Allostatic load associated with all-cause and cause-specific mortality: a large prospective cohort study Authors: Heng-Min Xu, Zong-Chao Liu, Xin-Ling Wang, Wen-Jing Zhao, Chao Zhang, Yang Zhang, Tong Zhou, Jing-Ying Zhang, Kai-Feng Pan, Wen-Qing Li Published: 2026 ABSTRACT AbstractBackgroundAllostatic load (AL) is a composite measure of cumulative physiological burden from chronic stress, quantifying multisystem dysregulation. Whether AL may alter the risk of overall, cause-specific mortality, and premature mortality remains unclear.MethodsWe analyzed UK Biobank participants aged 40–69 years (mean 56.6; 53.8% female) with complete data of AL metrics. AL was assessed by integrating 12 commonly applied metrics covering cardiovascular, metabolic, immune, and kidney systems (score range: 0–12), with higher scores indicating greater physiological dysregulation. Associations of AL scores with mortality from all cause and cause-specific mortality were examined. Stratified analyses assessed effect modification by premature mortality (< 70 years at death) and age at recruitment.ResultsOver a median follow-up of 15.4 years, 39,804 deaths were recorded, including 14,852 from cancer, 7118 from cardiovascular disease, 1938 from respiratory disease, and 1521 from digestive disease. Higher AL scores were significantly associated with increased risks of all-cause and cause-specific mortality in a dose-response manner (all P-trend < 0.001). Among specific causes, each one-point rise in AL score yielded stronger associations with digestive disease mortality (subdistribution hazard ratio [sHR] = 1.25, 95% confidence interval [CI]: 1.21–1.29) and cardiovascular disease mortality (sHR = 1.20, 95% CI: 1.18–1.21) than for cancer mortality (sHR = 1.08, 95% CI: 1.07–1.09) and respiratory disease mortality (sHR = 1.11, 95% CI: 1.09–1.14). Age significantly modified these associations (all P-interaction < 0.01), with higher HRs for premature mortality (age at death < 70 years) or younger individuals (age at enrollment < 50 years).ConclusionsHigher AL scores were associated with increased risks of all-cause and cause-specific mortality. Such associations were particularly pronounced for digestive and cardiovascular disease mortality, and appeared stronger for premature deaths and among younger individuals. These findings suggest that AL may serve as a marker of early physiological dysregulation and may help identify individuals at elevated risk of premature mortality.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12889-026-28127-0.Keywords: Allostatic load, Chronic stress, Mortality, Premature mortality, Non-communicable diseases, Risk stratification, Preventive medicine Introduction The global burden of mortality has become increasingly dominated by non-communicable diseases (NCDs), particularly cardiovascular disease (CVD), cancer, respiratory disease, and digestive disease [1, 2]. In 2021, NCDs accounted for at least 43 million deaths globally, including over 18 million premature deaths before age 70 [2]. Premature mortality is a core population-health metric because it captures avoidable early loss of life and is therefore widely used to monitor progress in NCD prevention and control [3]. Epidemiological studies have shown that premature death arises from a complex interplay of socioeconomic conditions, behavioral risk factors, environmental exposures, and biological susceptibility [4]. Although premature mortality from cancer is declining in high-income countries [5], several NCDs—such as colorectal cancer [6, 7], type 2 diabetes [8, 9], and heart failure [10]—are showing trends toward earlier onset and younger-age mortality. These shifting mortality patterns raise concerns about the long-term public health impact and underscore the need to identify individuals at elevated risk of early death. However, scalable indicators capable of capturing cumulative physiological burden before clinically manifest disease remain limited. Chronic exposures, such as unhealthy behaviors, sustained psychosocial stress and socioeconomic disadvantage, can lead to prolonged activation of the body’s adaptive regulatory systems, a process known as allostasis [11–13]. The cumulative physiological cost of maintaining allostasis under chronic stressors is conceptualized as allostatic load (AL), which can be quantified through a composite index of biomarkers across neuroendocrine, metabolic, immune, kidney, and cardiovascular systems [12, 14]. As an integrative measure of multisystem dysregulation, elevated AL could plausibly indicate reduced resilience and impaired adaptive capacity, thereby contributing to increased risk of mortality and premature death [15]. Although several prospective studies have examined associations between AL measures and mortality, evidence remains more consistent for all-cause mortality than for multiple cause-specific endpoints, partly due to limited numbers of cause-specific deaths in many cohorts [16–19]. Additionally, a few studies have specifically assessed the association between AL and mortality among specific clinical populations, such as patients diagnosed with breast or colorectal cancer [20, 21], but the generalizability of the findings to the broader, initially healthy population remains uncertain. Moreover, existing studies have rarely performed age-stratified analyses, and many cohorts have had relatively few deaths occurring at younger ages, limiting the ability to robustly evaluate premature mortality and age-specific associations. Given the growing public health importance of premature mortality and the limited evidence on its relationship with AL, further investigation in large population-based cohorts is warranted. Leveraging the large-scale UK Biobank (UKB) cohort with long-term follow-up, this study aimed to examine the associations between AL levels, constructed from 12 biologically relevant and commonly applied metrics, and mortality in the general population. Within a single prospective population-based