Prediction of Mortality Using Measures of Cardiac Autonomic Dysfunction in the Diabetic and Nondiabetic Population: The MONICA/KORA Augsburg Cohort Study

The MONICA survey S2 was part of the multinational World Health Organization MONICA project (10). In 1989–1990, a random sample aged 25–74 years (n = 4,940) was selected from an original population of n = 349,050 in the Augsburg region in southern Germany. All study participants aged 55–74 years were included in the present analysis, of whom n = 160 were classified as having diabetes if they reported a diagnosis of diabetes or if they were taking antidiabetes medication, while n = 1,560 subjects were considered nondiabetic if they did not meet these criteria. All participants were prospectively followed within the framework of KORA. After a follow-up period of 9 years, all-cause mortality and cardiovascular mortality were assessed. The survey was approved by the local authorities, and all participants gave written informed consent.

Blood pressure, body height, and body weight, were determined by trained medical staff (mainly nurses). All measurement procedures have been described elsewhere in detail (11–13). Information concerning sociodemographic variables, smoking habits, physical activity, and alcohol intake was assessed by standardized personal interviews. A regular smoker was defined as a subject who regularly smoked at least one cigarette per day. Alcohol consumption on the previous workday and during the previous weekend was calculated in grams per day. High alcohol intake was defined as ≥40 g/day in men and ≥20 g/day in women. A participant was considered physically active if he or she participated in sports in summer or winter for >1 h/week. Prevalent cardiovascular disease (CVD) was defined as the need for hospital treatment of myocardial infarction or stroke (13). Obesity was defined as BMI ≥30 kg/m². Hypertension was defined as a blood pressure of 160/95 mmHg or higher or use of antihypertensive medication, given that the subject was aware of having hypertension. Dyslipidemia was defined as a total–to–HDL cholesterol ratio ≥5 (14).

ECG examination was performed in a standardized manner as described previously (14,15). In brief, a 12-lead resting ECG was recorded over 20 consecutive seconds in the supine position using the digital stand-alone ECG data acquisition and analysis system SICARD 803 (Siemens Medizintechnik, Erlangen, Germany). Time domain measures including the SD of R-R intervals (SDNN), coefficient of variation (CV) of R-R intervals, and the difference between the maximum and minimum R-R interval (max-min difference) were computed (16). This approach was limited by not controlling for respiration and not using an index of HRV during deep breathing. QT intervals were determined from the 12-lead ECG strips. From each lead, three QT intervals were measured and, of these, the median values were computed. The longest of the 12 QT medians obtained was used as the representative QT interval for further analysis. Measurable QT intervals in eight leads were required as an acceptable minimum for this definition. QTc interval correction formulas for heart rate included the approaches by Bazett (17), the Framingham Heart Study (18), and Fridericia (19). QTD was measured as the difference between the longest and shortest QT intervals in 12-lead ECG. The cut points were the lowest quartiles for SDNN, CV, and max-min difference, while those for the QT indexes were QTc >440 ms and QTD >60 ms.

For the calculation of the autonomic function indexes, 241/26 nondiabetic/diabetic subjects had to be excluded due to atrial fibrillation or flutter (n = 37), left (n = 38) and right (n = 63) bundle-branch block, second- and third-degree atrioventricular or sinoatrial block (n = 12), treatment with antiarrhythmic agents, or treatment with agents known to prolong the QT interval (n = 48) (multiple nominations were possible). An additional 47/8 nondiabetic/diabetic subjects were excluded in the HRV assessment, respectively, and 64/9 nondiabetic/diabetic subjects were excluded in the QTc interval analyses because of multiple supraventricular or ventricular extrasystoles, pacemaker therapy, or missing data. Thus, 1,513 nondiabetic and 152 diabetic subjects were included in the multiple logistic regression analyses of HRV, while 1,496 nondiabetic and 151 diabetic subjects were included in the QTc interval model. The QTD model comprised 1,433 nondiabetic and 140 diabetic subjects as a resulut of incomplete 12 leads.

All continuous variables were described as means ± SD, and differences between groups were evaluated by t tests. Categorical variables were described by frequency tables and compared between groups using Fisher's exact test. All tests were performed two sided, and the level of significance was set at α = 0.05. Survival curves were estimated by the Kaplan-Meier method. The log-rank test was used to compare different survival curves. Multiple Cox regression models were fitted to analyze risk factors of mortality and potential confounders. Different models using fixed sets of independent variables were estimated and stratified for diabetic and nondiabetic subjects. The SAS statistical software package (version 8.2) TS2M0 was used for statistical analyses.

Compared with nondiabetic individuals, diabetic subjects were significantly older, had a higher BMI, faster resting heart rate, higher systolic blood pressure, lower HDL cholesterol, higher total–to–HDL cholesterol ratio, and higher fibrinogen levels as well as significantly higher proportions of hypertension and myocardial infarction or stroke (all P < 0.05) (Table 1). Furthermore, rates of death from all causes and coronary artery disease were significantly higher, while the percentages of physically active subjects and those with high alcohol intake were significantly lower in the diabetic compared with the nondiabetic group (all P < 0.05). The use of β-blocking agents tended to be higher in diabetic subjects (P = 0.072). No significant differences between the groups were noted for sex, diastolic blood pressure, total cholesterol, LDL cholesterol, or the percentage of regular smokers.

