about the causal direction between them. Our study did not include social determinants of health, such as socioeconomic status and social support, which are known to influence stress levels and may affect ALI and cognitive impairment [31]. Furthermore, due to the retrospective nature of the study and heterogeneity in medication records, detailed medication stratification was not feasible, potentially overlooking the influence of specific drug therapies. Finally, cognitive function was assessed solely using the Chinese version of the MMSE. While widely used in clinical practice in China, we acknowledge that it it is a screening tool rather than a diagnostic instrument. Furthermore, as this was a retrospective study, we utilized the clinically recorded MMSE scores, which may lack the granularity of prospective cognitive testing batteries. Future longitudinal studies utilizing multi-domain cognitive assessments are warranted to validate our findings.
The findings of this study may have practical implications for integrating ALI screening into stroke prevention programs. First, ALI can be incorporated into routine health assessments for high-risk stroke populations. Given that ALI is calculated using 10 easily measurable indicators that are already part of standard clinical tests, screening can be implemented without additional resource burden. Second, ALI stratification might help guide targeted interventions: individuals with higher ALI may benefit from lifestyle modifications such as diet control and exercise to reduce WHtR, which can help reduce ALI and potentially mitigate the risk of cognitive impairment. Third, ALI can serve as a dynamic monitoring indicator in long-term stroke prevention: repeated ALI measurements can track changes in chronic stress load, allowing clinicians to adjust intervention plans in a timely manner and optimize prevention efficacy.
Conclusion
This study demonstrates that a higher ALI is independently associated with an increased risk of cognitive impairment in individuals at high risk for stroke. Incorporating waist-to-height ratio into ALI framework may enhance its ability to capture stress-related physiological damage in this population. Although causality cannot be inferred from this retrospective analysis, the findings support the potential value of ALI as a quantitative tool for assessing cumulative chronic stress and identifying individuals at elevated risk of cognitive decline. Prospective, longitudinal studies are needed to confirm these associations and to evaluate whether ALI-guided interventions can effectively reduce cognitive impairment in stroke high-risk populations.
Supplementary Information
Acknowledgements
None.
Authors’ contributions
JLZ contributed to the study concept and study design. FJY and XHC performed statistical analysis and data interpretation. LWN and HFN were responsible for the quality control of data and algorithms. XJ and HDS performed literature research and data extraction. All authors contributed to writing of the manuscript and approved the final version.
Funding
This study was supported by the Scientific Research Projects of Medical and Health Institutions of Longhua District, Shenzhen (2021057); The Key Laboratory of Personalized Precision Treatment for Elderly Coronary Heart Disease, Longhua District, Shenzhen(Shen Long Hua Ke Chuang Ke Ji Zi (2024) No. 2); the Key Medical Discipline Construction of Shenzhen Longhua District.
Data availability
The original contributions presented in this study are included in the article, further inquiries can be directed to the corresponding author.
Declarations
Ethics statement and consent to participate
This study was reviewed and approved by the Medical Ethics Committee of Shenzhen Longhua District Central Hospital (Approval No.: 2020-124-01) and complies with the ethical guidelines set forth in the World Medical Association’s Declaration of Helsinki. Since our study included retrospective cases, there was no additional intervention, and all patient information was desensitized, and the need for written informed consent was waived.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Fengjuan Yan and Lvwen Ning contributed equally to this work.