framework, we evaluated a broad spectrum of mortality outcomes, including all-cause mortality, major categories of cause-specific mortality, and several specific diseases. Build on this, we further examine age-specific patterns of the associations, with a particular emphasis on premature deaths and mortality risk across different age groups. Methods Study population The current study was conducted based on the UKB (https://www.ukbiobank.ac.uk/). The UKB is a population-based prospective cohort study comprising over 500,000 individuals aged 40–69 years, who were recruited between 2006 and 2010 [22, 23]. Details of the study design and population have been described previously [22, 23]. The UKB received ethical approval from North-West Multi-center Research Ethics Committee (reference 11/NW/0382) and all participants provided written informed consent. Through an approved application (Application No. 90999), we have obtained access to all individual-level data of study participants, and have acquired a complete dataset for our research, covering demographics, environmental and lifestyle factors, medical history, biomarker measurements, linked longitudinal follow-up data, among other variables available in UKB. We then made subsequent selections of eligible participants according to prespecified eligibility criteria. In this study, we excluded individuals with missing data on any of the 12 metrics used to construct the AL index (n = 122,533), as the cumulative AL score depends on the full set of prespecified components and incomplete indicator panels would reduce comparability across individuals. Across the 12 AL metrics, the proportion of missing values ranged from 0.45% (waist-to-hip ratio) to 14.45% (HDL cholesterol). Participants who died within the first year of follow-up (n = 670) were also excluded to reduce potential reverse causation, leaving a total of 378,785 eligible participants (Fig-S1). Specifically, participants with relevant disease status at cohort baseline were additionally excluded from the corresponding cause-specific mortality analyses, including 13,141 participants with myocardial infarction, angina or stroke for CVD mortality (final n = 365,644); 27,956 with any cancer for cancer mortality (final n = 350,829); 2287 with chronic obstructive pulmonary disease (COPD) or emphysema for respiratory disease mortality (final n = 376,498); and 623 with liver failure, cirrhosis or alcoholic liver disease for digestive disease mortality (final n = 378,162). Assessment on allostatic load In this study, AL was operationalized as a cumulative index of physiological dysregulation derived from multiple biomarkers. Because there is no universally accepted formulation of AL, it often varies across studies depending on biological focus and data availability [15]. In general, biomarker-based AL indices are constructed to capture multisystem physiological dysregulation across major physiological domains such as cardiovascular, metabolic, immune, and organ-function systems. Consistent with this framework, we selected 12 biomarkers representing these major physiological systems based on their biological relevance, frequent use in previous AL studies, and availability in the UKB dataset [15, 20, 21, 24]. Specifically, the selected biomarkers covered allostatic regulation of the cardiovascular (pulse rate, systolic blood pressure (SBP), and diastolic blood pressure (DBP)), metabolic (glycated hemoglobin (HbA1c), albumin, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), triglycerides (TG), and waist-to-hip ratio (WHR)), immune (C-reactive protein (CRP) and insulin-like growth factor-1 (IGF-1)), and kidney systems (serum creatinine). Each metric was categorized into quartiles based on its distribution in the overall participants, and a score of 1 was assigned for values in the highest quartile (≥ 75th percentile), except for HDL-C, albumin, and IGF-1, for which a score of 1 was assigned for the lowest quartile (≤ 25th percentile); all other values were scored as 0. The definition of the high-risk quartile for each individual biomarker is shown in Table-S1. Scores for 12 metrics were summed to generate a total AL score ranging from 0 to 12, with higher scores reflecting greater multisystem physiological dysregulation. Assessment of outcomes During the follow-up, data on mortality were obtained through linkage with national death registries in the UKB, from NHS England (for England and Wales) and from the NHS Central Register, part of the National Records of Scotland (for Scotland) [22]. The International Classification of Diseases, 10th Revision (ICD-10) codes were used to identify the causes of death. The primary outcomes of our study were all-cause mortality and cause-specific mortality. For cause-specific analyses, we focused on four leading causes to NCD mortality [1, 2]: CVD (I00-I99), cancer (C00-D48), respiratory disease (J30-J98, with exception of J69 and J96), and digestive disease (K00-K93). In addition, mortality from major individual diseases were evaluated as secondary outcomes based on narrower ICD-10 groupings, such as C00-C14 (oral and pharyngeal cancer), C16 (gastric cancer), I21 (acute myocardial infarction), J41-J44 (COPD), and K70 (alcoholic liver disease). Statistical analyses To address concerns regarding selection bias, we compared the distribution of major demographic and lifestyle characteristics, as well as the overall and cause-specific mortality rate for the overall UKB cohort, included participants of the current study, and excluded participants due to missing AL metrics. Using the mortality linkage data provided by the UKB, the person-years of follow-up was calculated from the date of participant recruitment (between 2006 and 2010) until the date of death, loss to follow-up, or September 1st, 2024, whichever occurred first. We examined the associations of the AL scores with all-cause and cause-specific mortality due to cancer, CVD, respiratory disease, or digestive disease. The