Principal Investigator: Miss Louise Millard
Department: School of Social and Community Medicine
Institution: University of BristolTags: 16729, BMI, causality, Mendelian randomization, MR-pheWAS
1a: We aim to test a novel method for identifying potentially causal associations, which we call the Mendelian randomisation phenome-wide association study (MR-pheWAS) approach. We aim to identify potentially causal effects of BMI, as an exemplar. We have tested the MR-pheWAS approach using the Avon Longitudinal Study of Parents and Children (ALSPAC) dataset (with ~8K participants). Repeating this with Biobank will greatly improve statistical power, and means we investigate causal effects in adulthood. Research question: What novel, potentially causal effects of BMI can the MR-pheWAS approach identify? Health conditions: Body mass index, and several other traits. 1b: BMI is associated with a wide range of diseases. This project aims to improve understanding of the complex effects of BMI, by testing for the causal effects of BMI on a wide range of outcomes. Results from this project will prioritize a subset of hypotheses for replication and further investigation. Identifying potentially causal effects of BMI is important in order to inform policy makers on appropriate interventions. Interventions targeted at BMI are likely to impact a wide range of traits and diseases. This is becoming increasingly important as the UK is now experiencing increasing levels of obesity. 1c: We will use data of all participants from Biobank that have measurements for BMI and a set of genetic loci that have previously been found to be associated with BMI. We will test the causal effect of BMI on a wide range of outcomes. We will do this using a natural experiment (instrumental variable) constructed from the genome (a Mendelian randomization approach), to identify outcomes that may be affected by BMI. We use a hypothesis-free approach and so will test a large arbitrarily selected set of outcomes, rather than selecting particular outcomes. 1d: All participants with a value for BMI (var=21001) and BMI associated genetic loci. Extensions:
- Approved 03/08/2016:
|We would like to run MR-pheWAS analyses for the following traits (with field IDs): • Blood pressure • Education • Smoking • Female Puberty • Male Puberty • Height • Alcohol • Type 2 Diabetes • Genetic variants / phenotypes for which we would like to run the MR-pheWAS first stage only, that is the direct tests of genetic score on the set of outcomes, such that an actual variable for the phenotype is not needed – o AHRR methylation o HMGCoA o NPC1LI o PCSK9 • Blood measures – when the data becomes available|
- Approved 05/08/2016:
|We would like to run MR-pheWAS analyses for the following traits: • Prostate cancer|
- Approved 13/10/2016:
|We would like to run our pheWAS analysis with a set of additional genetic variants, provided in the attached file. We have evidence to suggest that these variants may have a regulatory impact on the S1PR1 gene, which we hypothesise to have a functional role on the aetiology of cardiovascular health. In the absence of any genetic associations with relevant traits thus far, we would like to undertake a pheWAS to potentially formulate new hypotheses and develop our understanding of S1PR1.|
- Approved 15/12/2016:
|We would like to submit an amendment, in order to investigate the usefulness of a new method for instrumental variable estimation based on a penalization-likelihood approach that was recently proposed by Kang et al (2016), on real Mendelian randomization studies. We will, as an exemplar, re-examine the causal relationship between BMI (exposure) and blood pressure (outcome) using the Kang (2016) method, as the positive causal relationship between these two variables is well-established. The hope is that the new robust method will still reaffirm this causal relationship, but now with robustness guarantees in case some of the genetic instruments are invalid. In addition, the new method can detect which instruments are invalid, which could be informative for future research using BMI variants in Mendelian randomization. Hence, this work will feed into our MR-pheWAS analyses by informing the choice of instrumental variable method to use, and also informing which genetic variants are invalid instruments and we may wish to remove from our analyses. The outcome is blood pressure (systolic or diastolic), the exposure is BMI, the instruments are genetic variants of BMI, and confounding variables are sex and age (with a quadratic term).|
- Approved 10/05/2017:
|We would like to perform a MR-pheWAS of Alzheimer disease, using a genetic risk score.|
COLLABORATING INSTITUTES: Lead Collaborator: Professor Guido Imbens Collaborating Institute: Stanford University Economics Graduate School of Business 655 Knight Way Stanford CA 94305-5015 United States of America Project extension: “We would like to investigate the causal link between severe mental illnesses (SMI; including schizophrenia, bipolar disorder and major depression) and cardiovascular disease (CVD, including conditions such as coronary heart disease, cerebrovascular disease and congestive heart failure), using a MR-pheWAS approach.” “We would like to perform an MR-pheWAS of non-response. Non-response refers to whether a participant responds to a data collection event (e.g. completing a questionnaire). We will use genetic variants that have been identified as being associated with non-response, as instrumental variables that proxy for being a non-responder. We will search for the causal effects of non-response using the MR-pheWAS approach.” “We would like approval to perform an MR-pheWAS of vitamin C.” Project extension: We aim to test a novel method for identifying potentially causal associations, which we call the Mendelian randomisation phenome-wide association study (MR-pheWAS) approach. We aim to identify potentially causal effects of BMI, as an exemplar. We have tested the MR-pheWAS approach using the Avon Longitudinal Study of Parents and Children (ALSPAC) dataset (with ~8K participants). Repeating this with Biobank will greatly improve statistical power, and means we investigate causal effects in adulthood. Research question: What novel, potentially causal effects of BMI can the MR-pheWAS approach identify? Health conditions: Body mass index, and several other traits. We will also perform MR-pheWAS of the following traits: ADHD Autism Schizophrenia Amino acids depression bipolar disorder lactase persistence Sex hormone-binding globulin testosterone Vitamin D Age at menarche We would like approval for the following: Collider effects describe artificial associations between variables that arise when conditioning on an outcome and can be of relevance within an MR-PheWAS setting. We argue that an investigation of collider effects might be used in reverse to assess whether variables are causally linked to an outcome. As a proof-of-concept, we would like to investigate the causal link between severe mental illnesses (ADHD, autism and major depression; phenotypically measured and genetically predicted) and birth weight (phenotypically measured and genetically predicted). We would like to run an analysis on all continuous variables to assess the impact of measurement error in Biobank. Firstly, using the observations with repeated assessments (circa 20k) we will estimate the scale of measurement error for each variable along with the corresponding level of statistical uncertainty by looking at the concordance between the initial and repeated measures. Findings will be applied to the whole Biobank database via correction factors to assess the potential effect of the measurement error on associations between the continuous exposures and standard outcome measures such as all-cause/CVD/Cancer mortality. In further analyses we will also assess the impact of measurement error on MR analyses by applying the correction factors, where applicable, to MR-pheWAS. As there are a very large number of continuous measures in UKB (~1000) we will develop an automated process for exploring agreement / concordance and generating the correction factors in the ~20K with repeat measures and consider a smaller subset (based on extent of error and potential relationship of that to confounders for observational analyses and violation of MR assumptions for MR analyses) for further analyses of the impact of bias. This work will also be applicable to other projects using Biobank data. We would like to perform a MR-pheWAS of miscarriage.
Age at start of school has been related to outcomes in school and beyond (e.g. grade averages, chances of being held back a year during school, income once left school). Previous analyses have used month of birth as an instrumental variable (in a regression discontinuity design) to assess the causal effect of age at school on education and economic outcomes. We propose to do a IV-PHEWAS in UKBB to investigate what other outcomes might be caused by age at school entry, using month of birth as the instrument. Identified significant associations will be taken forward for further investigation, including examining interactions with PRS for education and geographical area.
We would like to perform a MR-pheWAS of multiple sclerosis.
Last updated Jun 10, 2019