Utilizing data from the UK Biobank, a large-scale population-based biomedical database, this prospective cohort study included approximately 502,132 participants aged 40–69 years, recruited from 22 assessment centers throughout England, Scotland, and Wales during the period of 2006–2010. Participants underwent comprehensive baseline assessments including touchscreen questionnaires, physical examinations, and biological sample collection. The detailed study protocols and methodologies, previously published, can be accessed publicly via the UK Biobank website (https://www.ukbiobank.ac.uk). The North-West Multi-centre Research Ethics Committee (reference 16/NW/0274) approved the study, and all participants provided written informed consent in line with the Declaration of Helsinki principles.

Participants were excluded if they: (i) lacked measurement of BCAAs at baseline (n = 227,918), (ii) had incomplete BCAAs data (n = 213), (iii) had a history of RHD at baseline assessment (n = 406). Following these exclusions, our final analytical cohort comprised 273,595 participants who met all inclusion criteria and had complete baseline BCAA measurements available for analysis (Supplementary Fig. 1).

Nightingale Health Ltd. (Helsinki, Finland) developed the high-throughput NMR-based metabolic profiling platform used for measuring serum BCAAs concentrations (https://biobank.ndph.ox.ac.uk/showcase/label.cgi?id=220). This study analyzed isoleucine, leucine, and valine as the specific BCAAs, with the total BCAAs concentration calculated from the sum of these three amino acids. Individual BCAAs levels and total BCAA concentrations were then categorized into quartile (0.108 Q1 $\le$ 0.306 mmol/L; 0.306 Q2 $\le$ 0.353 mmol/L; 0.353 Q3 $\le$ 0.411 mmol/L; 0.411 Q4 $\le$ 1.305 mmol/L).

The outcome was the incidence of chronic RHD. The International Classification of Diseases (ICD-9 and ICD-10) coding systems were used to identify RHD cases through hospital inpatient records (primary or secondary hospital diagnosis) and death registry records (underlying or contributory cause of death) (Supplementary Table 1). Chronic RHD defined by $\ge$2 hospitalizations with ICD-9 or ICD-10 code, or RHD as primary cause of death in registries. Participants were followed up starting from their enrollment date. The follow-up ended when any of these events first occurred: the participant developed RHD, the participant passed away, or the study reached its cutoff date.

Four established inflammatory and immune-related biomarkers were calculated as composite indices to assess systemic immune status and inflammatory burden. The albumin-to-alkaline phosphatase ratio (AAPR) was determined by dividing the level of serum albumin by that of alkaline phosphatase. The lymphocyte-to-monocyte ratio (LMR) was calculated through division of the absolute lymphocyte count by the absolute monocyte count. Meanwhile, the lymphocyte-to-C-reactive protein ratio (LCR) was obtained via division of lymphocyte count by C-reactive protein concentration. The platelet-to-lymphocyte ratio (PLR) was figured out through the division of the platelet count by the lymphocyte count.

Demographic variables included participant age (categorized as 65 years and $\ge$65 years), sex (male and female), and race (Other and White). Socioeconomic indicators comprised educational attainment (university degree vs. no university degree) and annual household income (£18,000 vs. $>$£18,000). Anthropometric measurements encompassed body mass index (BMI), which was calculated based on height and weight. Lifestyle factors encompassed physical activity (low/moderate/high), smoking history (never/previous/current), and alcohol consumption patterns (never/previous/current). The quality of the diet was evaluated by the Dietary Approaches to Stop Hypertension (DASH) score, which came from the responses to a food frequency questionnaire. Medical history was ascertained through participant self-reports combined with linked health records, including previous diagnoses of hypertension, diabetes mellitus, cardiovascular disease, and cancer. Cardiovascular disease history included atrial fibrillation, ischemic heart disease, heart failure, and stroke.