AL score was classified into eight categories (0, 1, 2, 3, 4, 5, 6, and ≥ 7), with participants scoring 0 serving as the reference group. The highest category (≥ 7) combined scores that individually represented fewer than 5% of the study population. For each disease category, we further considered specific disease cause with at least 50 events. This empirical threshold was applied to reduce highly unstable estimates for relatively rare outcomes in subtype-specific analyses, as done in previous studies [25–27]. For all-cause mortality, Cox proportional hazards models were used to estimate the age- and sex-adjusted, as well as multivariate-adjusted hazard ratios (HRs) and 95% confidence intervals (CIs). The multivariable models therefore adjusted for age, sex, smoking status, alcohol consumption, Townsend deprivation index, household income, educational attainment, ethnicity, medication use, status of healthy diet, and physical activity (Supplementary Methods). Covariates were selected a priori based on established epidemiological evidence linking sociodemographic characteristics, lifestyle behaviors, and baseline health status to both AL and mortality risk [14, 19, 21, 24]. An indicator variable was created for missing values of each covariate. For cancer-specific mortality, family history of cancer in first-degree relatives was additionally adjusted for. For cause-specific mortality, Fine-Gray subdistribution hazard models were implemented to account for competing risks from other causes of death, yielding age- and sex-adjusted and multivariate-adjusted subdistribution hazard ratios (sHRs). The proportional hazards assumption was evaluated using Schoenfeld residual tests for Cox models and by examining time-by-covariate interactions in Fine-Gray models. When evidence of non-proportionality was detected, stratified analyses were conducted by follow-up time intervals. Dose-response relationships were assessed using the continuous measure of AL scores. We calculated the HRs (95% CIs) and P-values for trend (P-trend) per 1-unit increase in AL score, consistent with its nature as a count-based index and to improve interpretability. Restricted cubic splines were employed to examine the potential non-linear associations. For stratified analyses, AL scores were grouped as 0–1, 2–4, and ≥ 5 (reference: 0–1) to ensure adequate numbers of events and stable estimates within strata. Subgroup analyses were performed to examine the associations of AL scores with mortality risk by age at deaths in two (premature mortality, defined as death occurring before 70 years [2, 5, 28] or non-premature mortality) or five categories (< 50, 50–59, 60–69, 70–79, or ≥ 80 years). Stratified analyses were also conducted by age at participant enrollment or other stratum defined by major covariates. Likelihood ratio tests were used to calculate the P values for interactions (P-interaction) between AL scores and the subgroup variables. Sensitivity analyses were performed by removing one covariate at a time from the fully adjusted models to test the robustness of the results. Other sensitivity analyses were conducted to assess the robustness of findings, including the exclusion of deaths within the first two years of follow-up to mitigate reverse causality, the exclusion of deaths occurring post-COVID-19 pandemic onset to account for the potential period effects, as well as the exclusion of individuals with regular medication use or chronic conditions at baseline to partly address the concern on residual confounding. In addition, sensitivity analyses were also conducted by restricting to participants of non-White ancestry. All analyses were conducted using R version 4.4.2. Statistical significance was defined as the two-sided P < 0.05. Results Baseline characteristics This study included 378,785 participants for all-cause mortality analysis (mean age: 56.6 years; 53.8% female). Baseline characteristics across categories of AL scores are summarized in Table 1. Those with a higher AL score were more likely to be older and males, and tend to have lower educational attainment, reduced household income, and greater socioeconomic deprivation. Individuals in higher AL groups also demonstrated a higher prevalence of behavioral and clinical risk factors, such as smoking, medication use, poor diet status and insufficient physical activity. Baseline characteristics of the analytic sample were generally similar across the analytic sample and other participants in the UKB cohort, although mortality appeared higher among the excluded participants (Table-S2). Table 1Baseline characteristics of study participants by scores of allostatic load based on the UK BiobankCharacteristicOverallScores of allostatic load0123456≥ 7N378,78536,12464,80572,38166,92454,41939,31524,55320,264Age, year, mean (SD)56.6 (8.1)51.7 (7.8)54.3 (8.0)56.4 (8.0)57.5 (7.9)58.2 (7.7)58.6 (7.5)58.8 (7.4)59.0 (7.3)Townsend deprivation index, mean (SD)-1.34 (3.1)-1.55 (2.9)-1.54 (2.9)-1.48 (3.0)-1.39 (3.0)-1.28 (3.1)-1.11 (3.2)-0.93 (3.2)-0.73 (3.3)Female, %53.877.868.660.451.844.338.333.027.2Ethnicity, White, %88.688.788.688.888.688.488.388.288.4Education, college or university degree or equivalent, %33.044.139.535.431.729.226.324.722.2Annual household income ≥£52,000/year, %22.132.927.723.320.718.616.714.913.7Current smoking, %10.57.98.39.410.311.312.514.015.8Alcohol intake of ≥ 3 times/week, %43.646.845.944.943.842.641.239.637.9Current medication use, %21.317.418.419.821.523.224.425.626.2Health diet status, poor, %39.231.433.635.939.242.444.948.552.0Physical activity level, low %14.311.412.012.713.915.317.018.721.4Family history of cancer in first degree relatives, %35.032.934.535.335.335.235.935.635.3 Baseline characteristics of study participants by scores of allostatic load based on the UK Biobank Associations of AL scores with risk of all-cause mortality and cause-specific mortality from major disease categories During a median follow-up of 15.4 years (IQR: 14.6–16.1 years), we