Synthesizing evidence on different outcome scales: development of mapping methods for analysing childhood obesity data

Obesity in childhood is a central theme in life course epidemiology. As well as middle-age mortality, childhood obesity is associated with type 2 diabetes and heart disease in adulthood [1]. Obesity also tracks into adulthood where it is associated with increased risks for other conditions such as metabolic syndrome and some cancers [2].

Childhood obesity measurements can be reported in terms of standardized (against age and sex) and non-standardized BMI [3,4]. To allow coherent analysis of this data, we require methods to `map’ these outcomes onto a single scale. This poses a challenge when we only have access to aggregate (e.g. means) rather than individual data. This scenario is common in meta-analyses of childhood obesity, where we aim to combine data from multiple studies of the same question. Other continuous measurement scales also pose similar issues, giving this topic broad applicability.

Current strategies for dealing with this problem are restricted to crude standardization and ratio techniques. Limitations of these techniques include the requirement for non-negative measurement scales and the assumption that the different scales are linearly related, both of which are not fulfilled by BMI scales [5]. Therefore, the analysis of childhood obesity requires more sophisticated mapping methods.

The main aim is to develop novel sophisticated techniques for mapping between BMI measurement scales. Specific objectives include:

1. to examine and improve existing methods;
2. to refine and generalise sampling and optimization techniques using distributional assumptions and knowledge of individual-level scale conversions;
3. to compare the sophisticated methodologies with existing ‘crude’ strategies such as standardisation approaches and apply to real data sets.

Earlier work has produced preliminary methods for mapping between aggregate measures of standardized and non-standardized BMI. These use naïve sampling and optimization techniques relying on distributional assumptions and known relationships between individual measurements. The student will first improve and develop these methods in the context of BMI measurement scales. They will generalize the methods to any continuous outcome scale, assuming knowledge of individual-level mapping schemes and approximate distributions of the scales. New methods developments will then remove the requirement for distributional assumptions and knowledge of individual-level relationships. The new methods will be tested using simulation studies and real datasets and compared with simpler approaches such as standardization. We will re-analyse an existing childhood obesity dataset using the new mapping methods.

How to apply for this project

This project will be based in Bristol Medical School - Population Health Sciences in the Faculty of Health Sciences at the University of Bristol.

If you have secured your own sponsorship or can self-fund this PhD please visit our information page here for further information on the department of Population Health Science and how to apply.