Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.


Estimates of the marginal effect of measures of adiposity such as body mass index (BMI) on healthcare costs are important for the formulation and evaluation of policies targeting adverse weight profiles. Many existing estimates of this association are affected by endogeneity bias caused by simultaneity bias, measurement error and omitted variables.

The contribution of this study is to avoid this bias by using a novel identification strategy – random germline genetic variation in an instrumental variable analysis – to identify the presence and magnitude of the causal effect of BMI on inpatient hospital costs. We use genetic variant-level data to undertake much richer testing of the sensitivity of results to potential violations of the instrumental variable assumptions than is possible with existing approaches. Using data on over 300,000 individuals, we found effect sizes for the marginal unit of BMI over 50% as large as multivariable effect sizes. These effects attenuated under sensitivity analyses, but effect sizes remained larger than multivariable estimates for all but one estimator. There was little evidence for non-linear effects of BMI on hospital costs. Within-family estimates, intended to address dynastic biases, were null but suffered from low power.

This paper is the first to use genetic variants in a Mendelian Randomization framework to estimate the causal effect of BMI (or any other disease/trait) on healthcare costs. This type of analysis can be used to inform the cost-effectiveness of interventions and policies targeting the prevention and treatment of overweight and obesity, and for setting research priorities.


Padraig is currently an MRC research fellow. His three-year programme of work (2017-2020) uses Mendelian Randomization to study the causal effect of obesity, coronary artery disease and other health conditions and traits on healthcare costs and on quality of life. Padraig holds degrees in Economics from Trinity College Dublin (BA) and Nuffield College, Oxford (MPhil and DPhil), and in Health Economics from the University of York (MSc). His interests include applied economic evaluations and related research undertaken alongside randomised controlled trials and in the context of observational research designs.

Forthcoming events

Infectious Disease Seminar Series: Hepatitis B diagnosis, prevention and treatment: laboratory approaches to the elimination agenda

Monday, 06 February 2023, 1pm to 2pm @ BDI Seminar Room LG 0-1, Old Road Campus, Headington, OX3 7LF

Richard Doll Seminar: Edgar Sydenstricker: Household Equivalence Scales and the Causes of Pellagra

Tuesday, 07 February 2023, 1pm to 2pm @ Richard Doll Lecture Theatre, Richard Doll Building, Old Road Campus, OX3 7LF

Ethox seminar- Feminist-Ethical Perspectives on Digital (Health) Technologies

Tuesday, 14 February 2023, 11am to 12.30pm @ Big Data Institute, Lower Ground Seminar Room 1, Oxford Population Heath, University of Oxford

Richard Doll Seminar - E-Freeze trial results

Tuesday, 14 February 2023, 1pm to 2pm @ Richard Doll Lecture Theatre, Richard Doll Building, Old Road Campus, OX3 7LF

Infectious Disease Seminar Series: Informing on Neisseria gonorrhoeae treatment and management through pathogen genomics

Monday, 20 February 2023, 1pm to 2pm @ BDI Seminar Room LG 0-1, Old Road Campus, Headington, OX3 7LF

Richard Doll Seminar- Triangulation of evidence in aetiological epidemiology: principles, prospects and limitations.

Tuesday, 21 February 2023, 1pm to 2pm @ Richard Doll Lecture Theatre, Richard Doll Building, Old Road Campus, OX3 7LF

Aetiological epidemiology is concerned with the identification of causal influences on disease risk. Randomized controlled trials are, when possible, the cornerstone of knowledge as to whether interventions based on aetiological studies are merited. It is not feasible to subject all of the many candidate causes to large-scale RCTs, however, even in situations where they are in principle possible. Triangulation of evidence is an approach that attempts to formally combine findings from different domains to strengthen causal inference. Triangulation embraces the variety of evidence thesis, that inferential strength depends not only on the quantity of available evidence, but also on its variety: the greater the variety, the stronger the resulting support. An essential condition is that the systematic errors and biases are unrelated across different study types. For example, the effect of raising circulating HDL cholesterol on the risk of coronary heart disease can be estimated from RCTs or through Mendelian randomization using genetic variants related to HDL level. Both the results of RCTs and Mendelian randomization studies could be biased. However, the potential biases in one study design would not influence estimates from the other approach: the biases are unrelated to each other. In observational epidemiology approaches that can be applied include the use of negative control exposures or outcomes; the deliberate use of data from contexts in which confounding structures differ; the use of instrumental variables and related approaches, such as regression discontinuity; quasi-experimental studies; the estimation of the expected magnitude of associations generated by confounding and the incorporation of mechanistic data, amongst others. Pre-registration of protocols for the triangulation of evidence increases confidence in the findings produced.