Patient Subgroup Discovery

Principal Investigator: Dr Thomas House

This programme of work will address WP2 of UoM’s Turing application focusing on patient subgroups. The project is aligned to two Turing national priorities: to Revolutionise Healthcare; and to Supercharge Science (primarily Epidemiology).

Improving subgroup identification

This stream will focus on methods for looking for subgroups within the ‘Digital Phenotype’, in collaboration with Will Dixon of the Arthritis UK Centre for Epidemiology. The main technical challenge is the highly diverse volumes of data on each individual meaning that construction of feature vectors and definition of a distance between such vectors is not straightforward – progress has been made by this team on the Cloudy With a Chance of Pain data using distances between Dirichlet distributions with pseudocounts for the observed data.

Assessing subgroup structures

According to CDC: “The exposome can be defined as the measure of all the exposures of an individual in a lifetime and how those exposures relate to health.” Detailed knowledge of genomes allows the creation of dendrograms through hierarchical clustering; here consider how ‘exposome lineages’ can be constructed by creating algorithms suitable for use with detailed exposure data, building on previous work with colleagues at Warwick on Legionella using big socio-economic data. This will be useful for outbreak response, and other diseases with an environmental component, particularly COPD where we see possible synergies with other Turing@Manchester work.

Transferability

We will work with a framework where a stochastic process is solved for the vector for each cluster C, with parameters of this stochastic process expected to hold – or adapt more readily than those of unstructured models – across contexts. This should allow transferability, and we will see if inferences made for scabies in UK care homes can be extended to scabies in Ethiopia and Papua New Guinea in collaboration with the group of Jackie Cassell (UCL and Sussex). We will also collaborate on such mechanistic subgroup models with colleagues at Oxford, with a focus on conditions, like scabies, with significant burden in the developing world such as Ascaris.

Co-Investigators: Dr Ian Hall and Dr Lorenzo Pellis