Background
Due to the lack of valid Australian cardiovascular disease (CVD) risk assessment for type 2 diabetes (T2D) we previously developed the Fremantle Diabetes Study (FDS) equation for 5-year risk of hospitalisation for myocardial infarction (MI)/stroke or CVD death (3-point MACE, CVD1) using a Cox regression model. However, there have been substantial changes in the diagnosis, management and outcome of diabetes in the two decades since the original FDS cohort was enrolled. When the equation was applied to the contemporary FDS Phase II (FDS2) T2D cohort, calibration was poor, while improved survival after acute CVD events has seen hospitalisation for heart failure (HF) become a key additional outcome.
Aims
To develop and validate a contemporary CVD risk equation using conventional, routinely available variables, which considers adding hospitalisation for HF to CVD1 (CVD2) and assesses the use of Cox versus competing risk regression.
Method
FDS2 is a community-based longitudinal observational study. 1551 participants with T2D were followed from baseline (2008-11) for 5 years or until a first CVD event (hospitalisation for MI/stroke/HF or CVD death) or death from other causes. Incident events were identified from the Western Australian Hospital Morbidity Data Collection and Registry for Births, Deaths and Marriages. Missing covariates were multiply imputed (x20), defining models that included the outcomes of CVD and competing risk of death. Cox and competing risk regression modelling with backward elimination identified independent predictors of CVD1 and CVD2 within 5 years of study entry. For each of the four resulting models, discrimination was assessed by the area under the receiver-operating characteristic curve (AUC), calibration using the Hosmer and Lemeshow test (Ĉ-statistic) and accuracy by the Brier score. Positive and negative predictive values (PPV, NPV), sensitivity and specificity were determined for a 10% 5-year CVD risk cut-off. The best performing of the four risk equations was validated in 177 adults with T2D from the Busselton Diabetes Study. StataSE 15 was used for statistical analyses.
Results
The average age of the cohort was 66 years, 52% were men and the median diabetes duration was 9 years. During 7,011 and 6,896 person-years’ follow-up, respectively, 143 (9.2% (95% CI: 7.9%-10.8%)) participants had a CVD1 event and 245 (15.8% (14.0%-17.7%)) had a CVD2 event. Variables in the final risk models included age, sex, being Australian Aboriginal, heart rate, diabetes duration, ln(HbA1c), ln(total:HDL-cholesterol ratio), ln(urinary albumin:creatinine), eGFR, left ventricular hypertrophy, history of CVD, history of heart failure, and the presence of peripheral arterial disease. The performances of the four models are presented in the Table. The calibration of the CVD1 models was poor, dominated by overestimation of risk in the 10th decile with underestimation of risk in the lower deciles.  During 5-years’ follow-up, 20.9% (37 cases) of the validation cohort had an incident CVD2 event compared with a mean predicted risk of 12.5% (22 cases). There was good discrimination (AUC (95% CI): 0.81 (0.74-0.89), p<0.001), calibration (Ĉ-test p=0.16) and accuracy (Brier score: 0.15 (range: 0.00-0.92)). At a 10% risk cut-off the PPV was 43.1%, NPV 92.0%, sensitivity 75.7%, and specificity 73.6%.

Method	Outcome	AUC 
(95% CI)	Ĉ-statistic	PPV (%) 10% cut-off	NPV (%) 10% cut-off	Sensitivity (%) 10% cut-off	Specificity (%) 10% cut-off
Cox	CVD1	0.77 
(0.73-0.82)	<0.001	24.0	95.6	64.3	79.3
CRREG	CVD1	0.77
(0.73-0.81)	<0.001	24.3	95.5	62.9	80.0
Cox	CVD2	0.81
(0.78-0.84)	0.058	30.4	94.9	81.2	65.2
CRREG	CVD2	0.82
(0.79-0.85)	0.17	31.8	94.6	79.2	68.1

Discussion
In terms of performance, the preferred model was the competing risk one with HF included as an outcome, reflecting contemporary management and longer survival in Australians with T2D. Improvements in CVD risk prediction will require consideration of residual risk, risk variables not routinely measured, and novel risk factors.

A contemporary Australian cardiovascular risk equation for type 2 diabetes: The Fremantle Diabetes Study Phase II