Bio
Lilia is an Ear, Nost and Throat (ENT) surgeon in-training in London. In addition to her medical degree, she holds an undergraduate degree in Experimental Psychology, Philosophy, and Physiology from the University of Oxford and a MSc in Health Data Science from UCL. Lilia has combined academic and clinical training since qualifying as a medical doctor with her research routed in data sciences. She was awarded the MRC Clinical Research Training Fellowship to undertake her PhD at UCL. Her research focus is on using machine learning methods to better understand hearing loss subtypes. Outside of her primary roles, she actively pursues her interest in education and community engagement by serving as a school governor at the Boxing Academy in Hackney..
Research interests
Her project builds on ongoing work she has been involved in at University College London Hospital (UCLH) developing a hearing health database using Electronic Health Records (EHR). She is using unsupervised machine learning to identify hearing loss subtypes based on hearing test results for the most common type of hearing loss – sensorineural hearing loss. This is of clinical importance because more precise phenotyping is needed to inform therapeutic and preventative strategies as well as for prognostication. Currently, the only way to get information about subtypes is post-mortem. There have been human-crafted classification systems created to identify subtypes from hearing tests but recent developments in our understanding of the underlying pathology and genetics pathways of hearing loss challenge these systems. She is interested in exploring the association between different subtypes identified by the model and other features including social deprivation, ethnicity, and presence of co-morbidity (in particular dementia and diabetes). These fields are not reliably completed in the EHR. She will be working with the Structured Missingness in heterogenous data research theme led by Dr Robin Mitra to learn new methods to address the issue of data missingness as well as gain a deeper understanding of what it can tell us about the data and context the data is generated in.