Development of a novel meta-polygenic risk score for risk assessment and clinical manifestation prediction in systemic lupus erythematosus

Systemic lupus erythematosus (SLE) is a complex and heterogeneous autoimmune disease characterised by multisystem involvement and diverse clinical manifestations. This marked clinical variability often results in delayed diagnosis and suboptimal treatment, with many patients progressing to vital organ damage before definitive diagnosis, ultimately leading to poor prognosis and imposing a substantial burden on society. In recent years, the incidence of SLE has been rising globally and within China, 1 2 disproportionately affecting females who comprise approximately 90% of cases. Alarmingly, SLE ranks among the leading causes of mortality in young women, 3 reducing average life expectancy by approximately 22 years in females and 12 years in males compared with the general population. 4 Consequently, the development of strategies for early diagnosis and intervention in SLE is both critically important and urgently needed. The pathogenesis of SLE is highly complex, involving the interplay of genetic, environmental and lifestyle factors. Research has found that SLE exhibits strong familial aggregation, with relatives of patients with SLE having an elevated risk of developing SLE and other autoimmune diseases. The heritability of SLE has been estimated at 43.9%, 5 highlighting the potential need for genetic counselling in families with affected members. In recent decades, more than 200 genetic risk loci for SLE have been identified in East Asian populations through genome-wide association studies (GWAS). 6 These genetic variants, present from birth and preceding the emergence of clinical risk factors, offer promising opportunities for early disease prediction. Polygenic risk scores (PRSs) quantify an individual’s genetic predisposition to a disease by aggregating the effects of multiple single-nucleotide polymorphisms (SNPs) across the genome and have become an important tool in the advancement of personalised medicine. In recent years, several PRS models have been developed to predict the risk of SLE 7 10 which primarily focus on SNP loci directly associated with SLE development, without accounting for environmental and lifestyle factors that may influence disease risk over time. To improve the predictive performance of SLE risk models, studies using multivariate analyses have identified significant associations between SLE susceptibility and various risk factors, including lifestyle components (eg, smoking, alcohol consumption, sleep patterns) and environmental exposures (eg, sunlight, air pollution). 11 12 Although the integration of genetic and environmental factors has been shown to enhance the predictive performance of SLE risk models, this approach compromises one of the key advantages of PRS—its ability to identify high-risk individuals at birth, as many of these risk factors are acquired later in life. To address this limitation, a novel approach called the meta-polygenic risk score (metaPRS) has been proposed to enhance the prediction of complex diseases. Unlike traditional PRSs that focus solely on genetic loci directly associated with the disease, the metaPRS integrates susceptibility loci from related traits, including risk factors and comorbid conditions, to construct a composite score. This strategy has demonstrated improved risk stratification in several diseases, such as coronary artery disease, 13 intracranial aneurysms 14 and lung cancer, 15 outperforming conventional PRS approaches. However, this method has not yet been applied to autoimmune diseases, including SLE. This study established the first metaPRS of SLE for East Asian population and systematically compared its predictive performance with that of the conventional SLE PRS. Furthermore, we evaluated its utility in assessing clinical manifestations and enhancing risk stratification.

