Statistical analysis
We initially conducted cross-sectional logistic models in the observational models (Figure 1). To investigate whether a factor is a causal factor for asthma, we conducted MR two-stage least squares (2SLS) and MR sensitivity analyses for all relationships. MR 2SLS analysis is a fundamental method in MR for both binary and continuous variables of exposure and outcomes22. MR sensitivity analyses (MR inverse-variance weighted, Egger [slope], and weighted median method) were used to support the validity of the causal inference from the MR analysis23. To elucidate whether various factors predict incident asthma in later life, we also applied a discrete-time hazard model (DTHM). The DTHM is a survival analysis model that enables the estimation of hazard ratios (HRs) and 95% confidence intervals (CIs) when an event occurrence may be considered in a discrete period; baseline information on these factors was collected on disease-free individuals at age 11, and the outcomes were incident asthma at ages 12 and 17. The observational models and DTHM were adjusted for several potential confounders, such as age, sex, parental education, and family income. Odds ratios (ORs), HRs, and beta values were estimated using the IQR increase of the factors.
The GRS for a risk factor can be considered a genetic proxy. To compare the effect of these factors on asthma, as identified from observational and MR analyses, we further utilized the GRS to calculate the predicted prevalence of asthma. We computed the sample mean per quintile for each of the GRS to serve as the new dataset for prediction by logistic regression. To estimate the predicted probability of the GRS of an obesity-related factor, we included the means per quintile for asthma. All model regression and sensitivity analyses were performed using R software and R package (ivreg) MR 2SLS analysis (version 3.3.2, RStudio, Inc., Boston, MA, USA).