documented 39,804 all-cause deaths, among which 14,852 cancer deaths, 7118 CVD deaths, 1938 respiratory disease deaths and 1521 digestive disease deaths were identified. Participants who died during follow-up had a higher AL score compared to those who remained alive (Table-S3). A higher AL score was associated with elevated risks of all-cause mortality (Table 2). Compared with individuals having a score of 0, the multivariate-adjusted HR (95% CI) for all-cause mortality was 1.12 (1.05–1.19) for a score of 1 and 2.17 (2.04–2.31) for scores ≥ 7, displaying a clear upward trend in mortality with increasing AL scores (HR = 1.11, 95% CI: 1.11–1.12 per one point of score increase, P-trend < 0.001, Table 2). Restricted cubic spline models revealed no significant deviation from linearity of the associations (Fig-S2). Table 2Association of scores of allostatic load with all-cause and cause-specific mortality based on the UK BiobankaNo. of participantsNo. of deathsCrude mortality (per 100,000 person-years)Age- and sex-adjusted sHR/HR (95% CI)Multivariate-adjusted sHR/HR (95% CI)bAll-cause mortality Score categories 036,1241359246.2RefRef 164,8053681373.91.16 (1.09–1.23)1.12 (1.05–1.19) 272,3815824533.91.35 (1.27–1.43)1.25 (1.17–1.32) 366,9246989698.71.58 (1.49–1.68)1.40 (1.32–1.48) 454,4196896855.31.79 (1.68–1.90)1.52 (1.43–1.61) 539,31561121060.42.13 (2.01–2.26)1.72 (1.62–1.83) 624,55344661254.42.45 (2.30–2.61)1.90 (1.78–2.02) ≥ 720,26444771548.62.97 (2.79–3.16)2.17 (2.04–2.31) Per one point of score increase378,78539,804703.81.17 (1.16–1.17)1.11 (1.11–1.12) P-trend< 0.001< 0.001Cancer mortalityc Score categories 034,244658125.5RefRef 160,7431668180.41.13 (1.04–1.24)1.11 (1.01–1.22) 267,1732360232.31.22 (1.12–1.33)1.16 (1.06–1.26) 361,6292617282.81.34 (1.23–1.46)1.24 (1.14–1.35) 449,9102492335.21.47 (1.35–1.61)1.32 (1.21–1.45) 536,1892180408.51.72 (1.57–1.88)1.50 (1.37–1.64) 622,4011477451.61.84 (1.67–2.02)1.55 (1.41–1.71) ≥ 718,5401400525.02.06 (1.87–2.27)1.67 (1.52–1.84) Per one point of score increase350,82914,852282.31.11 (1.10–1.12)1.08 (1.07–1.09) P-trend< 0.001< 0.001Cardiovascular disease mortality Score categories 035,69614426.4RefRef 163,55047949.61.41 (1.17–1.70)1.36 (1.13–1.64) 270,46689083.71.90 (1.59–2.26)1.75 (1.46–2.08) 364,5031192123.42.44 (2.05–2.91)2.15 (1.81–2.56) 451,9581264163.72.93 (2.46–3.49)2.49 (2.09–2.96) 537,2541187216.43.62 (3.04–4.32)2.94 (2.46–3.51) 623,120906268.84.27 (3.57–5.11)3.30 (2.76–3.96) ≥ 719,0971056384.65.81 (4.86–6.94)4.26 (3.56–5.10) Per one point of score increase365,6447118130.01.25 (1.24–1.27)1.20 (1.18–1.21) P-trend< 0.001< 0.001Respiratory disease mortality Score categories 036,089417.4RefRef 164,67512012.21.18 (0.83–1.69)1.10 (0.77–1.57) 272,14124422.41.69 (1.21–2.36)1.41 (1.01–1.97) 366,54635035.22.30 (1.65–3.19)1.76 (1.27–2.45) 453,99735744.62.62 (1.89–3.65)1.85 (1.33–2.58) 538,92333358.33.20 (2.30–4.46)2.07 (1.48–2.88) 624,22024469.33.63 (2.59–5.10)2.13 (1.52-3.00) ≥ 719,90724987.44.34 (3.09–6.09)2.29 (1.63–3.23) Per one point of score increase376,498193834.41.22 (1.19–1.25)1.11 (1.09–1.14) P-trend< 0.001< 0.001Digestive disease mortality Score categories 036,116285.1RefRef 164,76710110.31.63 (1.07–2.48)1.54 (1.01–2.35) 272,29814213.01.75 (1.16–2.63)1.54 (1.03–2.32) 366,81425625.63.11 (2.09–4.62)2.59 (1.75–3.85) 454,29026232.63.67 (2.46–5.46)2.90 (1.95–4.32) 539,20926746.44.98 (3.34–7.42)3.68 (2.47–5.49) 624,48223064.76.70 (4.48–10.02)4.66 (3.11–6.98) ≥ 720,18623581.58.08 (5.40-12.11)5.22 (3.47–7.84) Per one point of score increase378,162152126.91.33 (1.29–1.37)1.25 (1.21–1.29) P-trend< 0.001< 0.001Abbreviations: AL Allostatic load, CI Confidence interval, HR Hazard ratio, sHR Subdistribution hazard ratioa Cox proportional hazards models were used for all-cause mortality, while Fine-Gray subdistribution hazard models were used for cause-specific mortality to account for competing risksb Adjusted for age, sex, smoking status (never, former, or current), frequency of alcohol consumption (never or occasionally, < 3 times/week, or ≥ 3 times/week), Townsend Deprivation Index, annual household income (<£18,000, £18,000–30,999, £31,000–51,999, or ≥£52,000), education level (basic education or below, vocational or A/AS-level equivalent, or college, university degree or equivalent), ethnicity (White or non-White), medication use (never or current), status of healthy diet (poor, fair, or ideal), and physical activity level (low, moderate, or high)c Additionally adjusted for family history of cancer among first-degree relatives Association of scores of allostatic load with all-cause and cause-specific mortality based on the UK Biobanka Abbreviations: AL Allostatic load, CI Confidence interval, HR Hazard ratio, sHR Subdistribution hazard ratio a Cox proportional hazards models were used for all-cause mortality, while Fine-Gray subdistribution hazard models were used for cause-specific mortality to account for competing risks b Adjusted for age, sex, smoking status (never, former, or current), frequency of alcohol consumption (never or occasionally, < 3 times/week, or ≥ 3 times/week), Townsend Deprivation Index, annual household income (<£18,000, £18,000–30,999, £31,000–51,999, or ≥£52,000), education level (basic education or below, vocational or A/AS-level equivalent, or college, university degree or equivalent), ethnicity (White or non-White), medication use (never or current), status of healthy diet (poor, fair, or ideal), and physical activity level (low, moderate, or high) c Additionally adjusted for family history of cancer among first-degree relatives AL scores were also significantly associated with cause-specific mortality in a dose-response manner, with higher sHR estimates observed for CVD (1.20, 95% CI: 1.18–1.21) and digestive disease mortality (1.25, 95% CI: 1.21–1.29) than for cancer (1.08, 95% CI: 1.07–1.09) and respiratory disease mortality (1.11, 95% CI: 1.09–1.14) per one-point increase in AL score. Because the proportional hazards assumption was not satisfied