A total of 2388 patients with SLE and 1132 healthy controls were initially enrolled in this study. Given the well-documented female predominance in SLE epidemiology and the recognised sex-specific effects of many known risk factors (eg, age at menarche, endometriosis), we restricted our analysis to female subjects only in this study. Following genetic QC procedures and the exclusion of male subjects, the final analytical cohort included 2209 female patients with SLE and 695 female controls ( figure 1 ). The baseline characteristics of patients are shown in table 1 . The median age at diagnosis among patients with SLE was 27.63 years (SD=13.57). The median follow-up duration was 29.35 months, with an average of 7.38 follow-up visits. The median of the highest recorded Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI-2K) during follow-up was 5 (IQR=8). The most frequently observed clinical manifestations were mucocutaneous involvement (69.35%), haematologic abnormalities (67.68%) and musculoskeletal symptoms (54.41%). In terms of laboratory profiles, anti-nuclear antibody was positive in 99.91% of patients, followed by anti-double-stranded DNA (anti-dsDNA) (79.4%), anti-Sjogren syndrome A (anti-SSA) (62.25%), anti-ribonucleoprotein antibody (anti-RNP) (45.09%) and anti-Smith antibody (anti-Sm) antibody (37.03%). Clinical manifestations were classified according to the British Isles Lupus Assessment Group index (BILAG-2004). anti-dsDNA, anti-double-stranded DNA antibody; anti-RNP antibody, anti-ribonucleoprotein antibody; anti-rRNP antibody, anti-ribosomal ribonucleoprotein protein antibody; anti-Sm antibody, anti-Smith antibody; anti-SSA antibody, anti-Sjogren syndrome A antibody; anti-SSB antibody, anti-Sjogren syndrome B antibody; SLE, systemic lupus erythematosus; SLEDAI-2K, Systemic Lupus Erythematosus Disease Activity Index 2000. Based on our literature review, four of the fourteen traits (high carbohydrate diet, Epstein-Barr virus infection, age at menarche and coeliac disease) did not have available GWAS data in East Asian populations. We therefore excluded these traits and proceeded to construct PRS models for the ten remaining traits. Subsequently, we selected those models that showed an association with SLE susceptibility (see Patients and method). Among these, five traits showed significant genetic associations with SLE ( figure 2 and online supplemental table 3 ), including SLE development, smoking initiation, serum selenium levels, endometriosis and Graves’ disease. As expected, the PRS of SLE development demonstrated the strongest associations with SLE (OR=2.03), followed by PRSs for Graves’ disease (OR=1.13), endometriosis (OR=1.12), smoking initiation (OR=1.12) and serum selenium levels (OR=0.89). The effect directions of all these five PRSs were fully consistent with findings from previous epidemiological studies ( online supplemental table 1 ). The top SNPs contributing to each trait-level PRS are provided in online supplemental tables 4–8 . To assess the independence of PRSs with distinct characteristics, pairwise correlation analysis was performed. As shown in online supplemental figure 2 , a significant correlation was observed between the PRSs for SLE and Graves’ disease (r=0.08, FDR=6.22e−5). After adjusting the coefficients using elastic net logistic regression ( online supplemental figure 3 and table 9 ), all five PRSs were retained and incorporated into the final metaPRS model, which comprised 75 427 unique SNPs. The overlap among these SNPs is illustrated in online supplemental figure 4 and table 10 . Finally, the metaPRS exhibited a stronger association with SLE development (OR=2.12, FDR<2e−16) than any individual trait-level PRS ( figure 2 ). As shown in figure 3A , the distribution of metaPRS was significantly higher in patients with SLE than in controls (p<2.2e−16). Dose–response analysis revealed an approximately linear increase in SLE risk across the metaPRS spectrum, with predicted risk rising from 46% to 94% as the metaPRS z-scores increased from −2 to +2 ( figure 3B ). Model calibration was further assessed using bootstrap-validated calibration curves by stratifying predicted probabilities into quintiles ( figure 3C ). The resulting curves demonstrated strong concordance between predicted risks and observed event rates, which was further supported by a non-significant Hosmer-Lemeshow goodness-of-fit test (p=0.799), indicating good calibration. To further enhance the clinical applicability of the metaPRS, the optimal cut-off value was determined by maximising the Youden index, which balances sensitivity and specificity ( online supplemental figure 5 ). A total of 57.9% of individuals