Evaluating and implementing block jackknife resampling Mendelian randomization to mitigate bias induced by overlapping samples

Fang, Shiqiang; Hemani, Gibran; Richardson, Tom G; Gaunt, Tom R.; Smith, George Davey

doi:10.1093/hmg/ddac186

Cited by 16 publications

(14 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Meanwhile, the block jackknife resampling framework requires access to individual-level phenotype and genotype data that we did not have a good way to evaluate. [42] Population overlapping could increase the possibility of false positives, but it was unlikely to affect our conclusion for our negative results.…”

Section: Discussionmentioning

confidence: 90%

Causal associations between estradiol and mouth ulcers: A Mendelian randomization study

Zhang,

Zhong,

Liang

et al. 2024

Medicine

View full text Add to dashboard Cite

People have difficulty in eating and speaking when they are suffering from mouth ulcers. Some studies suggest that estradiol is associated with the development and treatment of mouth ulcers, while some do not. To clarify the effect of estradiol on mouth ulcers, we performed 2-sample Mendelian randomization and multivariable Mendelian randomization (MVMR) analysis to evaluate their relationship. Data were obtained from the IEU OpenGWAS project and UK biobank, including male estradiol dataset (case/controls = 13,367/134,323), female estradiol dataset (case/controls = 37,461/126,524), mouth ulcers dataset (case/controls = 47,102/414,011). The causal associations were estimated by MR-Egger, weighted median, inverse-variance weighted (IVW) method, simple mode, and weighted mode. Cochran Q test, MR-Egger intercept test, MR-PRESSO tests, and leave-one-out analysis were used to examine sensitivity analyses. The MVMR controlling for depression, anxiety or panic attacks, severe stress and adjustment disorders was used to assess the effect of estradiol on mouth ulcers. Through screening, 13 single nucleotide polymorphisms (SNPs) of males and 2 SNPs of females in estradiol were used for harmonizing and MR analysis. The 2-sample MR analysis showed no causal association between estradiol of males and mouth ulcers (IVW, OR: 0.998, 95% confidence interval [95% CI]: 0.995–1.001, P = .18). Similar results were obtained between estradiol of females and mouth ulcers (IVW, OR: 1.000, 95% CI: 0.988–1.012, P = .97). No pleiotropy and heterogeneity were found and the results were robust (P > .05). After adjusting for the potential effects of confounders, estradiol of males and mouth ulcers still showed no causal association through MVMR analysis (P = .081). While MVMR analysis showed that the causal relationship between estradiol and mouth ulcers in women could not be statistical for the small number of SNPs. There was no evidence of a causal relationship between estradiol and mouth ulcers. The strategy of treating mouth ulcers with estradiol still needs to be confirmed by more studies.

show abstract

Section: Discussionmentioning

confidence: 90%

Causal associations between estradiol and mouth ulcers: A Mendelian randomization study

Zhang,

Zhong,

Liang

et al. 2024

Medicine

View full text Add to dashboard Cite

show abstract

“…Genetic IVs in a MR setting are conventionally selected from an independent dataset where the sample does not overlap with the dataset being analysed (53). This is to avoid overfitting bias.…”

Section: Methodsmentioning

confidence: 99%

A structural mean modelling Mendelian randomization approach to investigate the lifecourse effect of adiposity: applied and methodological considerations

Power,

Palmer,

Warrington

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

The application of a lifecourse approach to genetic epidemiology is key to better understanding causal effects of adversities on health outcomes over time. For some time-varying phenotypes, it has been shown that genetic effects may have differential importance in the development of an exposure at different periods in the lifecourse. Mendelian randomization (MR) is a technique that uses genetic variation to address causal questions about how modifiable exposures influence health. MR studies often employ conventional instrumental variable (IV) methods designed to estimate lifelong effects. Recently, several extensions of MR have been used to investigate time-varying effects, including structural mean models (SMMs). SMMs exploit IVs through g-estimation and circumvent some of the parametric assumptions of other MR methods. In this study, we apply g-estimation of SMMs to MR. We aim to estimate the period effects of adiposity measured at two different life stages on cardiovascular disease (CVD), type 2 diabetes (T2D) and breast cancer in later life. We found persistent period effects of higher adulthood adiposity on increased risk of CVD and T2D. Higher childhood adiposity had a protective period effect on breast cancer. We compare this method to an inverse variance weighted multivariable MR approach: a technique also using multiple IVs to assess time-varying effects, however, relying on a different set of assumptions and subsequent interpretations. We discuss the potentials and limitations of each approach and emphasise the importance of underlying methodological assumptions in the application of MR to lifecourse research questions.

