Using machine learning for the analysis of themes from the narrative on factors related to child deaths from child death review records

The National Child Mortality Database (NCMD) (https://www.ncmd.info/about/) national data collection and analysis system is the first of its kind anywhere in the world. It records comprehensive data, standardised across a whole country (England), on the circumstances of children’s deaths. The purpose of collating information nationally is to ensure that deaths are learned from, that learning is widely shared and that actions are taken, locally and nationally, to reduce the number of children who die. NCMD collates and analyses data on all children in England who die before their 18th birthday; with statutory death notification within 48 hours of death from the 58 Child Death Overview Panels (CDOPs) in England. Information is collected on statutory forms and includes the views and perspectives of families. There are over 1000 fields on the system, including categorical and free text answers. The final post-mortem report, and the results of any other investigations are also provided and available on the system as separate documents on Word or PDF format.

Aims

The aims of the NCMD programme are to:

• Capture, analyse and disseminate appropriate data and learning from child death reviews
• Drive the quality of child death review at every stage through bench-marking and quality improvement (QI) methodology
• Study and analyse the patterns, causes and associated risk factors of child mortality in England, providing information to target preventative health and social care and to assist in policy decisions

Supervisors

Professor Karen Luyt karen.luyt@bristol.ac.uk

Dr David Odd

There will be additional supervision with expertise in natural language processing.

Methods

The student will be given flexibility to explore and implement different methodologies to complete the project. However, the primary, proposed, methods would include Natural Language Processing (NLP) techniques for the analysis of narrative records to discern themes associated with child deaths. Initially, a comprehensive dataset comprising diverse narratives surrounding child fatalities will be curated. Subsequently, preprocessing steps such as tokenization, stemming, and lemmatization may be applied to enhance the efficiency of NLP algorithms. It is anticipated that Named Entity Recognition (NER) and topic modelling algorithms, such as Latent Dirichlet Allocation (LDA), will play a role in identifying themes like medical conditions, caregiver information, and socio-economic indicators, within the narratives. Machine learning models will be trained to categorise the identified themes and establish connections between these themes and potential risk factors, and modifiable factors, in child deaths.

How do I apply?

Candidates who wish to apply for this 3 year PhD should first complete the online application choosing `Translational Health PhD`. Please indicate in the funding box whether you have secured your own sponsorship or you are self-funding.

Informal initial enquiries are welcomed: karen.luyt@bristol.ac.uk