To the Editor: Late-onset multiple sclerosis (LOMS) is characterized by disease onset after the age of 50 years and comprises 5–10% of the population with multiple sclerosis (MS). Compared to typical adult-onset MS, LOMS presents greater diagnostic difficulty, accelerated disability progression, and poorer response to disease-modifying therapies, leading to increased disease burden for individuals and healthcare systems.1,2 Given the multifactorial etiology of MS involving complex gene–environment interactions, potential preventive measures specifically for LOMS remain particularly understudied.3 This gap underscores the need for a deeper understanding LOMS-specific modifiable risk factors to develop effective preventive measures and targeted public health interventions. Herein, this study used comprehensive data from the UK Biobank to systematically identify modifiable factors associated with LOMS, quantifying population attributable fraction (PAF)-based effects of modifiable domains on LOMS, and aiming to propose priority and synergetic strategies for LOMS prevention and improvements. To explore modifiable risk factors for LOMS, we conducted a UK Biobank-based nested case-control study, including 417 incidents of patients with LOMS and 1668 matched controls for analysis. A total of 275 modifiable factors from nine domains including sociodemographics, lifestyles, health condition, early life factors, local environment, cognitive function, psychosocial factors, biological measures, and physical measures were identified Supplementary Figure 1, https://links.lww.com/CM9/C876 and Supplementary Tables 1–3, https://links.lww.com/CM9/C877. Ethical approval of this study was obtained from the Northwest Multiple Centre Research Ethics Committee (No. 94533), and all participants provided informed consent. At the genetic level, the genome-wide association studies (GWAS) summary data for potential modifiable factors were obtained from UKBB GWAS Imputed v.3, and the GWAS data of MS was sourced from International Multiple Sclerosis Genetics Consortium, including 47,429 cases and 68,374 controls. Detailed information of participants inclusion, variables selection, and genetic data source is shown in Supplementary Methods, https://links.lww.com/CM9/C876. The 275 modifiable factors were transformed into indicator variables facilitated subsequent interdomain comparisons for eliminating dimensionality. Multiple imputation generated five imputed datasets and regression results from imputed datasets were summarized. Conditional logistic regression models were extensively used to assess the association between 275 modifiable factors and LOMS, adjusted for age, ethnicity, and assessment center to account for potential confounding effects, and the false discovery rate (FDR) less than 0.05 were considered significant. Gender-stratified analyses were conducted as sensitivity analysis, and mendelian randomization (MR) was used to further assess the causal association between modifiable factors and MS. The identified LOMS-associated modifiable factors were then grouped into distinct domains, and domain-specific scores (based on the effect direction and β coefficient in the regression model) were calculated and divided into three groups (unfavorable, intermediate, and favorable) for further intradomain and interdomain analyses. PAF was calculated to assess each domain’s contribution to the LOMS onset risk through the hypothetical scenario of eliminating the unfavorable or intermediate risk factors. Furthermore, the joint effect among the domains was analyzed for interdomain factors, and the protective effects of improvement in different combinations of domains were assessed. Finally, mediating analysis was conducted to analyze possible pathways by which modifiable factors may influence the onset of MS. All statistical analyses were performed using R software (v.4.2.1, R Development Core Team, Vienna, Austria) and STATA/MP (v.17.0, Stata-Corp LP, College Station, TX, USA). Further details of the analysis are provided in the Supplementary Methods, https://links.lww.com/CM9/C876. The study design was shown in Supplementary Figure 2, https://links.lww.com/CM9/C876. In this study, the majority of participants were white females, with a median age of 56 years at baseline and a median follow-up of 6.47 years for LOMS patients Supplementary Table 4, https://links.lww.com/CM9/C877. Thirty-eight LOMS-associated modifiable factors spanning six domains, including sociodemographics, lifestyle, health condition, psychosocial factors, and biological and physical biomarkers, were identified. Of these, 15 demonstrated protective effects, whereas 23 exhibited detrimental associations. Specifically, left-hand grip strength showed the strongest protective effect (odds ratio OR: 0.390; 95% confidence interval CI: 0.262–0.580; FDR <0.001) and allowance receiving showed demonstrated the most significant harmful association (OR: 3.707; 95% CI: 2.609–5.267; FDR <0.001). Figure 1A and Supplementary Figure 3, https://links.lww.com/CM9/C876. These findings were further validated through data imputation and regression of the original continuous variables Supplementary Figure 4, https://links.lww.com/CM9/C876 and Supplementary Table 5, https://links.lww.com/CM9/C877. The stratified analysis showed that the factors identified in females mirrored those found in the overall population, both in direction and magnitude, while males exhibited only partial overlap Supplementary Figure 5, https://links.lww.com/CM9/C876 and Supplementary Table 5, https://links.lww.com/CM9/C877.Figure 1: (A) Manhattan map shows the effect of each modifiable factor in nine categorical domains on late-onset multiple sclerosis risk. The red horizontal dotted line showed the significance threshold of FDR <0.05. (B) Weighted population attributable fraction for the six domains. Each point represents the generated population attributable fraction for each imputed dataset. (C) The joint association of sociodemographics, lifestyle, and health condition with late-onset multiple sclerosis. Dots