Genetic Liability and Early-Life Exposures Shape the Timing of Multiple Sclerosis Diagnosis: Evidence From a UK Biobank Survival Analysis
Nova and colleagues reframe the study of multiple sclerosis (MS) risk factors by moving from conventional retrospective case–control comparisons (typically summarized via odds ratios) to a prospective time-to-event framework that models instantaneous risk (hazard) across the life course. This distinction is not cosmetic: logistic regression implicitly collapses risk across heterogeneous follow-up times and assumes a constant effect over age, whereas survival analysis naturally accommodates censoring and enables explicit testing of age-varying effects—an especially relevant feature for MS, whose clinical manifestation and ascertainment can vary substantially by life stage.
Cohort, outcome definition, and exposure architecture
Using the UK Biobank, the authors constructed a longitudinal cohort restricted to unrelated participants self-identifying as White and born (and recruited) in England, following individuals from birth through 31 December 2022. MS cases were defined using ICD-10 code G35, with the earliest date drawn from linked hospital, primary care, death registry, or self-report sources; age at diagnosis was calculated from date of birth and used as a proxy for onset. Genetic liability was represented by a standardized MS polygenic risk score (MS-PRS, including the HLA region), while early-life factors included sex, birth year, sibship structure, maternal smoking around birth, breastfeeding, multiple birth status, and geographic cluster of birthplace; BMI genetic predisposition was included via a standardized BMI-PRS. Smoking initiation and infectious mononucleosis (IM; ICD-10 B27) were modeled as time-varying exposures to reduce immortal time bias.
Statistical strategy: IP-weighted Cox modeling and diagnostics
The primary analysis employed an inverse-probability (IP) weighted Cox proportional hazards model with age as the time scale, a deliberate choice to mitigate selection effects known to influence UK Biobank estimates. Missing covariate data were handled via multiple imputation by chained equations, with Rubin’s rules used for pooling across imputations. The investigators explicitly evaluated proportional hazards and linearity assumptions, introducing stratification or time-varying coefficients when proportional hazards were violated and using restricted cubic splines where needed. Model discrimination was assessed using Harrell’s C-index with bootstrap optimism correction, supporting an internally validated risk-ranking performance.
Central finding: age-dependent effects of sex and genetic burden
Among 345,027 participants (1,669 MS diagnoses), the authors report that both sex and MS-PRS exhibit age-dependent associations with MS diagnosis hazard, with larger hazard ratios observed in younger adults. The interpretation is clinically meaningful: elevated polygenic burden is not merely associated with higher lifetime probability of MS, but also with an apparent shift toward earlier diagnosis (used here as a proxy for earlier onset), and the female excess risk appears most pronounced at younger ages and attenuates later in life. This directly supports the paper’s core methodological claim: retrospective analyses that enforce constant effects may underestimate the magnitude of genetic and sex-associated risk during the age windows most relevant for MS emergence.
Environmental and early-life correlates: smoking, IM, BMI genetic propensity, and seasonality
The model corroborates established correlates of MS susceptibility within a survival framework: ever smoking on most or all days and prior IM diagnosis were associated with materially increased hazards, as was higher BMI-PRS (notably, higher values rather than low-to-average values). The analysis also identifies elevated hazard for being born outside the fall season (spring/summer/winter vs fall), consistent with hypotheses linking gestational sunlight exposure and maternal vitamin D status to later MS risk. In contrast, breastfeeding, maternal smoking around birth, multiple birth, and number of older siblings did not achieve statistical significance in the primary model, though the sibling effect trended toward modest protection.
Interaction and risk projection: combining MS-PRS with IM and smoking
Beyond main effects, the study interrogates gene–environment interplay on both additive and multiplicative scales. The authors report significant positive additive interactions between higher MS-PRS and several factors—female sex, IM diagnosis, later birth cohorts, and smoking—suggesting that combined exposures may yield more risk than would be expected from simple addition of individual effects (as quantified by measures such as RERI and attributable proportion). Translating model parameters into projected cumulative incidence, the paper illustrates how MS-PRS percentiles, together with assumed ages of smoking initiation and IM diagnosis, stratify lifetime diagnosis probability; for example, the most extreme combinations (very high MS-PRS plus IM and smoking) yielded markedly higher predicted cumulative incidence up to older ages, with higher absolute risk estimates in females than males under the modeled scenarios.
Interpretation, limitations, and implications for MS epidemiology and prediction
Methodologically, the work strengthens the argument that survival models should be used more routinely in MS etiologic research, particularly when effects plausibly vary with age; it also motivates “survival GWAS” approaches to detect variants whose effects are concentrated in earlier life stages. However, several limitations condition interpretation: diagnosis date was used as an onset proxy (implicitly assuming minimal and relatively uniform diagnostic delay), case definitions relied on administrative records and self-report (though sensitivity analyses excluding self-report-only diagnoses were consistent), and generalizability is constrained by the England-born White sample and by residual selection bias even after IP weighting. Finally, IM was used as an imperfect proxy for EBV-related immunologic exposure, omitting asymptomatic seroconversion. Despite these constraints, the study provides a rigorous template for integrating polygenic risk with time-indexed exposures to estimate age-specific MS risk and to prioritize prevention-oriented stratification in higher-risk groups.
Disclaimer: This blog post is based on the provided research article and is intended for informational purposes only. It is not intended to provide medical advice. Please consult with a healthcare professional for any health concerns.
References:
Nova, A., Di Caprio, G., Bernardinelli, L., & Fazia, T. (2024). Genetic and early life factors influence on time-to-multiple sclerosis diagnosis: A UK Biobank study. Multiple Sclerosis Journal, 30(8), 994-1003.