for CVD and respiratory disease mortality (P < 0.05; Table-S4), follow-up time-stratified Fine-Gray models were further applied, which yielded consistent associations for a follow-up time < 10 years or ≥ 10 years (Table-S5). These associations remained similar when AL was categorized into quartiles (Table-S6). Analysis of AL components revealed that each of the four components were significantly associated with both all-cause and cause-specific mortality, with the metrics related to inflammatory and kidney function exhibiting the strongest associations for all outcomes (Fig-S3). Associations by age at death and age at participant enrollment We examined AL scores associated with mortality stratified by age at death. Cumulative risk of mortality curves began to diverge by age 60, and the separation became pronounced by age 70, with higher scores showing progressively greater risks (Fig-S4). By age 85, the cumulative risk of mortality was 25% for score 0 and 49% for score 7, highlighting a marked absolute risk difference. The association with all-cause mortality was consistently stronger for premature deaths (< 70 years) than for deaths at ≥ 70 years (P-interaction < 0.001; Table 3 and Table-S7); the HR (95% CI) per one point of score was 1.17 (1.16–1.18) and 1.10 (1.09–1.11) respectively. Further stratified analysis by 10-year intervals of age at death corroborated the findings, which revealed a trend towards stronger associations for deaths at a younger age group (Table 3). A comparable shift in the risk profile was also noted for mortality from all major disease categories, with a more pronounced association estimate observed for deaths occurring at a younger age (Table 3 and Table-S7). The per-point-of-score sHR (95% CI) for premature vs. later deaths was 1.10 (1.09–1.12) vs. 1.06 (1.06–1.07) for cancer, 1.28 (1.25–1.30) vs. 1.16 (1.15–1.18) for CVD, 1.20 (1.14–1.26) vs. 1.09 (1.06–1.12) for respiratory disease, and 1.32 (1.26–1.38) vs. 1.22 (1.17–1.26) for digestive disease (all P-interaction < 0.05). Table 3Associations between allostatic load score and mortality risk stratified by age at death based on the UK BiobankAge at deathaNo. of participantsNo. of deathsCrude mortality (per 100,000 person-years)sHR/HR (95% CI)bP-interactionAll-cause mortality Premature mortality352,90713,926259.71.17 (1.16–1.18)< 0.001 < 50339,3653847.31.19 (1.12–1.25) 50–59344,205298055.91.20 (1.18–1.22) 60–69341,96110,562200.11.17 (1.15–1.18) Non-premature mortality364,85925,878466.31.10 (1.09–1.11) 70–79349,54320,654386.81.11 (1.10–1.12) ≥ 80359,635522495.41.08 (1.07–1.10)Cancer mortality Premature mortality342,0366059117.31.10 (1.09–1.12)< 0.001 < 50336,0961192.31.12 (1.01–1.24) 50–59337,288131125.61.14 (1.10–1.17) 60–69340,606462989.81.10 (1.08–1.12) Non-premature mortality344,7708793168.61.06 (1.06–1.07) 70–79343,5117534145.01.07 (1.05–1.08) ≥ 80337,236125924.51.06 (1.02–1.09)Cardiovascular disease mortality Premature mortality360,916239044.11.28 (1.25–1.30)< 0.001 < 50358,574480.91.38 (1.18–1.62) 50–59359,0204949.11.32 (1.26–1.39) 60–69360,374184834.11.27 (1.23–1.30) Non-premature mortality363,254472886.71.16 (1.15–1.18) 70–79362,204367867.61.17 (1.15–1.20) ≥ 80359,576105019.41.14 (1.10–1.18)Respiratory disease mortality Premature mortality375,0404808.61.20 (1.14–1.26)0.0048 < 50374,56660.11.28 (0.70–2.32) 50–59374,623631.11.24 (1.08–1.43) 60–69374,9714117.31.19 (1.13–1.26) Non-premature mortality376,018145825.91.09 (1.06–1.12) 70–79375,686112620.01.08 (1.05–1.12) ≥ 80374,8923325.91.12 (1.05–1.20)Digestive disease mortality Premature mortality377,1915509.81.32 (1.26–1.38)< 0.001 < 50373,736160.31.33 (0.99–1.80) 50–59373,8231031.81.32 (1.18–1.48) 60–69374,31159110.51.19 (1.14–1.25) Non-premature mortality377,61297117.21.22 (1.17–1.26) 70–79375,289156927.91.09 (1.06–1.12) ≥ 80374,2194998.91.11 (1.05–1.17)aPremature death was defined as death before age 70. Death age-specific stratification was performed using 10-year intervalsbHazard ratios (HRs) and subdistribution hazard ratios (sHRs) were estimated using Cox or Fine-Gray regression models, adjusted for age, sex, smoking status (never, former, or current), frequency of alcohol consumption (never or occasionally, < 3 times/week, or ≥ 3 times/week), Townsend Deprivation Index, annual household income (<£18,000, £18,000–30,999, £31,000–51,999, or ≥£52,000), education level (basic education or below, vocational or A/AS-level equivalent, or college, university degree or equivalent), ethnicity (White or non-White), medication use (never or current), status of healthy diet (poor, fair, or ideal), and physical activity level (low, moderate, or high). For cancer-specific mortality, additional adjustment was made for family history of cancer Associations between allostatic load score and mortality risk stratified by age at death based on the UK Biobank aPremature death was defined as death before age 70. Death age-specific stratification was performed using 10-year intervals bHazard ratios (HRs) and subdistribution hazard ratios (sHRs) were estimated using Cox or Fine-Gray regression models, adjusted for age, sex, smoking status (never, former, or current), frequency of alcohol consumption (never or occasionally, < 3 times/week, or ≥ 3 times/week), Townsend Deprivation Index, annual household income (<£18,000, £18,000–30,999, £31,000–51,999, or ≥£52,000), education level (basic education or below, vocational or A/AS-level equivalent, or college, university degree or equivalent), ethnicity (White or non-White), medication use (never or current), status of healthy diet (poor, fair, or ideal), and physical activity level (low, moderate, or high). For cancer-specific mortality, additional adjustment was made for family history of cancer A clear increase in all-cause mortality risk was observed with elevated AL levels and age at participant enrollment, in a dose-dependent manner. The HR (95% CI) for all-cause mortality was 2.24 (2.01–2.50) for AL score ≥ 5 in those aged < 50 years, 4.74 (4.34–5.17) in those aged 50–60 years, and 10.16 (9.36–11.03) in those aged ≥ 60 years (Fig-S5). The association between AL scores and all-cause mortality appeared strongest in participants aged < 50 years (HR = 1.22 per one score, 95% CI: 1.19–1.24) at cohort enrollment, followed by those aged 50–60 (HR = 1.17, 95% CI: 1.15–1.18) and ≥ 60 years (HR = 1.11, 95% CI: 1.10–1.12), suggesting an effect modification of age at enrollment on the association between AL scores and all-cause mortality (P-interaction < 0.001; Fig. 1). Similar statistical interaction was also found for cancer, CVD, and digestive disease mortality respectively (all P-interaction < 0.05; Fig. 1). Association between allostatic load scores and all-cause mortality and cause-specific mortality from major disease categories across subgroups by age at recruitment. Participants were stratified by age at baseline (< 50, 50–60, and ≥ 60 years). Analyses were conducted using multivariable-adjusted Cox proportional hazards for all-cause mortality and Fine-Gray competing risk models for cause-specific mortality, respectively. We modeled allostatic load (per score increase), stratified by age groups (< 50, 50–60, and ≥ 60 years). P for interaction between allostatic load scores and age group was derived from a multiplicative interaction term in Cox or Fine-Gray model including all covariates and treated allostatic load scores as a continuous variable. Abbreviations: CI, confidence interval; CVD, cardiovascular disease; sHR/HR, (subdistribution) hazard ratio Associations for cause-specific mortality from individual diseases We further examined the risk of mortality due to major individual diseases (with number of events > 50) (Fig. 2, Fig-S6 and Table-S8). Consistent with the stronger associations observed for major CVD and digestive disease mortality, several individual outcomes showed pronounced associations with AL. Specifically, AL scores were associated with elevated risk of mortality from major CVDs such as chronic ischemic heart disease (sHR = 1.24, 95% CI: 1.21–1.27 per one point of score increase), acute myocardial infarction (sHR = 1.24, 95% CI: 1.20–1.28), hypertensive heart disease (sHR = 1.27, 95% CI: 1.18–1.37), and ischemic stroke (sHR = 1.24, 95% CI: 1.14–1.34). For mortality due to digestive diseases, the highest effect estimates were concentrated in chronic liver diseases, such as liver fibrosis and cirrhosis (sHR = 1.52, 95% CI: 1.37–1.70), and alcoholic liver disease (sHR = 1.39, 95% CI: 1.30–1.49). Subdistribution hazard ratios for CVD and digestive disease mortality from major individual diseases by allostatic load scores based on the UK Biobank. Causes of death with at least 50 events were considered for the analyses. Fine-Gray model was used to assess the associations between score of allostatic load and cause-specific mortality, adjusted for age, sex, smoking status (never, former, or current), frequency of alcohol consumption (never or occasionally, < 3 times/week, or ≥ 3 times/week), Townsend Deprivation Index, annual household income (<£18,000, £18,000–30,999, £31,000–51,999, or ≥£52,000), education level (basic education or below, vocational or A/AS-level equivalent, or college/university degree), ethnicity (White or non-White), medication use (never or current), status of healthy diet (poor, fair, or ideal), and physical activity level (low, moderate, or high). Abbreviations: CI, confidence interval; sHR, subdistribution hazard ratio Although the magnitude of the association between AL scores and cancer mortality was relatively modest, higher AL scores were associated with increased mortality from digestive system cancers (Fig-S6), with the highest effect estimates observed for liver cancer (sHR = 1.27, 95% CI: 1.21–1.33), and bile duct and gallbladder cancer (sHR = 1.23, 95% CI: 1.12–1.36). Significantly increased cancer mortality was also demonstrated for kidney cancer (sHR = 1.18, 95% CI: 1.11–1.25), and several other cancers. In addition, an elevated risk was observed for myelodysplastic syndrome (sHR = 1.19, 95% CI: 1.04–1.35), whereas no significant associations were noted for other non-malignant neoplasms. For respiratory diseases, elevated AL scores were significantly associated with deaths from COPD, bronchiectasis and interstitial pulmonary diseases. In age-stratified analyses, stronger associations with premature mortality were found among younger participants across multiple diseases. Such patterns were evident for CVD outcomes including acute myocardial infarction, chronic ischemic heart disease and hemorrhagic stroke, as well as certain cancers and digestive diseases (Table-S9 and Table-S10; all P-interaction < 0.05). Sensitivity analysis Sensitivity analyses excluding deaths occurring within the first two years of follow-up, deaths post COVID-19 pandemic, participants with baseline medication use, those with pre-existing chronic diseases, or non-White ethnicity did not materially change the findings either for all-cause mortality, cause-specific mortality due to major disease categories (Table-S11) or from individual diseases (data not shown). Sensitivity analyses removing one covariate at a time from the fully adjusted models also yielded similar findings with the primary analyses, supporting the robustness of the results (data not shown). Subgroup analyses by demographic characteristics (other than age) and lifestyle factors Associations between AL scores and all-cause and cause-specific mortality were generally consistent across all subgroups (Table-S12). No substantial effect modification by demographic factors was observed for all-cause mortality, but stronger associations were noted among females for respiratory and digestive disease mortality (P-interaction < 0.05). Among lifestyle factors, associations with all-cause mortality were less pronounced in current smokers, with a similar effect modification by smoking status found for respiratory and digestive disease mortality (P-interaction < 