Diabetes increases the risk of heart disease (heart attack and stroke). While most patients receive treatments to manage heart disease risk, long-term interventions are costly and not without risks themselves. Ideally, treatments should be tailored to those in most need, who will likely benefit most as well.
A number of heart disease prediction rules are available, aiding medical professionals in quantifying a patients' risk, and tailoring treatment regimes. It is currently unclear: which prediction rule performs best in the United Kingdom, if rules predicting any heart disease are sufficient or if heart disease specific rules are more relevant (e.g., separately predicting heart attack or stroke), and whether there is a need for subgroup-specific rules (e.g., for gender, age, duration of diabetes). Finally, it is unclear after how much time rules should be updated to correct for changes in patient- and treatment-characteristics, and whether the usual practice of predicting 10 years risk is sufficiently relevant for diabetes patients.
By harnessing electronic healthcare data, we will for the first time, systematically compare prediction rules within the same data, tackling the above stated knowledge gaps and deriving novel prediction rules accounting for difference between patients, and type of heart disease.
Objective:
To compare performance of existing cardiovascular disease (CVD) prediction rules in type 2 diabetes (T2DM) patients and derive novel rules accommodating differences between patient subgroups, and type of CVD.
Methods and data analysis:
For the first time we will use a single dataset to externally validate existing CVD prognostic rules in T2DM patients on: overall performance, discrimination, calibration and risk classification ( cut-offs <5%, 5%-20% and >20%). Performance will be stratified on: calendar time periods (per year and clinically relevant periods), duration of T2DM diagnosis (</=1,3,5, >/=8 years), history of coronary heart disease (CHD) or stroke, type of CVD, ethnicity, geographical location (as proxy for differences in case-mix).
Novel prediction rules will be derived by extending the Cox proportional hazard model to account for competing risks using the Fine and Grey methodology (Fine & Gray, 1999), allowing for joint prediction of individual CVD elements (e.g., stroke, MI, heart failure [HF]), as well as correcting for risk over-estimation due to all-cause mortality (Lau, Cole, & Gange, 2009). To further increase applicability, a novel nonparametric multivariate method (Andreas C. Damianou, 2013) will be used which can flexibly account for interactions between patient characteristics, and changes in prognostic ability over time (i.e., disease duration).
The primary outcomes are cardiovascular disease (CVD) and the individual CVD elements. All-cause mortality can be considered a secondary outcome as it plays an important role as a competing risk.
- Cardiovascular disease
- Stroke
- All-cause mortality
- Coronary heart disease
- Myocardial infarction
- Heart Failure
- Angina
Amand Schmidt - Chief Investigator - University College London ( UCL )
Amand Schmidt - Corresponding Applicant - University College London ( UCL )
Bob Wilffert - Collaborator - University of Groningen
Eelko Hak - Collaborator - University of Groningen
Folkert Asselbergs - Collaborator - University College London ( UCL )
Katarzyna Dziopa - Collaborator - University College London ( UCL )
Kenan Direk - Collaborator - University College London ( UCL )
Liam Smeeth - Collaborator - London School of Hygiene & Tropical Medicine ( LSHTM )
Mihaela van der Schaar - Collaborator - University of Oxford
Nishi Chaturvedi - Collaborator - University College London ( UCL )
Riyaz Patel - Collaborator - Barts Health and UCLH NHS Trusts
HES Admitted Patient Care;HES Outpatient;ONS Death Registration Data;Patient Level Townsend Score