represent odds ratios, and vertical lines indicate corresponding 95% confidence intervals. * P <0.05. FDR: False discovery rate.The MR analyses revealed significant causal associations in the domains of sociodemographics, lifestyle, health condition, and biological measures with LOMS risk Supplementary Figure 6, https://links.lww.com/CM9/C877 and Supplementary Table 6, https://links.lww.com/CM9/C877. Notably, the beneficial effect of vitamin D on MS was consistently observed across various instrumental variable selections and multiple MR methods (OR: 0.593–0.740; P = 0.012), and the effects of physical activity remained robust. MR analyses conducted separately for male and female populations were shown in Supplementary Tables 7 and 8, https://links.lww.com/CM9/C877. Domain-specific LOMS risks were assessed using 30 identified modifiable factors, consolidated across imputed datasets, with correlation checks ensuring unbiased scoring Supplementary Figure 7, https://links.lww.com/CM9/C876. The analysis for unweighted and weighted risk scores both revealed that higher risk levels within the domains of sociodemographics (OR: 1.633; 95% CI: 1.157–2.305; P = 0.005), lifestyle (OR: 1.713; 95% CI: 1.268–2.315; P <0.001), health condition (OR: 2.813; 95% CI: 1.993–3.971; P <0.001), and physical measures (OR: 1.598; 95% CI: 1.182–2.162; P = 0.001) are associated with an increased risk of LOMS in mutual adjustment model Supplementary Figures 8 and 9, https://links.lww.com/CM9/C876. The PAF analysis was conducted to assess the risk weights of each domain. The results showed that the PAF for each domain was as follows: health condition (19.0%), lifestyle (15.1%), physical measures (12.4%), sociodemographics (11.8%), biological measures (11.8%), and psychological factors (9.50%), in descending order Figure 1B and Supplementary Table 9, https://links.lww.com/CM9/C877. Moreover, when considering the combination of health condition, lifestyle, and sociodemographics, achieving high levels in lifestyle, health condition, and improvements in socioeconomic indicators, provided the maximum protective effect against LOMS onset (OR: 0.249; 95% CI: 0.148–0.418; P <0.001) Figure 1C and Supplementary Figure 10, https://links.lww.com/CM9/C876. Mediation analyses showed biological indicators, such as vitamin D levels and neutrophil counts, mediated the effect of modifiable factors on LOMS risk both epidemiologically and genetically Supplementary Figures 11 and 12, https://links.lww.com/CM9/C876 and Supplementary Table 10, https://links.lww.com/CM9/C877. Specifically, the overall health rating had an indirect effect on MS through neutrophil count (β = 0.054, P = 0.005), and physical activity lowered MS risk by vitamin D (β = −0.043, P = 0.008). The extensive nested case-control study identified several LOMS-associated modifiable factors across different domains. In the sociodemographic domain, socioeconomic indicators may bidirectionally affect LOMS risk by influencing both disease onset and postdiagnosis management, though the current evidence remains uncertain. Future research should consistently analyze their combined impact on LOMS beyond isolated factors. In health condition and lifestyles domain, our study corroborated prior evidence that frailty, comorbidities, and low physical activity elevate LOMS risk, while exercise and restorative sleep are protective.4 These findings align with observed risk reductions through lifestyle modifications, thereby underscoring the value of health monitoring and targeted behavioral interventions for prevention. In addition, our findings also reinforced the significance of psychosocial factors, particularly anxiety and depression, suggesting that psychosocial factors may act as early warning signs of LOMS. In biological measurements, vitamin D deficiency and immune cell dysregulation were identified as key LOMS biomarkers, offering actionable targets for early detection.5 Our findings highlight multidimensional domain contribution in LOMS risk and recognized impactful modifiable factors for providing targeted interventions guidance. We have shown that optimizing lifestyle and health status, alongside targeted socioeconomic support, yields the greatest protective effects against LOMS. We hope that these findings will inform the development of effective public health initiatives. This study has several key strengths. The age distribution of the UK Biobank participants is particularly advantageous for studying LOMS, and its prospective design addresses limitations in existing cross-sectional studies. Additionally, rigorous data collection reduced attrition bias, and PAF analysis offered actionable insights for translating scientific evidence into practical preventive measures. However, the potential for “healthy volunteer” selection bias and overrepresentation of White ethnicity individuals in the UK Biobank dataset may constrain generalizability of our findings to the broader population, and the sample size constraints further classification into specific MS phenotypes, such as secondary progressive multiple sclerosis and primary progressive multiple sclerosis. Future studies should aim to address these limitations by including more diverse populations and validate these findings in large prospective cohort study. In conclusion, the onset of LOMS is associated with several modifiable factors, including sociodemographics, lifestyle, health condition, psychosocial factors, biological measures, and physical measures. The estimation of PAF emphasizes the importance of promoting health condition, lifestyle, and social support. Moreover, adjusting lifestyle and health status to the upstream level, coupled with socioeconomic support, yields the most substantial protective benefits against LOMS onset and should be prioritized in public health decision-making. Funding This study was funded by STI2030-Major Projects Grant (No. 2022ZD0204704), and the National Natural Science Foundation of China (Nos. 82371404 and 82271341). Conflicts of interest None.
Chen et al. (Mon,) studied this question.