Advanced machine learning guiding cardiovascular therapy decisions in type 2 diabetes patients.

Recent studies show that 70% percent of all deaths among type 2 diabetes (T2DM)
patients are due to cardiovascular disease (CVD) involving coronary heart disease,
stroke, heart failure and peripheral arterial disease. CVD treatment in subjects without
T2DM is guided by the Qrisk3 prognostic rule, however while multiple CVD prognostic
rules are available, it is unclear which rule is most appropriate for T2DM patients.
Furthermore, it is unclear whether it is possible to develop more advanced rules, jointly
predicting individual CVD elements (e.g., CHD and stroke), while allowing for competing
risk and disease trajectories.
As a first step in introducing risk-stratified CVD management in T2DM care I will validate
existing CVD prognostic algorithms on a British sample of 200,000+ T2DM CPRD
subjects. The scope of the project includes exploring average out-of-sample
performance, as well as presenting stratified results for geographical location, diagnosis
setting, disease duration, year of diagnosis and patient type.
The next phase of the project focuses on developing novel CVD prediction rules in
diabetes patients using advanced competing-risk and multi-stage methodologies.
Moreover, we want to leverage genomics data to improve prognostic performance. The
scope of the project also includes exploring the utility of EHR data and real-time analytics
to update CVD rules in diabetes population, designing methods for predicting repeated
episodes and predicting disease trajectory.