For covariates with missing data, multiple imputation by chained equations was applied, performing five imputations. The predictor variables used in the imputation model included age, sex, race, education, income, physical activity, smoking, alcohol consumption, DASH score, BMI and history of hypertension, diabetes mellitus, cardiovascular disease (CVD), and cancer. The proportion of missing covariates was all within 30% (Supplementary Table 2), and they were missing randomly (Supplementary Fig. 2). Continuous variables were reported as means alongside standard deviations (SD) or as medians with interquartile ranges. Categorical variables, on the other hand, were shown as counts and percentages. For normally distributed variables, group comparisons were conducted using ANOVA or Student’s t-test. For non-normally distributed variables, the Kruskal-Wallis rank sum test or Mann-Whitney U test was applied. Categorical variables were assessed using the chi-squared test or Fisher’s exact test as appropriate.

To explore the connection between BCAAs and RHD risk, we utilized multivariable Cox proportional hazard models to derive hazard ratios (HRs) and 95% confidence intervals (CIs). The BCAAs data underwent transformation into z-scores, with HRs assessed for each SD unit rise. Additionally, BCAAs were stratified into quartiles to scrutinize the influence of quartile increments on RHD risk and to conduct trend tests. Four progressive models were established: crude model (unadjusted), Model 1 adjusted for sociodemographic characteristics (age, sex, race, education, and income); Model 2 further adjusted for life-style (physical activity, smoking, alcohol consumption, and DASH score); and Model 3 additionally adjusted for disease and health status (BMI and history of hypertension, diabetes mellitus, CVD, and cancer). To investigate the dose-response relationship between BCAA levels and RHD events, restricted cubic splines (RCS) were employed. The RCS analysis was fully adjusted for all covariates included in Model 3.

Using the “CMAverse” R package, causal mediation analysis was performed to explore how immune markers (AAPR, LMR, LCR, and PLR) might mediate the relationship between BCAAs and RHD risk. This approach employs counterfactual-based mediation analysis, which allows for the dissecting the total effect into natural direct and indirect effects within the counterfactual framework [19]. Bootstrap resampling (n = 100) was used to calculate 95% CI for the mediation effects. The mediated proportion was determined by dividing the natural indirect effect by the total effect, representing the percentage of the total association that operates through each immune marker.

To evaluate potential disparities in study outcomes among different population groups, stratification was performed based on age (65 vs. $\ge$65 years), sex, education, race, income, physical activity, smoking status, and alcohol consumption. Interaction terms were tested to evaluate effect modification. For the subgroups with a small sample size, we conducted a post hoc power analysis using the powerSurvEpi R software package (version 0.1.5, Vienna, Austria). We performed several sensitivity analyses to test the robustness of our study findings. Firstly, the accelerated failure time model with Weibull distribution was employed as an alternative analytical approach. Secondly, we conducted Cox regression models using age as the time scale. Thirdly, to evaluate the influence of missing data imputation, participants with incomplete baseline covariate data were excluded. Fourthly, to reduce the reverse causality effect, participants who developed RHD within the first 2 years of follow-up were excluded. Finally, participants with a history of CVD (ischemic heart disease, heart failure, and stroke) were excluded to reduce potential confounding.

Statistical analyses were performed using R software (version 4.5.0, R Foundation, Vienna, Austria). The statistical significance of the differences between groups was determined by a two-tailed p value less than 0.05. In this study, we analyzed four exposure factors and four immune markers. The significance level was adjusted using the Bonferroni correction method: for the association between BCAAs and RHD, the significance level was set at 0.05/4 = 0.0125; for the mediation analysis, the significance level was set at 0.05/16 = 0.003.

We included a total of 273,595 participants in the final analysis. The distribution of covariates remains essentially consistent before and after imputation for missing data (Supplementary Table 3). Over a median follow-up period of 13.71 $\pm$ 1.17 years, chronic RHD occurred in 6051 (2.21%) participants. Baseline characteristics demonstrated distinct patterns across BCAA quartiles (Table 1). Participants in higher BCAA quartiles were more likely to be male (25.90% in Q1 vs. 61.13% in Q4), had higher BMI (25.76 $\pm$ 4.40 kg/m² in Q1 vs. 28.89 $\pm$ 4.87 kg/m² in Q4), and showed increased prevalence of diabetes mellitus (2.67% vs. 9.71%), hypertension (25.20% vs. 36.73%), and CVD (6.17% vs. 10.24%). Conversely, DASH scores decreased progressively across quartiles (5.01 $\pm$ 1.34 in Q1 vs. 4.80 $\pm$ 1.34 in Q4).