References
- 1.Juster RP, McEwen BS, Lupien SJ. Allostatic load biomarkers of chronic stress and impact on health and cognition. Neurosci Biobehav Rev. 2010;35(1):2–16. [DOI] [PubMed] [Google Scholar]
- 2.Parker HW, et al. Allostatic load and mortality: A systematic review and Meta-Analysis. Am J Prev Med. 2022;63(1):131–40. [DOI] [PubMed] [Google Scholar]
- 3.Borrell LN, Rodriguez-Alvarez E, Dallo FJ. Racial/ethnic inequities in the associations of allostatic load with all-cause and cardiovascular-specific mortality risk in U.S. Adults. PLoS ONE. 2020;15(2):e0228336. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Rodriquez EJ, et al. Allostatic load: Importance, Markers, and score determination in minority and disparity populations. J Urban Health. 2019;96(Suppl 1):3–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Caraballo C, et al. Excess mortality and years of potential life lost among the black population in the US, 1999–2020. JAMA. 2023;329(19):1662–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Guan Y, et al. Association between allostatic load and breast cancer risk: a cohort study. Breast Cancer Res. 2023;25(1):155. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Johnson NB, et al. Effects of allostatic load on Long-Term survival after stroke. Stroke. 2025;56(1):87–94. [DOI] [PubMed] [Google Scholar]
- 8.McEwen BS. Protective and damaging effects of stress mediators: central role of the brain. Dialogues Clin Neurosci. 2006;8(4):367–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Lenart-Bugla M, et al. The association between allostatic load and brain: A systematic review. Psychoneuroendocrinology. 2022;145:105917. [DOI] [PubMed] [Google Scholar]
- 10.Zhang X, et al. High-risk population and factors of stroke has changed among middle-aged and elderly Chinese-Evidence from 1989 to 2015. Front Public Health. 2023;11:1090298. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Collins GS, Moons KGM, Dhiman P, et al. TRIPOD + AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Yan F, et al. Association between the stress hyperglycemia ratio and 28-day all-cause mortality in critically ill patients with sepsis: a retrospective cohort study and predictive model establishment based on machine learning. Cardiovasc Diabetol. 2024;23(1):163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Lucente M, Guidi J. Allostatic load in children and adolescents: A systematic review. Psychother Psychosom. 2023;92(5):295–303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.McEwen BS, Stellar E. Stress and the individual. Mechanisms leading to disease. Arch Intern Med. 1993;153(18):2093–101. [PubMed] [Google Scholar]
- 15.Sterling P. Allostasis: a model of predictive regulation. Physiol Behav. 2012;106(1):5–15. [DOI] [PubMed] [Google Scholar]
- 16.McEwen BS. Physiology and neurobiology of stress and adaptation: central role of the brain. Physiol Rev. 2007;87(3):873–904. [DOI] [PubMed] [Google Scholar]
- 17.Fava GA, et al. Clinical characterization of allostatic overload. Psychoneuroendocrinology. 2019;108:94–101. [DOI] [PubMed] [Google Scholar]
- 18.McEwen BS, Wingfield JC. What is in a name? Integrating homeostasis, allostasis and stress. Horm Behav. 2010;57(2):105–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.McEwen BS, Seeman T. Protective and damaging effects of mediators of stress. Elaborating and testing the concepts of allostasis and allostatic load. Ann N Y Acad Sci. 1999;896:30–47. [DOI] [PubMed] [Google Scholar]
- 20.Seeman TE, et al. Price of adaptation–allostatic load and its health consequences. MacArthur studies of successful aging. Arch Intern Med. 1997;157(19):2259–68. [PubMed] [Google Scholar]
- 21.Juster RP, et al. Allostatic load and comorbidities: A mitochondrial, epigenetic, and evolutionary perspective. Dev Psychopathol. 2016;28(4pt1):1117–46. [DOI] [PubMed] [Google Scholar]
- 22.Castagne R, et al. Allostatic load and subsequent all-cause mortality: which biological markers drive the relationship? Findings from a UK birth cohort. Eur J Epidemiol. 2018;33(5):441–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hwang AC, et al. Predicting all-cause and cause-specific mortality by static and dynamic measurements of allostatic load: a 10-year population-based cohort study in Taiwan. J Am Med Dir Assoc. 2014;15(7):490–6. [DOI] [PubMed] [Google Scholar]
- 24.Seeman TE, et al. Allostatic load as a marker of cumulative biological risk: MacArthur studies of successful aging. Proc Natl Acad Sci U S A. 2001;98(8):4770–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Seeman T, et al. Education, income and ethnic differences in cumulative biological risk profiles in a National sample of US adults: NHANES III (1988–1994). Soc Sci Med. 2008;66(1):72–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Rodriquez EJ, et al. Relationships between allostatic load, unhealthy behaviors, and depressive disorder in U.S. adults, 2005–2012 NHANES. Prev Med. 2018;110:9–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Busetto L, et al. A new framework for the diagnosis, staging and management of obesity in adults. Nat Med. 2024;30(9):2395–9. [DOI] [PubMed] [Google Scholar]
- 28.Pettigrew C, Soldan A. Defining cognitive reserve and implications for cognitive aging. Curr Neurol Neurosci Rep. 2019;19(1):1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Gianaros PJ, et al. Prospective reports of chronic life stress predict decreased grey matter volume in the hippocampus. NeuroImage. 2007;35(2):795–803. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Zhou Y, et al. Allostatic load effects on cortical and cognitive deficits in essentially Normotensive, Normoweight patients with schizophrenia. Schizophr Bull. 2021;47(4):1048–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Allen CL, Bayraktutan U. Oxidative stress and its role in the pathogenesis of ischaemic stroke. Int J Stroke. 2009;4(6):461–70. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The original contributions presented in this study are included in the article, further inquiries can be directed to the corresponding author.