Supporting Evidence-based Public Health Interventions using Text Mining

Evidence-based public health (EBPH) reviews play a central role in public health policy, practice and guidance. Their development currently involves first searching, then screening and synthesizing evidence from the vast amount of literature. Unlike systematic reviews, EBPH reviews require dynamic and multidimensional views of relevant information from the literature, without relying on a priori research questions.
As a result, EBPH reviewing is a time consuming and resource intensive process that can take more than a year to complete. Since crucial information can be difficult to locate, and indeed understand given the complex nature of EBPH problems, the multiple causes and interrelations between interventions, diseases, populations and outcomes can remain hidden.

This project will address these limitations by exploring new research methods, which combine text mining and machine learning to produce novel search while screening tools for public health reviews. Text mining methods will discover automatically knowledge from unstructured data and machine learning will support the prioritisation and ranking of the extracted information into meaningful topics. The combination of text mining and machine learning methods will reduce the burden of producing public health reviews which will be completed more quickly, thus meeting policy and practice timescales and increasing their cost efficiency. They also allow more timely and reliable reviews, thus improving decision making across the health sector.

The project will be informed throughout by close interaction with the Centre of Public Health at NICE , who will also carry out qualitative and quantitative evaluation based on the implementation of a novel search while screening pilot system. Evaluation will be carried out on reviews related with non-communicable diseases related with prevention of hazardous and harmful drinking and excessive alcohol consumption. Moreover, given the national and international importance of EBPH reviewing, the project has developed a multistrand pathways to impact document to engage with a variety of key EBPH stakeholders both in the UK and internationally.

This project will address current limitations in Evidence-based public health (EBPH) interventions by exploring new research methods, beyond the PICO framework, which combine text mining and machine learning to produce novel search while screening tools for public health. The PICO framework is conventionally used to structure pre-defined research questions matching clinical needs, helping to identify the Population, the Intervention, the Comparator and the Outcome. However, it is not well suited to the needs of EBPH reviews such as those conducted by the National Institute for Health and Care Excellence (NICE). PH questions are always complex involving behaviour, culture and organizations and often need to be described using abstract, fuzzy terminology in ways that make defining all the parameters in PICO a priori extremely problematic
We will investigate novel approaches to EBPH reviewing based on text mining-based unsupervised methods for the discovery of direct and indirect associations to support a dynamic and multi-dimensional relevance required for public health reviews. In particular, it will build on distributional semantics methods to improve term and document similarity measures by including contextual information in a novel way. Novel descriptive clustering algorithms will be developed that will use these measures to group documents, to analyse their topics to yield meaningful cluster labels and to simultaneously yield high quality document and label clusters. The project will also produce new ranking algorithms to order and visualise meaningful associations in an interactive manner, suitable for EBPH reviewing.
A pilot Web-based system providing a quick, interactive way of screening while searching, and visualisation, based on terms, their associations, descriptive clusters, labels and their ranking will form the basis for quantitative and qualitative evaluation for public health reviews related with alcohol misuse and consumption.

Economic and societal impact.
Text mining and machine learning based searching will contribute towards transforming evidence-based public health and influencing development of guidelines at a national and international level via NICE. Specifically, the new research will minimise the impact of publication bias in reviews and will extract more accurate and pertinent information from the literature, thus cutting both costs and time. Public health guidelines and best practice methods rely on EBPH: by producing more reliable and timely reviews, we will improve decision making in the health sector. The outcomes of the project will impact on the way policy makers, reviewers and researchers access information and discover knowledge. Policy impacts will be explored for national strategies in the area of obesity by Prof. Kelly, who is director of the Centre for Public Health Excellence at NICE, via public engagements.
The global economic impact of preventable ill health (WHO) will continue to increase at an alarming rate and we must work to improve awareness of diseases at different levels: societal, financial, clinical, psychological, etc. Thus, methods that provide cost-effective approaches to understanding interconnections between topics and better coverage of EBPH contribute towards mitigating the cost of public health.
Our proposed methods will also help improve UK competitive position in a digital market through better language technology products and services. Our project brings together a mixture of unsupervised techniques, which are re-usable and re-targettable in supporting and enabling language technology based access (via semantic search). Thus, our advanced search and screening techniques will be applicable in almost any other domain, such as energy, security, national libraries, institutional repositories, etc.
Other economic impacts will arise from exploitation of our software products in a range of health management and monitoring products.

Academic impact
Unsupervised ways of combining text mining and machine learning will impact on the digital economy in many domains. Semantic search via knowledge technology is a sine qua non to address the UK's competitive advantage across many sectors (biology, energy, security, etc.). The advanced methods developed here will be generalisable to literature reviewing in general and could thus lead to significant annual gains in academics' productivity - in excess of the £157m p.a. in additional productivity that basic text mining is estimated to yield for literature reviewing by UK academics, according to JISC.
We will provide training in our novel search while screening methods to evidence-based public health (EBPH) reviewers: for EBPH project partners and at summer schools and other events.
The proposed project also addresses important, longstanding but little understood problems associated with systematic review methods. Current information retrieval methods for systematic reviews are grounded on a narrow and restrictive theoretical basis. The result is that a single 'one-size-fits-all' method is the only available option for every type of information retrieval challenge in the increasingly complex portfolio of evidence synthesis types and methods. The proposed project has the potential to transform the theoretical basis of systematic review search methods, not only for public health problems, but for all types of systematic review and evidence synthesis.