Im a statistician with interests incausal inference, techniques in medical statistics, statistical genetics, Bayesian statistics and decision theory. I developed an interest in statisticsthrough my undergraduate degree in actuarial science (BAFS, 2000). I then pursued a MSc. in statistics at UCD (2001), and stayed at UCD for a lecturing rolefrom 2001-2002. Based on my MSc. resultsI received a travelling studentship in Mathematical Science from the National University of Ireland, which I used to travel to the U.S. topursue a PhDin statistics at Yale University. During my PhD, I receiveda thorough grounding in both the theoretical and applied aspects of statistics. I received the Leonard J. Savage prize for statistical writing for myPhD dissertation, which focused on statistical analyses that show robustness properties over a range of choices of statistical loss (or risk) functions.After graduating with a PhD in statistics in2009, I undertook a joint postdoctoral role in statistical genetics and biostatistics with Professor Hongyu Zhao (in the biostatistics group atYale) and Professor Judy Cho in the genetics group. During this time, I developed statistical methodology to address a range of challenges in statistical genetics such as reducing winners curse bias in genetic studies, association tests that combine multiple rare variants in a gene region to improve statistical power and joint analysis of multiple histone libraries. Ialso worked in collaboration in geneticists and other scientists on a range of studies involving high throughput sequencing. After this postdoctoral role, I undertook a position (as a Visiting Assistant Professor) in George Washington University where I taught courses in statistical decision theory, probability and basic statistical inferenceat both postgraduate and undergraduate level.In 2013, I returned to Ireland. I worked at UL as a research fellow in biostatistics from 2013-2015.During this time, I developed novel statistical methods in record linkage that we used to link hospital laboratory records longitudinally over time, as cross sectionally to CSO records. I also worked as the principle statistician on a range of applied projects pertaining to the progression of chronic kidney disease.In 2015, I undertook a position as Senior Research Fellow, located at the Clinical Research Facility atthe University of Galway. Since 2015, I have developed statistical methods to better estimate population attributable fractions, a causal metric that refers to the level of disease in a population that would be avoided in the absence of a harmful exposure (such as smoking). I received a HRB Emerging Investigator award in 2017, to extend this work, with a focus on better incorporation known causal pathways (involving risk factors, confounders and disease)into these statistical methods. Many of the new methods that I(together with postdocs) developed as part of the award are implemented in the R package graphPAF, and are summarised in the associated paper in the European Journal of Epidemiology. In 2021, I assumed the position of lecturer (above the bar) in biostatistics in Medical Biostatistics located in Clinical Research Facility,and since January 2024 I have been lecturer (above the bar) in Statistical Science in the School of Mathematical and Statistical Sciences.

My research interests pertain to the development of improved statistical methodology, with applications in several applied domains, such as medicine, population health and genetics.Some areas of current focus are listed under headings below. Better confidence intervalsfor functions of multiple parameters : If we have an exact 95% confidence interval for a parameter, the method of substitution allows us to generate a 95% confidence interval for any monotonic function of that parameter by simply plugging the endpoints of the interval for the parameter into f ; as an example, think of how one typically generates a confidence interval for an odds ratio or risk ratio, by exponentiating a symmetric interval for the log-ratio. Propagation of imprecision, devised by Newcombe, 2011, extends this idea to real valued functions of multiple parameters. This approach has advantages over the Delta method as the intervals will usually be better behaved than symmetric intervals, when the distribution of is skewed, and does not require the arbitrary setting of a random seed to be reproducible, as would Bootstrap or MCMC approaches. In recent work, I have justified propagation of imprecision in more general settings than Newcombe recommended,including when estimators of individual parameters are correlated and when f is non-monotonic. Most importantly, I have introduced a new related approach `Approximate Propagation of Imprecision, an asymptotically equivalent algorithm that produces very similar intervals to Propagation of Imprecision, that removes the requirement of grid search in Newcombes original algorithm, which facilitatesapplication of the method for functions of K gt;3 parameters. Approximate propagation of imprecision is quick, even in large dimensions, automatic (relying on automatic differentiation algorithms), and easy to implement. Estimation of PAF: Population attributable fractions (PAF) are estimates ofthe probable level of disease in the complete absence of a disease risk factor. For instance, one might like to know what would be the incidence of lung cancer in Ireland if nobody smoked. In 2017, I received a HRB-emerging investigator award for work in causal inference applied to population health metrics like PAF. I have since developed novel estimation approaches for types of attributable fractions, using causal Bayesian networks (for the estimation of joint and sequential attributable fractions) mediation techiques(for pathway-specific PAF), and motivated new pertinentestimands depending that estimate the contribution of differing mechanisms by which a risk factor might cause disease. Ive also worked on other ideas related to attributable fractions. For instance, two papers regarding PAF have been recently published in the European journal of epidemiology. The first relates biases that may result from using Levins approach, which is a very old formula used to estimate PAF in epidemiology but which is only applicable in very limited circumstances, and another relating to graphPAF, a R package I wrote to perform a wide range of attributable fraction calculations and inference, under a causal inference framework. Eliminating Winners Curse, with applications to 2-sample Mendelian Randomisation: In 2-sample Mendelian randomization (MR), causal effects between the exposure trait (say body mass index) and outcome (say cardiovascular disease) are estimated by meta-analysing ratios of genetic effects: ab , with b representing a statistical association of the instrument, a genetic marker, on the exposuretrait, and a, the association between the same marker and the outcome, with a and b estimated in genetic studies. Usually the instrument SNPs that are meta-analysed are selected according to a threshold for genome-wide statistical significance, and as a result they incur a Winners Curse: estimated associations from a sample under selection will tend to be biased-above for their population counterparts. This bias propagates in a complex way into MR estimates, interweaving with other biases like weak instrument bias. Im currently working on methods to minimise or eliminate Winners curse bias in Mendelian randomisation. In the past, Ive worked on methods (in particular Empirical Bayes approaches) to reduce the effect of this same bias on the SNP-disease association estimates. Some recent work, in PLOS genetics, also investigates a novel Bootstrap estimator to reduce the effect of Winners curse on GWAS estimates. Methods for analysingGWAS data jointly over several related traits: Many diseases and traits share common genetic backgrounds, examples being psychiatric disorders, such as schizophrenia or bi-polar disorder, and autoimmune diseases, such as Type I diabetes, Crohns disease and ulcerative colitis. However, often GWAS datasets for differing diseases are analysed individually and independently. Im currently developing novel Empirical Bayes approaches for combined analysis of Genome Wide association data for several traits. The method effectively converts p-values at a given genetic marker (which are independently estimated for differing related traits using standard approaches) to posterior probabilities of association that correctly incorporate additional information in related traits. In addition to improving statistical power to find genetic associations, other favourable properties of the proposed approach include conceptual simplicity, relative computational speed, reliance only on summary associations and p-values - which are more often available than individual level data, an ability to incorporate data from both case control and cohort study designs, and a coherent method to identify markers having particular association patterns. For instance, a geneticist may be interested in markers that are very likely to be associated with Heart disease, but not associated with BMI. The method will generate posterior probabilities for each possible association pattern at a marker. Initial tests of the method, using UK Biobank GWAS data on the traits height, BMI and Type I Diabetes, shows promising performance.