exceed this threshold, exhibiting a 3.33-fold higher risk of SLE compared with those below the threshold (95% CI: 2.79 to 3.98; p<2e−16). Next, to assess the robustness of the metaPRS, internal validation was performed via bootstrap resampling, confirming stable predictive performance across datasets, with a 95% confidence interval for the AUC ranging from 0.67 to 0.71 ( figure 3D ). Subsequently, to evaluate the risk stratification capability of the metaPRS relative to the PRS SLE (the PRS derived exclusively from SLE risk genetic variants), we categorised participants into low-risk, intermediate-risk and high-risk groups according to metaPRS or PRS SLE tertiles. As shown in figure 3E , individuals in the low-risk group defined by metaPRS had an even lower risk of SLE than those in the PRS SLE (OR=0.96, p=0.646). In contrast, both the intermediate-risk and high-risk groups defined by metaPRS exhibited significantly higher risks of SLE development (OR=2.13, p=3.33e−13; OR=4.93, p<2e−16), indicating superior risk stratification performance. Finally, to quantitatively assess the discrimination power of the metaPRS, ROC curves were plotted for both PRS SLE and metaPRS. As shown in figure 3F , the metaPRS demonstrated significantly improved discriminative performance, with a higher AUC compared with PRS SLE (0.69 vs 0.68, p=0.046, Delong test), highlighting its enhanced predictive accuracy. We first investigated the relationship between the metaPRS and clinical features among patients with SLE ( online supplemental table 11 ). As shown in figure 4 , higher metaPRS were found to be significantly associated with several autoantibodies, including anti-dsDNA, anti-Sm, anti-RNP, anti-SSA and anti-rRNP ( online supplemental table 11 ). To further explore the genetic determinants underlying these clinical associations, we decomposed the metaPRS into two components: (1) PRS SLE and (2) the risk factor PRS (PRS riskfactor ), which aggregated the genetic contributions from four established risk factors (smoking initiation, serum selenium levels, endometriosis and Graves’ disease). Genetic association analyses revealed that PRS SLE showed a markedly stronger association with SLE susceptibility (OR=2.03, p<2e−16) than PRS riskfactor (OR=1.26, p=4.66e−07), with both components demonstrating independent effects. In the predictive modelling, PRS SLE contributed the majority (75.4%) of the total predictive capacity, while PRS riskfactor accounted for the remaining 24.6% of the explained variance. Clinical phenotype analyses demonstrated striking differences between the genetic components. PRS SLE showed significant associations with multiple disease manifestations including positivity for anti-dsDNA, anti-Sm, anti-RNP, anti-SSA and anti-rRNP ( online supplemental table 11 ). In contrast, PRS riskfactor showed no significant associations with any clinical features in our cohort ( online supplemental table 11 ). To examine potential interaction effects, we conducted stratified analyses by PRS SLE levels. Even within matched PRS SLE strata, different PRS riskfactor groups still showed no differential associations with clinical phenotypes ( online supplemental table 12 ), suggesting that risk factor genetics do not substantially modify disease presentation regardless of baseline SLE genetic risk. Further contrasting two extreme subgroups: HighSLE-LowRiskFactor (patients with high PRS SLE but low PRS riskfactor , n=242) and LowSLE-HighRiskFactor (patients with low PRS SLE but high PRS riskfactor , n=245), revealed distinct clinical profiles. The HighSLE-LowRiskFactor group exhibited a higher proportion of mucocutaneous manifestations (73.6% vs 68.8%, p=0.031), positivity rates for anti-SSA (67.4% vs 58.2%) and anti-rRNP (34.8% vs 22.9%, p=0.004) antibodies ( online supplemental table 13 ). Interestingly, although no association was observed between metaPRS and age at SLE diagnosis, distinct predictive performances of PRSs were identified in early-onset and late-onset SLE subgroups. Patients were stratified into two subgroups: childhood-onset (cSLE, diagnosis age<16 years) and adult-onset (aSLE, diagnosis age≥16 years). As shown in figure 5 and online supplemental table 14 , the metaPRS outperformed alternative models in discriminating both cSLE and aSLE cases, with a slight improvement in risk stratification observed in cSLE (high risk vs low risk: OR=6.24, p<2e−16) relative to aSLE (high risk vs low risk: OR=5.03, p<2e−16). Although the difference in effect sizes between cSLE and aSLE did not reach statistical significance, these findings suggest the metaPRS’s potential to guide early intervention strategies tailored to disease-onset age.

