Modern modelling of co-morbidities
Approved Research ID: 57382
Approval date: September 14th 2020
Morbidity trends are poorly understood and it is becoming increasingly important to understand, model and estimate the interdependency between factors/risks affecting individual health and longevity. Identifying these will result in improved projection models that estimate the range of mortality and morbidity outcomes for the UK population, or subsets of it, down to an individual level.
This will aid better decision making, e.g around treatments, costs/budgeting, resource allocation, forward-planning, targeted preventative measures across populations and individuals, and being able to better influence lifestyle decisions.
This research can lay the foundations to create a centre of excellence for improving understanding of health and care outcomes to help tackle social, health and demographic inequalities in the UK.
In particularly, this research will experiment with a range of 'big data' analytic techniques to try and fit the Biobank data on morbidity in a more informative way; driven by the fact that traditional survival analysis approaches do not allow for competing risks or the differences between individuals in a very sophisticated way. They make 2 simplifying assumptions that are in stark contrast to clinician views - see A4 for more detail.
Instead, our approach is based on accepting that you cannot identify a model from maximum likelihood approach, but you can seek the most likely model (from a range of models) given the observed data. Applying a Bayesian framework can help bring us closer to the solution we're looking for.
The (co-)morbidity outcomes we will ultimately assess will be determined by the Biobank data available. Until we see the data, examples of some of the morbidity outcomes are:
*cardiovascular disease, strokes, diabetes, Alzheimer's/dementia, cancers etc. I've listed the major diseases/illness affecting the population, however we may also investigate other morbidities that may exist in the data.
*links between treatments of health issues and their impact on other morbidities, for example successful cancer treatments and their impact on cardiovascular disease
The risk factors we will ultimately analyse will be determined by the Biobank data provided. However, examples of risk factors may be: general (age, gender, etc), medical factors/measures (general health, treatment type, medication, historic illnesses etc), socio-demographic (location/post code, employment history, etc), lifestyle (diet/fitness). Please note that we plan to analyse for latent classes where risk factors may not exist in the data, but the identification of these latent classes will support further investigation and analysis.