RHD participants were older (61.96 $\pm$ 6.18 vs. 56.44 $\pm$ 8.08 years), had higher BMI, and demonstrated greater comorbidity burden compared to controls (Supplementary Table 4). In the RHD group, total BCAA concentrations were significantly higher (p 0.001), with similar patterns observed for individual components (Table 2). Immune markers revealed significancal difference between two groups. Specifically, RHD patients had notably lower AAPR (0.530 vs. 0.563), LMR (3.800 vs. 4.196), LCR (1.065 vs. 1.408), and PLR (128.743 vs. 132.632), with all p-values being less than 0.001.

Serum BCAAs were significantly associated with RHD risk, as shown by multivariable Cox proportional hazard regression analyses (Table 3). We have verified the Cox proportional hazards assumption through Schoenfeld residual tests (Supplementary Tables 5,6; Supplementary Figs. 3–11). In Model 3, a 1-SD rise in total BCAAs corresponded to a 4.8% increased risk of RHD (HR = 1.048, 95% CI: 1.023–1.074, p 0.001). Upon quartile analysis, individuals in the top quartile (Q4) exhibited an 8.3% higher risk than those in the bottom quartile (Q1) (HR = 1.083, 95% CI: 1.007–1.164, p = 0.032). HRs across BCAA quartiles, denoting the BCAAs concentration threshold ($>$0.411 mmol/L) for RHD risk. Among individual BCAAs, valine exhibited the strongest association with RHD risk, with each 1-SD increase conferring a 6.1% elevated risk (HR = 1.061, 95% CI: 1.035–1.088, p 0.0001). Isoleucine showed a modest but significant association (HR = 1.039, 95% CI: 1.014–1.065, p = 0.002), while leucine demonstrated the weakest association (HR= 1.030, 95% CI: 1.004–1.056, p = 0.021). Restricted cubic spline analyses confirmed dose-response relationships between BCAAs and RHD risk (Fig. 1).

Fig. 1.

Association of the BCAAs on RHD using RCS. (A) Association of the BCAAs on RHD. (B) Association of the leucine on RHD. (C) Association of the isoleucine on RHD. (D) Association of the valine on RHD. Models were adjusted for age, sex, race, education, income, physical activity, smoke, alcohol, DASH, BMI, and the history of hypertension, diabetes mellitus, CVD, and cancer. BCAA, Branched-chain amino acid; DASH, Dietary approaches to stop hypertension; BMI, Body mass index; CVD, Cardiovascular disease; RHD, Rheumatic heart diseases; RCS, Restricted cubic spline; HR, Hazard ratio.

Consistent associations between BCAAs and RHD risk across different population strata were demonstrated in subgroup analyses (Fig. 2). The most notable finding was a significant age-related interaction (p for interaction = 0.032). Compared to older participants ($\ge$65 years: HR = 1.013, 95% CI: 0.974–1.053, p = 0.523), younger participants exhibited stronger associations (65 years: HR = 1.074, 95% CI: 1.040–1.108, p 0.001). Significant interactions were not found for sex, education, income, or lifestyle factors including physical activity, smoking, and alcohol consumption. Individual BCAA analyses revealed consistent age-related patterns (Supplementary Figs. 12–14), with leucine, isoleucine, and valine all demonstrating stronger associations in younger participants. Age-stratified restricted cubic spline analyses further confirmed these differential associations (Fig. 3). In participants 65 years, all BCAAs exhibited progressive dose-response relationships with steeper risk curves, particularly for total BCAAs and valine.

Fig. 2.