The overall design of this study is illustrated in figure 1 . A total of 2388 patients with SLE were enrolled from Peking Union Medical College Hospital, all of whom had been registered in the Chinese SLE Treatment and Research (CSTAR) registry cohort between May 2009 and December 2024. All patients met at least one of the following classification criteria for SLE: the 1997 American College of Rheumatology (ACR) revised criteria, the 2012 Systemic Lupus International Collaborating Clinics criteria or the 2019 European Alliance of Associations for Rheumatology/ACR criteria. Demographic, clinical and laboratory data were collected through the CSTAR registry. The procedures for sample collection and data acquisition adhered to standardised protocols described in our previous publication. 16 Clinical manifestations such as mucocutaneous involvement, haematologic involvement and renal involvement were defined as positive if they had occurred at any point prior to December 2024, according to the criteria specified in the British Isles Lupus Assessment Group index (BILAG-2004). The control group consisted of 1132 individuals from the Shunyi Study, a population-based cohort in China, 17 excluding those with any diagnosed autoimmune diseases. DNA genotyping of blood samples from patients with SLE was performed using the Illumina Infinium Asian Screening Array-24 V.1.0 on the Illumina iScan platform, following the manufacturer’s protocol. DNA genotyping data for the control group were previously generated using the same platform. 18 After merging the genetic data of SLE cases and healthy controls, we applied a quality control (QC) pipeline at both the variant and sample levels. For variant-level QC, SNPs were excluded if they met any of the following criteria: (1) Hardy-Weinberg equilibrium p value<1×10⁻⁶ or (2) call rate≤90% across all samples. For sample-level QC, individuals were excluded if they: (1) showed sex discordance based on X chromosome heterozygosity; (2) had excessive heterozygosity (>3 SD from the mean); (3) exhibited low call rates (≤90%); (4) were identified as outliers in principal component analysis (PCA); or (5) displayed evidence of relatedness (PI_HAT >0.2 in identity-by-descent analysis), in which case the lower-quality sample from each related pair was removed. The remaining high-quality samples and variants were subsequently phased using SHAPEIT2 (v2.r904) and imputed using Minimac4 (V.1.0.2) with default parameters, with the 1000 Genomes Project Phase 3 V.5 (1000G p3v5) serving as the reference panel. Only common variants (minor allele frequency>1%) with imputation quality scores (Rsq)>0.3 were retained for downstream association analyses. Furthermore, PCA using common SNPs indicates no obvious population stratification between cases and controls ( online supplemental figure 1 ). Based on a comprehensive literature review, we selected 14 traits implicated in SLE pathogenesis to construct the metaPRS. Traits were included only if supported by prospective cohort studies or by both case–control and Mendelian randomisation (MR) studies, ensuring that selected traits are associated with SLE development and may have potential causal relevance ( online supplemental table 1 ). These traits encompassed: (1) SLE development; (2) lifestyle behaviours, including smoking (current smoking, cigarettes per day, smoking initiation), drinking (drinks per week, drink status), sleep disorders (insomnia) and high carbohydrate diet; (3) essential nutrients, including serum vitamin D levels (serum 25-hydroxyvitamin D levels), serum selenium levels and serum total calcium levels; (4) infection (Epstein-Barr virus infection); (5) female hormone-related traits (age at menarche); (6) diseases associated with increased risk of SLE, including endometriosis, Graves’ disease, coeliac disease and periodontitis. For each trait, we obtained summary data from the largest GWAS in East Asian populations. Traits without East Asian GWAS data were excluded from further analyses. Detailed information on each trait-specific PRS is provided in online supplemental table 2 . SNP effect sizes used for PRS construction were derived from the corresponding GWAS results. Each trait-level PRS was calculated using the pruning and thresholding method implemented in PRSice2. 