The influence of baseline risk on the relation between HbA1c and risk for new cardiovascular events and mortality in patients with type 2 diabetes and symptomatic cardiovascular disease

For this study we used data from patients with type 2 diabetes and cardiovascular disease enrolled in the Second Manifestations of ARTerial Disease (SMART) cohort. The design and rationale of the SMART study have been described previously [23]. To shortly summarise, patients aged 18–79 who are referred to the University Medical Center Utrecht with atherosclerotic vascular disease or for treatment of cardiovascular risk factors are included in the database. Physical and laboratory examinations are performed in the hospital after an at least 8-h fast. Information about a history of CVD, vascular risk factors and a detailed medical history are obtained via a questionnaire. Follow-up information is obtained via questionnaires sent to patients every 6 months or via the general practitioner (GP) if the patient is unwilling to fill out questionnaires and gives consent for GP consulting. The ethics committee of the University Medical Center Utrecht approved this study, all eligible patients received written and oral information about the study and all included participants gave informed consent [23].

For the current study, patients with type 2 diabetes and clinically manifest vascular disease at baseline were selected (n = 1205). Patients with a missing high-sensitivity C-reactive protein (hsCRP) value (n = 22) or a hsCRP value above 15 mg/L (n = 83) were excluded. The latter were excluded because we think such a high hsCRP value is more indicative of infection or a different cause of inflammation than the low-grade inflammation seen in patients with atherosclerosis [24]. Finally, we recalculated missing data regarding LDL using the Friedewald formula. Data that remained missing after this procedure (n = 6) was imputed. The participants for whom the SMART risk score could not be calculated (n = 4) were also excluded, bringing the final study population to 1096 participants.

Diabetes mellitus was defined as a referral diagnosis of type 2 diabetes, self-reported type 2 diabetes, a fasting serum glucose ≥7.0 mmol/l at inclusion with initiation of glucose lowering treatment within 1 year or the use of oral anti-hyperglycemic agents or insulin at baseline. Participants with known type 1 diabetes were excluded from our analyses. History of clinically manifest vascular disease is defined as ever having had clinical manifest vascular disease [23]. HbA1c was studied as a continuous variable.

The cardiovascular outcomes defined are myocardial infarction (MI), ischemic stroke, the composite of major cardiovascular events, cardiovascular mortality, and all-cause mortality. Their definitions have been described elsewhere [23]. Each notification of a possible event is thoroughly checked by collecting all available information about the patient regarding the event. This includes all correspondence and documented investigations concerned with the particular event. All events were audited by three independent committee members of the Outcome Committee.

Data are presented as mean ± standard deviation (SD) or median with interquartile range for variables with a skewed distribution. For all 1096 participants, the SMART risk score was calculated using the equation described by Dorresteijn et al. [22]. This equation predicts the 10-year risk of recurrent major vascular events (myocardial infarction, stroke and vascular death) in patients with CVD [22]. All analyses were performed for the whole cohort and in quartiles (Q1-Q4) of SMART risk. Non-HDL cholesterol was calculated for all patients by subtracting the measured HDL cholesterol from the measured total cholesterol value. To assess whether baseline risk as defined by the SMART risk score has a modifying effect on the relation between HbA1c and new outcomes, the interaction term between HbA1c and SMART risk score was used. We considered the effect modification by the SMART risk score significant when the p value of the interaction term was <0.05. Cox regression modelling was used to assess the relation between HbA1c and cardiovascular outcomes and mortality. Cubic spline analysis was rejected based on the inability of this technique to measure effect modification. Results are given as hazard ratios with 95 % confidence interval, and denote the increase in risk for a cardiovascular outcome or mortality related to a 1 % increase in HbA1c. Three models were constructed. The first model was a univariate model which only included HbA1c. The second model was adjusted for sex and age, and the third model was additionally adjusted for the known confounders current smoking, non-HDL cholesterol, diabetes duration, systolic blood pressure and eGFR (MDRD). We performed sensitivity analyses including the earlier excluded patients with a hsCRP value above 15 mg/L (n = 83) and additional analyses adjusted for all components of the SMART risk score. All analyses were performed in an imputed dataset, as missing data is seldom missing at random. We imputed missing data on eGRF(1, 0.09 %), LDL cholesterol (6, 0.5 %) and HbA1c (100, 9.1 %). All analyses were performed using SPSS 21 (IBM, New York). Graphs were created using R 3.1.1, (The R Foundation for Statistical Computing, Austria).

The mean age of the participants in the cohort was 62.6 ± 8.8 years and 76 % of the participants were male. The SMART risk score ranged from a predicted 5–100 % 10-year risk for recurrent vascular events. The creation of quartiles based on the SMART risk score resulted in four groups with a SMART risk score range of 5–16, 16–24, 24–37 and 37–100 % respectively (Table 1).

Mean age increased from 55.8 ± 7.1 years in the lowest risk quartile to 70.0 ± 6.5 years in the highest risk quartile, and the percentage of males increased from 72 to 76 %. In addition, the mean duration of diabetes increased from 4.0 to 6.0 years and time since first vascular event increased from 0 to 9 years.

During a mean follow-up period of 6.9 years 243 patients’ deaths were reported. The mean follow-up period for vascular events was 6.4 years, in which period 223 cases were reported. The separate outcomes had follow-up periods of 5.8 years for MI and 6.7 years for stroke, in which period 84 and 48 events were reported, respectively.

In the whole cohort of patients with type 2 diabetes and vascular disease, a 1 % increase in HbA1c level was associated with a higher risk for all-cause mortality (HR 1.18, 95 % CI 1.06–1.31). There was no relation between HbA1c and risk for new cardiovascular events during follow-up (HR 1.03, 95 % CI 0.91–1.16) (Table 2).

In patients with the highest predicted 10 years risk for new cardiovascular disease or mortality (37–100 %), a 1 % increase in HbA1c was associated with an increased risk for cardiovascular mortality (HR 1.29, 95 % CI 1.03–1.61) and all-cause mortality (HR 1.33, 95 % CI 1.11–1.60) (Table 2). The results for the lowest SMART risk quartile were similar, albeit not significant (HR 1.36, 95 % CI 0.93–1.98). An inverse relation was suggested between HbA1c and risk for MI in all quartiles (Fig. 1). This was most apparent in the second risk quartile (HR 0.71, 95 % CI 0.46–1.10).

However, the interaction term between HbA1c and the SMART risk score was not significant for the composite cardiovascular outcomes and for cardiovascular or all-cause mortality (p = 0.225, p = 0.259 and p = 0.179 respectively). Thus, baseline risk as estimated by the SMART risk score did not modify the relation between HbA1c and cardiovascular events or mortality in patients with type 2 diabetes and cardiovascular disease.

None of our sensitivity analyses changed the direction or magnitude of the modifying effect of the SMART risk score on the relation between HbA1c and vascular outcomes and mortality.