Tools that assess the risk of complications associated with type 2 diabetes mellitus estimate the probability that people with diabetes will experience a heart attack, stroke, heart failure, kidney disease, nerve disease, vision loss, or death. These tools have been used to studies comparing the value for money of different diabetes treatments, and can guide clinical decision-making. To date, the most commonly used tools to assess the risk of complications associated with type 2 diabetes mellitus were based on the UK Prospective Diabetes Study (UKPDS), conducted from 1977 to 1999 among 5,102 participants with newly diagnosed type 2 diabetes. Because of major changes in the epidemiology and treatment of type 2 diabetes over the last two decades, the tools based on the UKPDS have been found to be inaccurate for modern populations. Recently, we derived new Risk Equations tools to assess the risk of complications associated with Type 2 Diabetes (RECODe) from US population studies. Our purpose in this study is to extend our work in the US to derive similar risk equations tools for the UK CPRD population to enable more accurate estimation of the risk of diabetes complications among the UK adult population seen in modern clinical practice.
We propose a retrospective open cohort study of adults with Type 2 Diabetes in the CPRD. Adults with Type 2 Diabetes over the age of 18 will be studied over the period 1 January 2008 through 31 December 2017. The outcomes of interest: fatal coronary heart disease event or non-fatal acute myocardial infarction or coronary revascularization; fatal or non-fatal stroke (haemorrhagic or ischemic); first exacerbation of congestive heart failure; retinopathy leading to photocoagulation, or vitrectomy; nephropathy defined as microalbuminuria or stage 3-5 renal disease; neuropathy defined based on diagnostic codes or pressure sensation loss by monofilament test; lower-extremity non-traumatic ulcer; and death from any cause. Cox models will be fit to the time-to-event data for each outcome among people without history of the outcome at the beginning of the study period. Covariates in the models will include patient demographics, biomarkers, co-morbid disease history, and medications. An additional analysis will be performed by comparing equations developed through traditional Cox models to models alternatively derived using a novel machine learning approach called DeepSurv, which enables a neural network to be estimated to address potential complex interactions between covariates, both right- and left-censoring, and meaningful missing data patterns.
Diagnosis of acute myocardial infarction
Diagnosis of heart failure
Diagnosis of stroke
Diagnosis lower-extremity, non-traumatic amputation
Diagnosis of nephropathy
Diagnosis of neuropathy
Diagnosis of retinopathy
Death from any cause
Christopher Millett - Chief Investigator - Imperial College London
Eszter P Vamos - Corresponding Applicant - Imperial College London
Joseph Rigdon - Collaborator - Imperial College London
Raffaele Palladino - Collaborator - Imperial College London
Sanjay Basu - Collaborator - Stanford University
HES Admitted Patient Care;ONS Death Registration Data;Practice Level Index of Multiple Deprivation