Dr. Anna O’Donnell is a Dean’s Postdoctoral Research Fellow in the Graduate Entry Medical School at the University of Limerick, Ireland. Anna’s research interests are in health economics and applied econometrics with a particular focus on personalised health and medicine, subgroup analysis, treatment effects in randomised controlled trials and machine learning methods.
Name: Dr. Anna O’Donnell
EHS Department of School you are affiliated to: Graduate Entry Medical School
Dean’s Fellowship Research Area: Personalised Health and Medicine
Research Leader: Professor John Forbes
What attracted you to the EHS Dean’s Fellowship? I was particularly interested in the supports provided by the fellowship. In addition conducting innovative research that can vastly contribute to patient care and well-being, the EHS Dean’s fellowship offers opportunities to learn about progressing my research career particularly in areas of grant writing and funds sourcing. These are important skills to have as a professional researcher and I look forward to properly learning about the protocols and avenues that ensure future research is financially viable.
In addition to providing supports that are essential for an early stage researcher, the fellowship fosters an environment of research collaboration. By bringing together five researchers with multidisciplinary backgrounds, this provides opportunities to conduct research beyond our own boundaries of experience by building on each other’s strengths and insights. Ultimately, this fellowship appealed to me because it encourages us to build a research portfolio that is financially secure for future prospective research and is diverse in areas of expertise achieved through collaboration and partnership.
Research Background (Education and work experience to date):
I was awarded a PhD in Economics with an emphasis in Health Economics from the University College Cork in 2016. Previously, I graduated with a BA Honours Degree in Economics from Metropolitan State University and an MSc in Economic Development from the University of Glasgow. My PhD focused on establishing the determinants of subjective well-being of Irish workers during the economic recession. Econometric issues such as sample selection bias and endogeneity bias were addressed in identifying the relationship between well-being and job insecurity for public and private sector workers. This research led me to present at a number of international and Irish conferences as well as recent submissions of academic articles for publication that are currently under peer review.
While completing my PhD I taught a number of undergraduate and postgraduate economics courses at both the University College Cork and the University of Limerick that had a particular focus on data analysis and econometrics. It proved beneficial to incorporate my existing research into the classroom when explaining the link between economic theory and application to the real world. The reverse is also true whereby teaching these modules ensured fundamental econometric assumptions were adhered to in my research which proved essential when addressing issues of reverse causality and sample selection bias in my doctoral dissertation.
My interest in individual well-being ultimately led me to apply for the EHS Dean’s Postdoctoral Fellowship under the area of personalised health and medicine. I am keen to research and apply the innovative methods for identifying individual-level responses to medical treatments. This research has the potential to identify ways of streamlining tailored medical interventions by employing innovative data-driven methods that identify well-being enhancing outcomes for subgroups of the population that under standard subgroup analyses would be ignored or lost in population average effects.
Research Expertise (Your area of focus and contribution to your field):
Currently, standard subgroup analyses in randomised controlled trials (RCTs) examine whether treatment effects vary according to patient characteristics, method of administering treatment, or approach to measuring the outcome. The aim of these analyses is to assess whether these indicators moderate or mediate the treatment effect by identifying subgroups of patients with similar characteristics. These ad hoc subgroups have been shown to yield biased and inconsistent estimates making it difficult for clinicians, investigators and policymakers to believe the apparent effect of the given treatment.
An approach that is more desirable for identifying patient subgroups and individual-level treatment effects is machine learning methods. The objective of these methods is to gain new insight from the data, without specifying a priori assumptions. These data-driven methods are beneficial when identifying baseline covariate profiles of specific patients who benefit from a treatment, rather than standard subgroup analyses which tend to only explain a small fraction of the individual-level treatment effect.
The innovation of this research lies in the fact that machine learning methods have only recently crossed over from computer science to economics and medicine. Traditionally, in order to calculate a causal treatment effect comparisons were made between outcomes observed and the counterfactual outcome under a different treatment. However, this is not feasible when identifying individual-level treatment effects because the counterfactual outcome is not known. Machine learning methods are being developed that partition data into subpopulations that inherently differ in the magnitude of their treatment effects. The algorithm can further be used to identify individual causal effects based on these partitioned subpopulations without knowing the “ground truth” or counterfactual. Data-driven techniques like machine learning methods allows the data to build models free of assumptions that balance the trade-off between bias and variance. The benefits of machine learning methods are slowly making their way into economic discourse and are on a path to eventually become the gold standard for building and testing empirical models.
Context of your Research (Area of application or real world impact):
The amount of data coming from RCTs is growing in both size and quality resulting in an increased interest in personalised statistical analyses. The increased ability of technology to collect and store big datasets has required the application of techniques and algorithms commonly found in computer science. Under the umbrella of personalised health and medicine, the ability to exploit the richness of available data to identify individuals believed to benefit from treatment is crucial for all stakeholders in patient care such as clinicians, policymakers and healthcare payers. This, however, is not an easy task as it requires a revolution in thinking towards patient-centred diagnosis and treatment. In the words of Hippocrates “It is far more important to know what person the disease has than what disease the person has.”