In the diabetic group, survival probability was significantly lower in subjects with the max-min R-R interval difference “as measured from a 20 s standard 12-lead ECG without controlling for depth and rate of respiration” (see research design and methods) at the 1st quartile vs. 2nd–4th quartiles (P = 0.0447)(Fig. 1A), whereas no significant difference was noted in the nondiabetic group (P = 0.4227)(Fig. 1C). The corresponding mortality rates were 30 of 79 (38.0%) vs. 19 of 80 (23.8%) in diabetic subjects and 42 of 366 (11.5%) vs. 119 of 1,179 (10.1%) in nondiabetic participants, respectively. In subjects who had a QTc interval >440 vs. those with a QTc interval ≤440 ms, survival probability was significantly lower in both the diabetic (P < 0.0001) (Fig. 1B) and nondiabetic (P = 0.0026) groups (Fig. 1D).

In the nondiabetic group, male sex, age, regular smoking, low physical activity, and hypertension (all P < 0.05), but not low max-min difference, were significant predictors of mortality, while CVD tended to predict mortality (P = 0.06) (Table 2). Among diabetic individuals, only low physical activity and dyslipidemia were significant predictors of mortality (both P < 0.05), while low max-min difference, male sex, CVD, and regular smoking tended to predict mortality (P = 0.067 to P = 0.089). The increase in relative risk of mortality for the 1st quartile of max-min was 73% in diabetic persons. The results were similar when using the SDNN and CV as alternative indexes of HRV (data not shown).

In the nondiabetic group, prolonged QTc interval, male sex, age, regular smoking, low physical activity, and hypertension (all P < 0.05) were significant predictors of mortality, while CVD tended to predict mortality (P = 0.076) (Table 2). Among diabetic individuals, prolonged QTc interval, male sex, age, and low physical activity were significant predictors of mortality (all P < 0.05), while a trend to predict mortality was noted for regular smoking and dislipidemia (P = 0.087, to P = 0.081). The increase in relative risk of 9-year mortality in subjects with QTc >440 ms was twofold and threefold in the nondiabetic and diabetic group, respectively. The results were similar when using the Framingham Heart Study and Fridericia formulas for QTc (data not shown).

In the nondiabetic group, male sex, age, high alcohol intake, regular smoking, low physical activity, and hypertension (all P < 0.05), but not increased QTD, were significant predictors of mortality (Table 2 C). In the diabetic group, age and low physical activity (both P < 0.05), but not increased QTD, were significant predictors of mortality.

After introducing both the max-min difference and QTc interval and their possible interaction into the model, QTc interval prolongation remained a significant predictor of all-cause mortality showing risk ratio 2.23 (95% CI 1.32–3.75), P = 0.003, in the nondiabetic group and 7.09 (1.80–27.94), P = 0.005, in the diabetic group, respectively. In contrast, low max-min difference did not predict mortality in the nondiabetic group (0.96 [0.64–1.42]; P = 0.829) but tended to in the diabetic group (1.96 [0.99–3.89]; P = 0.054). There was no interaction between these two variables in predicting mortality (data not shown).

Prolonged QT interval predicted cardiovascular mortality in the entire cohort and in the nondiabetic group (3.91 [95% CI 2.14–7.14]) P < 0.001, and 4.47 [2.44–9.22], P < 0.001, respectively) but not in the diabetic group (1.39 [0.26–7.43], P = 0.698). The max-min difference and QTD did not predict cardiovascular mortality in either of the groups studied, possibly because of the relatively small number of cases (data not shown).

The results of this study suggest that prolonged QTc interval is an independent predictor of mortality over 9 years in the nondiabetic and diabetic elderly general population, respectively. Diabetic patients with a QTc prolongation >440 ms had a threefold increased risk of mortality. In contrast, increased QT dispersion did not predict mortality in nondiabetic or diabetic subjects. Low HRV showed a trend toward an increased risk of mortality by 73% in the diabetic but not the nondiabetic elderly general population. Thus, while QTc interval prolongation represents a general prognostic index independent of the presence of diabetes, reduced HRV appears to be a more specific marker only in the context of diabetes.

Several studies have previously reported that QTc interval prolongation predicts the risk of mortality in the elderly (20) and middle-aged (21) general population, although in apparently healthy subjects this risk was weak (21). Likewise, an increased QTD has been identified as a predictor of increased cardiac mortality at the population level (22,23). However, recent studies evaluating the predictive role of prolonged QTc or increased QTD in diabetic patients have reported conflicting results. In a 23-year follow-up of the World Health Organization Multinational Study of Vascular Disease in Diabetes, QTc was associated with long-term mortality in subjects with type 1 diabetes but not in those with type 2 diabetes (24). In contrast, in the Strong Heart Study, including American Indians with type 2 diabetes, QTc predicted all-cause mortality after a mean follow-up of 4.7 years (25), and in the Dundee cohort of the UK Prospective Diabetes Study, including newly diagnosed type 2 diabetic patients, both QTD and QTc predicted cardiac mortality after a mean follow-up of 12.7 years (26). Moreover, QTD was identified as an independent predictor for total cardiovascular events and for cardiac deaths among type 2 diabetic patients with arterial hypertension (27). However, another study found that QTc but not QTD was an independent predictor of all-cause and cardiovascular mortality in type 2 diabetic patients (28). However, these were clinic-based studies and, hence, not representative of any certain population. We confirm at the population level that prolonged QTc but not increased QT dispersion is a predictor of all-cause mortality in both elderly nondiabetic and diabetic subjects. However, there is ongoing controversial discussion about the value of measuring QT dispersion (29,30). Among the reasons for the divergent study results are difficulties in determining the end of the T-wave, the absence of standards for this method, circadian rhythm, and the lack of normative data.

