Automatically mapping and assessing inequalities in public health research

RESEARCH QUESTION

Can text mining be used to maintain a ‘living’ database of public health research, including information about topics, methods and inequalities? BACKGROUND

The ability to find and interpret evidence from a complex research literature is key to effective decision-making in public health, identifying effective interventions, and understanding their impact on inequalities. The heterogeneous nature of public health evidence makes these tasks more difficult than for other areas of medicine (e.g. clinical) while evidence about equity, an important factor in decision-making, presents particular challenges.

Automatic text mining methods have been shown to reduce the effort required to identify and assess evidence, but are less developed for public health evidence than other areas. AIMS AND OBJECTIVES

The project aims to develop and apply text mining methods to analyse the public health evidence base with a particular focus on equity. It will achieve this aim through the following objectives: - Develop a ‘living’ online database of public health studies, regularly updated with new research

- Carry out automatic analysis of the evidence base to provide additional information about the studies in the database: a) topics/themes discussed; b) PROGRESS-Plus dimensions; c) study type

- Engage with researchers, decision makers, funders, research commissioners, patients and the public to ensure that the project’s outputs are appropriate and useful METHODS

Achieving the project’s goals is possible because of substantial pre-existing research and development. We have a system providing a live flow of evidence from Microsoft Academic, and the ability to publish an online database of bibliographic data. This project will extend that work by developing and evaluating methods to:

- identify and automatically extract key data from public health research articles

- automatically rank research articles with reference to study design, relevance to public health, broad topic area, and health inequalities (PROGRESS-Plus dimensions)

- automatically model and present the topics of the identified studies, to help users to make sense of large volumes of research literature quickly

We will re-use existing data to 'train' machine learning models to recognise PROGRESS-Plus dimensions and the types of study being described, and use 'topic modelling' for theme identification. MILESTONES Month 2: Initial engagement with users Month 5: Development/assessment of methods to identify themes in public health research

Month 6: Development/assessment of tools to identify equity information (PROGRESS-Plus dimensions), including training data sets and computational tools (open access)

Month 9: Publication of freely available live system and evaluation by user groups. Development of final presentation to NIHR. IMPACT

The project will increase capacity for generating robust evidence of effectiveness by facilitating the efficient identification of relevant research and increasing the profile of inequalities in understanding effectiveness.

It will demonstrate the value of automated techniques for the identification and interpretation of public health evidence, thereby encouraging their use within complex decision-making problems. It will also advance text mining methods in public health.

Dissemination routes include the freely available online database, publications, presentations, and release of datasets and source code