19 Briefly, clumping was first performed to remove SNPs in linkage disequilibrium (r 2 >0.1) within a 250 kb window. Second, for each trait, PRSice2 systematically tested a range of p value thresholds (from 5×10 −8  to 1) to generate multiple SNP subsets. Next, to identify the optimal PRS per trait for predicting SLE, logistic regression models were applied using the top five genetic principal components as covariates. For each trait, the PRS with the highest Nagelkerke’s R 2 in the training dataset was selected as the optimal PRS. To construct the metaPRS, we first selected traits whose individual PRSs remained significantly associated with SLE risk after p value adjust (false discovery rate (FDR)<0.05) and standardised each PRS value to a z-score for further analysis. Elastic net logistic regression was tuned using 10-fold cross-validation. A grid of α and λ values was evaluated, and for each α, cross-validation was used to identify the λ that achieved the highest mean area under the receiver operating characteristic curve (AUC) across folds. The α and λ combination that yielded the highest mean AUC was selected as the final hyperparameters, which were then used as weights for the individual PRSs in the composite score. The final metaPRS for SLE was computed as a weighted sum: where n is the total number of PRSs, β j is the coefficient from the elastic net logistic regression and σ j is the empirical SD of each PRS. Binary logistic regression was employed to assess the associations between PRSs and SLE development and to estimate ORs with corresponding 95% CIs. Pearson correlation analysis was used to evaluate pairwise correlation coefficients among trait-specific PRSs. SNP overlaps across different PRSs were visualised and analysed using UpSet plots generated via R package ‘UpSetR’. Internal validation was performed through bootstrap resampling (1000 iterations) using the ‘boot’ package in R. For calibration analysis, predicted probabilities were divided into quintiles of equal frequency. Within each quintile, the mean predicted probability was plotted against the observed event rate (proportion of actual cases), with 95% CIs estimated from 1000 bootstrap resamples to account for sampling variability. Model discrimination was evaluated using the unadjusted AUC. A linear hypothesis test (Wald test) was employed to assess whether the ORs of the metaPRS, PRS SLE and PRS riskfactor differed in predicting cSLE versus aSLE. DeLong’s test was applied to compare AUCs between the metaPRS and PRS SLE models. The optimal threshold for clinical application was determined by maximising the Youden index (J=sensitivity+specificity−1) from the ROC analysis, providing a cut-off that best balances both sensitivity and specificity. Based on the distribution of the metaPRS or PRS SLE , individuals were stratified into low (first tertile), intermediate (second tertile) and high (third tertile) genetic risk groups. Logistic regression was then applied to assess and compare the risk-stratification performance of the two scores. Clinical characteristics were summarised as counts (percentages) for categorical variables and as means (SD) or medians (IQR) for continuous variables, depending on data distribution. The Kruskal-Wallis test was used to compare continuous variables across groups when distributions were non-normal. χ 2 or Fisher’s exact tests were applied for categorical comparisons. Logistic regression was used to evaluate associations between PRSs and clinical manifestations, with p values corrected using the FDR. A two-sided p<0.05 was considered statistically significant. All statistical analyses were performed using R V.4.4.1.