Whether reduced HRV is a predictor of mortality has previously been addressed at the population level in two studies (31,32). We suggest that low HRV is not an independent predictor of mortality in the nondiabetic population. This is in line with the results of the Hoorn study (31) and the Atherosclerosis Risk in Communities (ARIC) study (32). On the other hand, in the Zutphen study (33) the 5-year age-adjusted relative rate of total mortality was 2.1 (95% CI 1.4–3.0) in middle-aged men and 1.4 (0.9–2.2) in elderly men with low HRV. This finding does not necessarily contradict ours, since that study included only a male population. We found that male sex is an independent predictor of mortality in the nondiabetic population, and the prognostic value of low HRV may therefore differ between male and female or mixed cohorts. However, as shown above, the risk associated with low HRV was considerably lower and no longer significant in elderly men.

The Hoorn study (31) followed a population aged 50–75 years over 9 years similar to our cohort aged 55–74 years. The results of our study are compatible with those of the Hoorn study, demonstrating that diminished HRV is a predictor of mortality in the diabetic as opposed to the nondiabetic population. In contrast, in the ARIC study (32), low HRV did not predict fatal coronary artery disease or non–coronary artery disease mortality in diabetic subjects after an average of 8 years of follow-up. However, in the ARIC cohort the age range of 45–64 years was considerably lower than in our cohort. The findings of the Hoorn study (31) and ours indicate that after adjustment for the various confounding factors, the value of low HRV in predicting excess mortality is only moderate. In line with the borderline independent effect of low max-min difference on mortality observed in our study, in the Hoorn study the increased risk of mortality in the diabetic subjects was noted in only two of six indexes of HRV measured (31). This risk level may be underestimated because this and other epidemiologic surveys have performed less extensive assessment compared with the majority of the clinic-based studies. Indeed, it has been suggested in a meta-analysis that the pooled relative risk of mortality in clinic-based studies that used more than one index was considerably higher than that observed in studies that used one measure only (3).

One limitation of this study is the relatively short period (20 s) of ECG recording without control for respiration. However, since we have not employed frequency domain indexes of HRV, we believe that computing time domain indexes from these recordings is relatively accurate. Relatively short recordings of ECG leads have been reliably used in several other epidemiological studies, e.g., three or more consecutive cycles (34) or recordings over 10 s in the Rotterdam Study (6) and the Diabetes Prevention Program (35) or 15–30 s in the Zutphen-Study (33). Nonetheless, adding indexes of vagal function obtained during controlled deep breathing would have likely strengthened the HRV results.

In conclusion, after adjustment for various well-known prognostic factors such as male sex, age, smoking, physical activity, hypertension, obesity, dyslipidemia, and CVD prolonged QTc interval and, to a lesser degree, without controlling for respiration as a result of the restricted setting of an epidemiological study, diminished HRV represent independent predictors of 9-year mortality in the diabetic elderly general population. In contrast, increased QT dispersion did not predict mortality in nondiabetic or diabetic individuals. Low HRV showed a trend of borderline significance toward an increased risk of mortality by 73% in the diabetic population but was not a predictor of mortality in the nondiabetic elderly general population. Thus, cardiac autonomic dysfunction characterized by prolonged QTc and/or diminished HRV is associated with a high risk of excess mortality particularly in diabetic subjects. Recent studies indicate that some cardioprotective agents may increase diminished HRV (36) and shorten prolonged QT interval (37) and, hence, have the potential to improve or worsen prognosis in selected patient populations. Against this background, the findings of the present study may have consequences for treatment of diabetic patients with CAN provided that the simple measures of HRV and QTc interval described herein are being used for risk stratification in clinical routine. Measurement of the QTc interval may be favored in this context given that it is simple to do and may represent a stronger prognostic marker than HRV.

The KORA and the MONICA Augsburg studies were initiated and financed by the Helmholtz Zentrum München - German Research Center for Environmental Health, which is funded by the German Federal Ministry of Education, Science, Research and Technology and by the state of Bavaria. Mortality follow-ups in 1997–1998 and 2002–2003 were also supported by grants from the Federal Ministry of Education, Science, Research and Technology (01 ER 9701/4) and the German Research Foundation (DFG) (TH 784/2-1), respectively.

We thank all members of the GSF Institute of Epidemiology and the field staff in Augsburg involved in the planning and conduct of the study.