Having multiple long-term conditions is a growing problem in health and social care. Both research and actual recent events have shown that those suffering from this, have greater health and care needs and worse health outcomes. This is expensive for the health and social care system and often puts certain groups of people at a specific disadvantage.
This often affects certain groups, especially the most vulnerable (for example those of low socioeconomic status or advanced age), but we do lack accurate ways of predicting the patterns. So far, most research has focused on how to define the problem. We aim to look at how multiple long-term conditions develop and to see whether we can predict this happening. Specifically, we do not understand how individual health journeys evolve: how do individual diseases, medicines, health behaviours, mental health, geography, income, etc. contribute to these patterns?
One of the reasons we do not know this is that the amount of information is vast, and therefore difficult to organize, analyse and present. We now can program computers to learn as they go along, to more quickly and efficiently curate large amounts of data.
This project aims to analyse patients’ data using two different but complementary methods of computer learning. This study will look at a large set of general practice health care records (almost 40% of individuals with primary care records in the UK) using two different approaches to see which approach, or combination of approaches, most accurately predicts patterns of disease.
Although multimorbidity is associated with age, there is little understanding of common patterns. We aim to characterise health trajectories of multimorbidity using CPRD.
We target eighteen conditions/comorbidities: anxiety, asthma, atrial fibrillation, cancer, coronary heart disease, chronic kidney diseases, chronic obstructive pulmonary diseases, dementia, depression, diabetes, heart failure, hypertension, Parkinson’s disease, peripheral vascular diseases, schizophrenia, stroke, rheumatoid arthritis, and osteoporosis.
Study design: four cohorts of adults aged:18-44, 45-64, 65-84, and 85+ on 30th June 2005 and followed up to date. Minimum of 2-years follow-up from index date, ideally with long-term follow-up (~15 years). Population of interest (each cohort): confirmed cases of any target condition (described above).
First primary objective: to characterise the ageing trajectories of individuals based on: (EXPOSURES) characteristics (e.g. age, sex), biomarkers (e.g. cholesterol, creatinine), health behaviours/risk factors (e.g. smoking, alcohol use), target conditions, and socioeconomic factors (e.g. deprivation score). Outcomes targeted will be hospitalisation and all-cause mortality. Bayesian methods for identification of Latent Trajectories will integrate longitudinal data (EXPOSURES above) to estimate trajectories akin to biological age(BA) trajectories, which will support prevention of negative health outcomes.
Second primary objective: to identify clusters of multimorbidity for characterisation of new target groups. Using AI-Supported Clustering on longitudinal data (i.e. observed or modelled trajectories) based on the outcomes of hospitalisation and all-cause mortality will create clinically-relevant clusters. Two different inputs for these methods based on both a) original biomarkers and patient characteristics and b) latent trajectories obtained as estimates of BA will allow comparison of model outputs and will identify groups at risk potentially reducing health inequalities.
Analyses will be carried out by experts in each one of these fields (DS and TZ). They will be supported by expertise in electronic health records and large scale epidemiology (CB, SS, RP, DB, BC) as well as clinical experts (BN, CH, and KB).
For both first and second primary aims, the primary outcomes of interest will focus on hospitalisation and all-cause mortality. These are broad outcomes but clinically relevant and important across all target conditions studied.
For the first primary aim, the focus will be on derivation of models for prediction of individualised risk. Deriving and validating these models would provide a solid basis for early identification and, in the case of relevant interventions, early prevention. For the latent trajectories work, system (e.g. renal, respiratory, etc) specific (secondary) outcomes will be explored based on their importance (e.g. end-stage renal disease, COPD, etc).
For the second primary aim, the use of the outcomes will be for the creation of clinically-relevant clusters. These will allow characterisation of at risk groups and targeted public health interventions aiming to reduce health inequalities.
For health utilisation outcomes, we will estimate the relationship of C-MLTC clusters with life expectancy, DALYs, costs, and inequalities, and identify and prioritize cost-effective interventions for C-MLTC.
Rafael Perera - Chief Investigator - University of Oxford
Rafael Perera - Corresponding Applicant - University of Oxford
Anica Alvarez Nishio - Collaborator - University of Oxford
Benjamin Cairns - Collaborator - Our Future Health
Brian Nicholson - Collaborator - University of Oxford
Carl Heneghan - Collaborator - University of Oxford
Clare Bankhead - Collaborator - University of Oxford
David Steinsaltz - Collaborator - University of Oxford
Derrick Bennett - Collaborator - Nuffield Department of Population Health
Henrique Aguiar - Collaborator - University of Oxford
Kamaldeep Bhui - Collaborator - University of Oxford
Maria Christodoulou - Collaborator - University of Oxford
Subhashisa Swain - Collaborator - University of Oxford
Tingting Zhu - Collaborator - University of Oxford
HES Accident and Emergency;HES Admitted Patient Care;HES Outpatient;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Rural-Urban Classification