In this study, we developed a metaPRS for SLE in East Asian populations by integrating genetic data of SLE and four SLE-related risk factors. The metaPRS demonstrated improvement in predictive accuracy and enhanced risk stratification for SLE compared with conventional PRS based on SLE-associated variants alone. Within the metaPRS, the SLE-specific component (PRS SLE ) contributed more significantly to predictive capacity than the risk factor component. Furthermore, elevated metaPRS levels showed significant associations with clinical manifestations, particularly in specific autoantibody profiles, with these correlations primarily driven by PRS SLE . Notably, the metaPRS appeared to show a trend toward stronger predictive power in cSLE compared with aSLE. These findings underscore the clinical potential of the metaPRS as a valuable tool for early disease screening and personalised risk assessment, which could facilitate timely interventions and reduce the risk of organ damage in East Asian patients with SLE. So far, four PRS models have been developed for predicting SLE development in East Asian populations, 7 10 all of which were constructed using GWAS or meta-GWAS summary statistics for SLE alone, resulting in limited predictive performance. Hocaoǧlu et al subsequently developed an environmental risk score (ERS) comprising 37 risk factors and demonstrated that combining ERS with PRS could substantially improve prediction. 12 Another emerging strategy, the metaPRS approach, integrates genetic variants associated with related traits and has been successfully applied to several complex diseases such as coronary artery disease, 13 stroke, 20 intracranial aneurysms, 14 lung cancer, 15 type 2 diabetes 21 and bipolar disorder. 22 However, this strategy has not yet been explored for SLE. Here, we constructed the first metaPRS of SLE by incorporating additional SLE-associated genetic variants from related traits, achieving a 4.43% increase in OR over the SLE-only PRS, which shows clear potential for early risk stratification and personalised disease prediction. We initially evaluated 10 traits (including SLE and 9 previously reported risk or protective factors) encompassing lifestyle behaviours, essential nutrients, infections and comorbid conditions associated with SLE risk. However, only four traits demonstrated significant genetic correlations with SLE development: smoking, serum selenium levels, endometriosis and Graves’ disease. These findings are partially supported by a few MR studies. 23 25 Nevertheless, an MR study has reported no causal effect of smoking on SLE risk, 26 which may be related to differences in ancestry and the p value thresholds used for SNP inclusion. In our study, such an association was only observed when the p value threshold was relaxed to 0.003, resulting in the inclusion of 44 978 SNPs. Further research is required to resolve this inconsistency. Notably, the four risk factors validated in this study may contribute to the development of SLE through multiple mechanisms. First, different diseases may share a common genetic architecture. 27 For example, Graves’ disease 28 and endometriosis 29 may exert their effects by sharing specific risk loci with SLE. Second, gene–environment interactions also represent an important mechanism through which these environmental factors exert their effects. For example, Cui et al reported a strong additive interaction between an SLE genetic risk score and current or recent smoking. 30 Third, these risk factors intersect with the core pathophysiological processes of SLE. Soni et al demonstrated that selenium supplementation suppresses the pathogenesis of SLE in a mouse model by specifically inhibiting the activation and differentiation of B cells and macrophages, leading to reduced germinal centre B cells and autoantibody production, suggesting its potential as a therapeutic supplement. 31 Research also found that selenium supplementation alleviated lupus symptoms in mice by reducing oxidative stress–induced loss of intestinal epithelial γδT cells and restoring gut barrier integrity. 32 It is particularly noteworthy that, although some traditional risk factors (such as serum vitamin D levels, drinking and sleep disorders) did not show genetic associations in this study, this may be because these factors primarily influence SLE risk through acquired modifications, rather than innate genetic predispositions. This finding provides a new perspective for understanding gene–environment interactions in SLE. By integrating these diverse genetic influences, metaPRS provides a more holistic risk assessment than single-trait PRS. Our findings reveal a fundamental dichotomy in the genetic architecture of SLE: while both PRS SLE and PRS riskfactor independently contribute to disease susceptibility, their roles in shaping clinical manifestations are strikingly distinct. The PRS SLE component, derived from SLE-specific loci, not only exhibited a stronger association with disease development (OR=2.03 vs 1.26 for PRS riskfactor ) but also drove the majority of phenotypic expression. This suggests that PRS SLE captures core disease mechanisms—such as dysregulated immune pathways and end-organ targeting—that directly mediate clinical severity. In contrast, PRS riskfactor , despite aggregating genetic effects from four established risk factors, showed no association with any clinical features, implying that its contribution is limited to modulating susceptibility thresholds rather than phenotype determination. This dissociation aligns