Subgroup analysis of association of BCAAs on chronic RHD. Models were adjusted for age, sex, race, education, income, physical activity, smoke, alcohol, DASH, BMI, and the history of hypertension, diabetes mellitus, CVD, and cancer. CI, Confidence interval.

Fig. 3.

RCS analysis of association between BCAAs with RHD stratified by age. (A) Association between BCAAs with RHD stratified by age. (B) Association between leucine with RHD stratified by age. (C) Association between isoleucine with RHD stratified by age. (D) Association between valine with RHD stratified by age. Models were adjusted for age, sex, race, education, income, physical activity, smoke, alcohol, DASH, BMI, and the history of hypertension, diabetes mellitus, CVD, and cancer.

In the subgroups with relatively small sample sizes, specifically “other” in race, “Current” in smoking status, and “Never” and “Previous” in alcohol consumption, no statistically significant associations were observed between BCAAs and RHD. To further elucidate this, we conducted post hoc power analyses for Cox regression with survival data using the powerSurvEpi R package. The statistical power to detect significant effects in these four subgroups was 2.5%, 2.6%, 4.2%, and 5.5%, respectively. These low power values suggest that the absence of statistical significance in these subgroups is likely attributable to insufficient statistical power, rather than a true absence of association.

Causal mediation analyses revealed that immune markers significantly mediated the associations between BCAAs and RHD risk (Supplementary Table 7, Fig. 4). Among the four immune indicators examined, LCR demonstrated the strongest mediating effect, accounting for 30.8% (95% CI: 32.53–52.87%) of the total association between BCAAs and RHD risk. The AAPR served as the second most important mediator, explaining 17.45% (95% CI: 17.47–32.58%) of the relationship. LMR contributed 11.07% (95% CI: 7.99–17.68%) to the mediation, while PLR had the smallest mediating effect at 7.92% (95% CI: 4.89–12.12%). Individual BCAAs analyses demonstrated consistent patterns with varying magnitudes (Supplementary Tables 8,9, Supplementary Figs. 15,16).

Fig. 4.

Mediating proportion of immunity in the relationship between BCAAs and RHD. (A) Mediating proportion of AAPR in the relationship between BCAAs and RHD. (B) Mediating proportion of LMR in the relationship between BCAAs and RHD. (C) Mediating proportion of LCR in the relationship between BCAAs and RHD. (D) Mediating proportion of PLR in the relationship between BCAAs and RHD. Models were adjusted for age, sex, race, education, income, physical activity, smoke, alcohol, DASH, BMI, and the history of hypertension, diabetes mellitus, CVD, and cancer. AAPR, Albumin to alkaline phosphatase ratio; LMR, Lymphocyte-to-monocyte ratio; LCR, Lymphocyte and C-reactive protein Ratio; PLR, Platelet-to-Lymphocyte Ratio.

Multiple sensitivity analyses confirmed the robustness of our primary findings. The accelerated failure time model yielded consistent results, with each 1-SD increase in total BCAAs correlating to a 2.4% increased risk of RHD (HR = 1.024, 95% CI: 1.011–1.037) (Supplementary Table 10). Cox regression using the time scale of age produced nearly identical results to our primary analysis (Supplementary Table 11), with total BCAAs maintaining significance (HR = 1.048, 95% CI: 1.023–1.074). After excluding individuals with missing baseline covariates, the associations remained significant though slightly attenuated, with total BCAAs showing a 4.2% increased risk per 1-SD rise (HR = 1.042, 95% CI: 1.008–1.077) (Supplementary Table 12). To address potential reverse causality, excluding individuals with incident RHD during the initial 2-year follow-up period, total BCAAs correlated with a 5.1% higher risk (HR = 1.051, 95% CI: 1.024–1.077) (Supplementary Table 13). Finally, after excluding participants with baseline CVD history, the link stayed significant for total BCAAs (HR = 1.027, 95% CI: 1.011–1.057) (Supplementary Table 14). Corresponding RCS analyses consistently demonstrated dose-response relationships across all sensitivity analyses (Supplementary Figs. 17–20).