0.05). Additionally, a more pronounced effect magnitude was observed for CVD mortality among individuals with higher annual household income, and for respiratory disease mortality among those with higher educational attainment. Discussion Leveraging data from a large population-based cohort of 378,785 individuals, we found that higher AL scores were associated with progressively elevated risk of all-cause mortality and mortality from major disease categories, with particularly prominent associations for CVD and digestive disease mortality. Notably, the associations appeared stronger for the risk of premature mortality (death at age < 70 years) and among individuals with a younger age at enrollment, underscoring the potential impact of early-life stress and cumulative physiological burden on long-term survival. Our findings extend previous evidence on the relationship between AL score and mortality in several aspects. In this large population-based cohort with long-term follow-up, we were able to systematically evaluate associations of AL score with a broad range of mortality outcomes, including both major cause-specific categories and several individual diseases. In addition, by examining premature mortality and age-specific patterns of association, our analyses provide new insights into the life-course relevance of multisystem physiological dysregulation. These findings suggest that elevated AL scores may represent an early biological signature of cumulative physiological burden that is associated with the development of major chronic diseases and increased mortality risk, particularly at younger ages. Mortality risks of major non-communicable diseases [1, 2] were positively associated with elevated AL scores in our analyses, which appeared pronounced for CVD and digestive disease mortality. Biologically, the associations between scores of AL and mortality risks in our study consistently reflect the cumulative toll of chronic stress on physiological systems [15, 29]. Perceived stress activates the hypothalamic-pituitary-adrenal (HPA) axis and related neuroendocrine pathways, contributing to glucocorticoid dysregulation, metabolic imbalance, inflammation, mitochondrial dysfunction, and cellular aging [30–32]. For CVD mortality, elevated AL scores were associated with increased mortality for leading causes of death such as chronic ischemic heart disease, acute myocardial infarction, and ischemic stroke. Besides, we observed more pronounced associations of AL components reflecting inflammation and kidney function with CVD mortality than of other components. This pattern is consistent with evidence that stress-related neural activity, inflammageing, and trained immunity may sustain vascular inflammation and plaque vulnerability, thereby exerting cumulative stress to CVD mortality [33–35]. In addition, pronounced associations were also observed for chronic liver diseases and several digestive system cancers. These findings point to heightened susceptibility of hepatobiliary and kidney systems to cumulative physiological burden from chronic stress, which has been supported by experimental evidence implicating ferroptosis [36], mitochondrial dysfunction [37], and immune-mediated hepatic injury in stress-related liver pathology [38]. AL scores were more strongly associated with premature mortality, suggesting that cumulative physiological dysregulation may play a particularly important role in shaping early mortality risk. Premature deaths may be driven by multiple determinants, with large-scale analyses showing that environmental and socioeconomic exposures play a substantial role in shaping early mortality risk [4], and recent integrative analyses further suggest that environmental factors may explain more variation in mortality than genetic predisposition alone [39]. In this context, AL may reflect cumulative physiological responses to environmental, behavioral, and psychosocial exposures across the life course, capturing cumulative physiological burden across systems that may contribute to early mortality risk. Consistent with this pattern, stratified analyses by baseline age demonstrated significant associations between AL scores and mortality risk across all age categories, with relative risks being most pronounced among younger adults. These age-stratified findings should nevertheless be interpreted cautiously, given the relatively limited number of deaths in younger age groups. A prior meta-analysis reported stronger associations between AL and all-cause mortality in studies with a mean age below 65 years [18]; however, that study lacked individual-level age data and was thus unable to investigate the associations among individuals within specific age subgroups. Leveraging a well-established prospective mega-cohort with long-term follow-up, our current study was able to refine the age gradient and demonstrate particularly pronounced associations in participants younger than 50 years. These findings suggest that AL may serve as an indicator of early physiological dysregulation associated with premature mortality and highlight the potential relevance of early-life intervention strategies targeting elevated physiological burden. Despite overall declines in NCD mortality in some regions, the burden of premature deaths—particularly those considered avertable—remains substantial and unevenly distributed across populations and causes [40, 41]. A proportion of such deaths can be prevented through effective policies and health care, but the extent of preventability varies greatly by disease categories and settings [41]. In this context, AL has the potential to offer added value to population risk stratification and surveillance, which may be achieved by pinpointing subgroups more prone to experience avertable premature deaths, especially among younger individuals at high risk. By capturing heterogeneity in premature mortality, AL provides a framework to evaluate the efficacy of interventions targeting stress-related pathways