with emerging evidence that environmental risk factors may operate through distinct biological pathways (eg, epigenetic priming, microbiome dysbiosis) to initiate disease, while SLE-specific variants dictate its clinical course. The lack of association between PRS riskfactor and SLE clinical phenotypes may also be explained by the fact that the traits included in our study primarily capture overall SLE susceptibility rather than genetic mechanisms directly related to specific organ involvement. This intrinsic characteristic likely limits the ability of PRS riskfactor to discriminate among clinical subtypes. Future studies should focus on integrating trait-level PRSs that more directly capture the biology of specific organ systems. Using lupus nephritis as an example, incorporating PRSs related to lupus nephritis could improve the prediction of this SLE subphenotype. However, current predictive models for subphenotypes mostly rely on laboratory or demographic indicators, such as hypocomplementaemia and anti-dsDNA antibody, 33 which are challenging to translate into PRSs. Therefore, it may be more feasible to include PRSs associated with renal function or to identify potential loci through pathways implicated in the pathogenesis of lupus nephritis, thereby enhancing the predictive performance for SLE subphenotypes. Nevertheless, this null association should be interpreted with caution, as the relatively modest effect size of PRS riskfactor may have limited the statistical power to detect subtle phenotype correlations, raising the possibility of a type II error. Further studies in larger cohorts are warranted to validate this finding. In summary, these insights advocate for a stratified risk-assessment framework: PRS riskfactor could identify high-risk individuals for preventive measures, whereas PRS SLE may guide personalised monitoring for complications in diagnosed patients. We have to acknowledge that the increase in AUC from 0.68 to 0.69 is modest, even though the difference is statistically significant, but it is consistent with findings from other complex diseases. For example, the metaGRS for Alzheimer’s disease increased the AUC from 0.642 to 0.645 compared with the reference PRS model. 34 Similarly, the metaGRS for intracerebral haemorrhage improved the C-index from 0.686 to 0.695. 14 These results suggest that while integrating additional trait information provides measurable improvements, the predictive ceiling for complex diseases such as SLE remains difficult to surpass. Future work with larger sample sizes and more advanced analytical approaches will be needed to further investigate and improve predictive performance in this area. Nonetheless, the metaPRS retains meaningful clinical value. Our model effectively stratifies individuals into distinct genetic risk categories, enabling early identification of those at high risk for closer monitoring and preventive interventions before irreversible organ damage occurs. In clinical trials, the metaPRS could also serve as an efficient enrichment tool to identify participants with elevated genetic susceptibility, thereby enhancing statistical power to assess preventive strategies. Furthermore, integrating the metaPRS with clinical and serological markers may improve individualised risk prediction, guide tailored surveillance and management plans and ultimately support precision medicine approaches in SLE care. This study has several strengths. We conducted a case–control study in China with detailed clinical characteristics of enrolled patients with SLE, allowing for a comprehensive evaluation of the association between genetic risk and clinical phenotypes. To our knowledge, both existing risk factor-incorporated prediction models were derived from European-dominated public databases, 11 12 whereas our study specifically targets East Asian populations. Furthermore, this is the first study in the field of autoimmune diseases to integrate disease-related trait genetic information into metaPRS construction—an approach that may prove crucial for improving the prediction of complex, multifactorial disease. Third, our model combines the strengths of two established approaches: it retains the ability of conventional PRS to estimate disease risk from birth, while enhancing predictive accuracy through the integration of SLE-associated traits. This provides a promising tool for early diagnosis, intervention and risk stratification. However, several limitations should be acknowledged. First, the limited sample size prevented the use of an independent validation cohort and also resulted in a small number of cSLE cases. External validation in larger, independent cohorts is therefore needed in future studies. Second, the lack of individual-level environmental and lifestyle data (eg, smoking, serum selenium levels) limited our ability to assess how these exposures interact with genetic susceptibility in SLE development. Third, future research could focus on genetic features related to specific organ involvement, which may help better characterise distinct clinical subtypes of SLE.

We developed the first metaPRS in SLE by integrating genetic characteristics from multiple SLE-related risk factors, which outperformed conventional PRS in risk prediction and stratification. The metaPRS was associated with multiple autoantibodies. Decomposition of the metaPRS into PRS SLE and PRS riskfactor revealed that both components contributed to SLE susceptibility, while associations with clinical manifestations were exclusively driven by PRS SLE .