show abstract

“…Although summary statistics from GWAS of both sepsis [ 24 ] and HDL subclass measures [ 70 ] are available already, the largest samples (by an order of magnitude) are in UK Biobank, leading to sample overlap in the exposure and outcome datasets, which leads to “winners curse”, biasing estimates [ 55 , 71 ]. In order to maximise sample size while reducing overfitting, we performed a sensitivity analysis using block jacknife resampling [ 72 ]. We split UK Biobank into multiple blocks using a block jacknife approach.…”

Section: Methodsmentioning

confidence: 99%

“…To measure outcomes in the remaining ~ 260,000 participants, we used a block resampling approach [ 72 ]. To do this we split this group into ten further samples, each with around 26,000 participants.…”

Section: Methodsmentioning

confidence: 99%

Low levels of small HDL particles predict but do not influence risk of sepsis

Hamilton,

Pedersen,

Ghazal

et al. 2023

Crit Care

View full text Add to dashboard Cite

Background Low levels of high-density lipoprotein (HDL) cholesterol have been associated with higher rates and severity of infection. Alterations in inflammatory mediators and infection are associated with alterations in HDL cholesterol. It is unknown whether the association between HDL and infection is present for all particle sizes, and whether the observed associations are confounded by IL-6 signalling. Methods In the UK Biobank, ~ 270,000 individuals have data on HDL subclasses derived from nuclear magnetic resonance analysis. We estimated the association of particle count of total HDL and HDL subclasses (small, medium, large, and extra-large HDL) with sepsis, sepsis-related death, and critical care admission in a Cox regression model. We subsequently utilised genetic data from UK Biobank and FinnGen to perform Mendelian randomisation (MR) of each HDL subclass and sepsis to test for a causal relationship. Finally, we explored the role of IL-6 signalling as a potential causal driver of changes in HDL subclasses. Results In observational analyses, higher particle count of small HDL was associated with protection from sepsis (Hazard ratio, HR 0.80; 95% CI 0.74–0.86, p = 4 × 10–9 comparing Quartile 4, highest quartile of HDL to Quartile 1, lowest quartile of HDL), sepsis-related death (HR 0.80; 95% CI 0.74–0.86, p = 2 × 10–4), and critical care admission with sepsis (HR 0.72 95% CI 0.60–0.85, p = 2 × 10–4). Parallel associations with other HDL subclasses were likely driven by changes in the small HDL compartment. MR analyses did not strongly support causality of small HDL particle count on sepsis incidence (Odds ratio, OR 0.98; 95% CI 0.89–1.07, p = 0.6) or death (OR 0.94, 95% CI 0.75–1.17, p = 0.56), although the estimate on critical care admission with sepsis supported protection (OR 0.73, 95% CI 0.57–0.95, p = 0.02). Bidirectional MR analyses suggested that increased IL-6 signalling was associated with reductions in both small (beta on small HDL particle count − 0.16, 95% CI − 0.10 to − 0.21 per natural log change in SD-scaled CRP, p = 9 × 10–8).and total HDL particle count (beta − 0.13, 95% CI − 0.09 to − 0.17, p = 7 × 10–10), but that the reverse effect of HDL on IL-6 signalling was largely null. Conclusions Low number of small HDL particles are associated with increased hazard of sepsis, sepsis-related death, and sepsis-related critical care admission. However, genetic analyses did not strongly support this as causal. Instead, we demonstrate that increased IL-6 signalling, which is known to alter infection risk, could confound associations with reduced HDL particle count, and suggest this may explain part of the observed association between (small) HDL particle count and sepsis.

show abstract

Evaluating and implementing block jackknife resampling Mendelian randomization to mitigate bias induced by overlapping samples

Cited by 16 publications

References 51 publications

Causal associations between estradiol and mouth ulcers: A Mendelian randomization study

Causal associations between estradiol and mouth ulcers: A Mendelian randomization study

A structural mean modelling Mendelian randomization approach to investigate the lifecourse effect of adiposity: applied and methodological considerations

Low levels of small HDL particles predict but do not influence risk of sepsis

Contact Info

Product

Resources

About