and to generate evidence for refining prevention strategies. Beyond serving as a tool for risk stratification, AL may act as a traceable surrogate indicator that reflects or mediates exposure to cumulative stress-related physiological dysregulation. However, the present study only investigated the association between AL and mortality, and did not further explore the potential mediating role of AL in the pathway between upstream risk factors and mortality. Furthermore, some adjusted covariates may lie partly along the causal pathway between AL (or its upstream exposures) and mortality, which underscores the need for future studies using causal inference and mediation analyses. Collectively, these points represent limitations of our study and warrant further investigation in future research. Several other limitations should be acknowledged. First, the AL score was derived from a single time-point measurement, precluding evaluation of its longitudinal changes or cumulative trajectories, which may offer more nuanced insights into dynamic physiological dysregulation and mortality risk [17]. Second, although we adjusted for key sociodemographic and behavioral variables, genetic predisposition to cause-specific mortality was not fully accounted for, given the varying heritability across disease outcomes. Notably, only cancer-specific models included family history as a proxy. Future studies should explore gene-AL score interactions to enhance understanding of individualized risk. In addition, residual confounding due to unmeasured or imperfectly measured factors cannot be excluded. Third, although the UKB is a well-established mega-cohort, it represents a selected population, and participants in our study only included those with complete data on all 12 metrics used to construct the AL score, which may raise concerns regarding the representativeness of our study sample. Besides, the UKB predominantly comprises individuals of European ancestry, limiting generalizability to other populations due to smaller non-European samples. Finally, despite the large cohort size, cause-specific analyses were confined to mortality from four major NCD categories. Within these categories, analyses of individual diseases were restricted to outcomes with an empirical cutoff of at least 50 deaths to minimize unstable estimates. As such, these subtype-specific results for rare causes of death should be interpreted as exploratory. Given the comparatively modest number of events, other crucial NCDs, such as diabetes, were not included in the analyses. Conclusions In summary, we found elevated AL scores were significantly associated with increased risks of all-cause and multiple cause-specific mortality, with larger relative estimates observed for CVD and digestive disease mortality. These associations were more pronounced among younger individuals and for premature mortality, suggesting that AL captures early physiological dysregulation that precedes clinically manifest disease. Our findings underscore the potential utility of AL as an indicator for identifying individuals at elevated risk of premature deaths. Further studies are warranted to examine whether AL, as a marker of cumulative physiological burden stemming from chronic stress, can be integrated into risk stratification frameworks to identify populations at elevated risk of adverse health outcomes and premature mortality, particularly among younger individuals. Supplementary Information Supplementary Material 1. Acknowledgements We thank all individuals who participated in this study and donated samples. The UK Biobank resource was used for this research under Application No. 90999. We acknowledge the contributions of participants and investigators of UK Biobank. Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research. Abbreviations Allostatic load Chronic obstructive pulmonary disease Cardiovascular disease Diastolic blood pressure Glycated hemoglobin High-density lipoprotein cholesterol Hazard ratio International Classification of Diseases,10th Revision Insulin-like growth factor-1 Interquartile range Low-density lipoprotein cholesterol Non-communicable disease Systolic blood pressure Subdistribution hazard ratio Triglycerides UK Biobank Waist-to-hip ratio Authors’ contributions HM Xu, ZC Liu and WQ Li conceived and designed the study. HM Xu and ZC Liu drafted the manuscript. HM Xu, ZC Liu, XL Wang, CZ, YZ, TZ, and JY Zhang contributed to data collection and analysis. KF Pan provided critical revisions. WQ Li supervised the study and is the guarantor of the study. All authors reviewed and approved the final manuscript. Funding This study was funded by the Noncommunicable Chronic Diseases-National Science and Technology Major Project (No. 2023ZD0501400; 2025ZD0545200), National Natural Science Foundation of China (No. 82273704, 82504482), Beijing Hospitals Authority’s Ascent Plan (DFL20241102), China Postdoctoral Science Foundation (2024M760152), Postdoctoral Fellowship Program of CPSF (GZB20250192), Beijing Hospitals Authority Clinical Medicine Development of Special Funding Support (No. ZLRK202325), Peking University Medicine Fund for World’s Leading Discipline or Discipline Cluster Development (No. BMU2022XKQ004), Science Foundation of Peking University Cancer Hospital (No. BJCH2024BJ02, XKFZ2410, and BJCH2025CZ04). Data availability Data are available in the UK Biobank upon application (http://www.ukbiobank.ac.uk). No additional data are available. Declarations Ethics approval and consent to participate The UKB received ethical approval from North-West Multi-center search Ethics Committee (reference 11/NW/0382) and all participants provided written informed consent. Consent for publication Not applicable. 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. Heng-Min Xu and Zong-Chao Liu contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement Data are available in the UK Biobank upon application (http://www.ukbiobank.ac.uk). No additional data are available.