Mortality and Predictors of Mortality in a Cohort of Brazilian Type 2 Diabetic Patients

Characteristics of the study patients have been detailed elsewhere (12,14). Briefly, all adult diabetic outpatients (diagnosed according to 1985 World Health Organization [WHO] criteria) who had standard electrocardiograms (ECGs) recorded from July 1994 to June 1996 were consecutively enrolled in the study. After applying clinical and electrocardiographic exclusion criteria, as previously described (12,14), the total cohort numbered 471 type 2 diabetic patients. The study protocol complied with the 1975 Declaration of Helsinki and was approved by the local ethics committee.

The baseline procedures and criteria for diagnosing clinical variables have been previously described (14). All subjects were given a thorough clinical examination, with special attention to signs and symptoms of cardiovascular diseases and diabetic degenerative complications. Laboratory evaluation included fasting glycemia, fructosamine, serum creatinine, triglycerides, total and HDL cholesterol, and 24-h proteinuria. Mean values of all office systolic blood pressure (SBP) and diastolic blood pressure (DBP) measurements and laboratory examinations performed in the first year of follow-up were recorded. Arterial hypertension was diagnosed for mean SBP ≥140 mmHg, DBP ≥90 mmHg, or if antihypertensive drugs had been prescribed.

From standard resting 12-lead ECGs, abnormalities were registered according to the Minnesota code, except left ventricular hypertrophy, which was verified by voltage criteria (either Sokolow-Lyon or Cornell sex specific). Premature ventricular contractions (PVCs) were considered frequent if present in >10% of all cycles recorded. The ECG intervals were measured as previously reported (14) by a single independent observer. The maximum heart rate−corrected QT interval duration (QTc_max) and QT interval dispersion (QTd; difference between maximum and minimum QT intervals) were recorded. To assess reproducibility, 45 randomly chosen ECGs were analyzed twice, with at least 6 months between the measurements. Intraobserver mean relative errors were 1.1% for QTc_max and 12% for QTd.

The patients were evaluated regularly at least two times a year until June 2001. Those who failed to present at the hospital were contacted annually to determine vital status. Causes of death during the follow-up period were ascertained from medical records, death certificates, and interviews with attending physicians and families, using a standard questionnaire reviewed by an independent observer. Causes of death were coded according to the International Classification of Diseases. We defined diabetes-related mortality as that caused by infection, renal failure, or cardiovascular causes. Cardiovascular mortality was defined as death from any cardiac, cerebral, aortic, or peripheral vascular disease. The observation period for each patient was the number of months from the date of the measured ECG to the date of death or 30 June 2001. Overall, 43 (9.1%) patients were lost from follow-up and were considered as censored observations at the date of their last hospital visit.

Statistics were performed using the STATA version 7.0 software. Continuous data were described as medians and 5–95% percentiles. The comparison of mortality between this cohort and the population of Rio de Janeiro (using the midpoint 1996 population) was made by calculating the standardized mortality ratios (SMRs). The 95% confidence intervals (CIs) for age- and sex-adjusted SMRs were calculated under Poisson assumption. The Kaplan-Meier estimation of survival curves (compared by log-rank tests) and uni-and multivariate proportional hazards Cox models were used for survival analysis. Serum creatinine was log₁₀ transformed because of its positive skewed distribution. Data for 24-h proteinuria (25%) and HDL cholesterol (32%) were frequently missing. Deleting subjects with a missing value on one predictor variable included in multivariate models commonly leads to biased results and surely to loss of power (15). Therefore, to decrease bias and increase statistical efficiency, we imputed missing data using the expectation maximization method. Variables with a P < 0.10 in Cox univariate analysis entered the multivariate models. Sex was forced into the multivariate analysis. To remain in the multivariate models, a value of P < 0.10 was necessary. Different multivariate models were fitted in a forward stepwise strategy for all-cause and diabetes-related mortality. Assumptions of the proportional hazards models and interactions were tested (16), and no violation or significant interaction was observed. Results are presented as the hazard ratio with 95% CI. A two-tailed P < 0.05 was considered statistically significant.

After a median follow-up of 57 months (range 2–84 months), 121 (25.7%) patients died, 91 (75.2%) from diabetes-related causes: 44 (36.4%) from cardiovascular diseases, 40 (33.1%) from infections, and 7 (5.8%) from renal failure. The most frequent non-diabetes−related cause of death was cancer (13 patients, 10.7%). Table 1 shows the baseline characteristics of survivors and deceased patients. Nonsurvivors were older and had a longer duration of diabetes, a higher prevalence of cardiovascular diseases and diabetes complications, and more electrocardiographic and laboratory abnormalities than did survivors. Patients lost from follow-up had baseline characteristics identical to those who completed follow-up.

Table 1 also shows the results of univariate Cox analysis for diabetes-related mortality. Figures 1 and 2 show Kaplan-Meier survival curves for patients grouped according to the presence or absence of six prognostically important variables. Table 2 shows the predictive multivariate Cox regression models for all-cause and diabetes-related mortalities. The independent mortality predictors were older age, increased 24-h proteinuria, preexisting peripheral vascular disease (and cerebrovascular disease for diabetes-related deaths), frequent PVCs and QTc_max prolongation >470 ms^1/2 on baseline ECGs, and a lower serum HDL cholesterol. The use of β-blockers was a protective factor for mortality.

Table 3 shows all-cause and cardiovascular SMRs for patients stratified according to sex and age ranges. Type 2 diabetic patients had a more than threefold excess mortality adjusted for age and sex than the background population of Rio de Janeiro. Although the increased mortality persisted in both sexes until age 79 years, it was most prominent in age ranges <70 years, largely because of an increased cardiovascular mortality risk.

This prospective study with up to 7 years of follow-up had two main findings. First, Brazilian type 2 diabetic patients had a more than threefold excess mortality compared with the general population. Second, some clinical/demographic (older age, preexistent vascular disease), laboratory (increased 24-h proteinuria, decreased HDL cholesterol), and electrocardiographic variables (presence of frequent PVCs, prolonged QTc_max) were independent predictors of this increased mortality.

This is, to the best of our knowledge, the second study on the impact of type 2 diabetes on mortality in Brazil and the first to report its predictive factors. We found sex- and age-adjusted SMRs to be higher than those in investigations from most other countries (1,4–6), but similar to some (7,17). A report from the WHO Multinational Study of Vascular Disease in Diabetes (2) has shown that overall SMRs vary widely among different cities and countries, from as low as 1.38 in men and 1.26 in women in Tokyo, Japan, to as high as 3.70 in men and 4.35 in women in Havana, Cuba. However, SMRs from distinct centers should be interpreted cautiously, as the figures might not be directly quantitatively comparable because of different background mortality rates (2). Of note was the finding that in our cohort, the excess mortality was most important in age ranges <69 years and was largely explained by an increased cardiovascular mortality. The excess all-cause mortality persisted until the 8th decade of life. These findings are supported by those from other studies (4,7,17). Also, we did not demonstrate any difference in diabetes-related mortality risk between sexes, confirming the assumption that the presence of diabetes equalizes mortality risks between men and women (8).

Some of the predictors of mortality observed in this analysis are well-established risk factors in patients with type 2 diabetes, such as older age (5,6,18,19), preexisting vascular disease (1,6,19–21), increased proteinuria (1,5,6,19,22), and decreased HDL cholesterol (21–23). Two electrocardiographic variables, the presence of frequent PVCs and maximum QTc interval prolongation, were shown to add prognostic information for overall survival beyond these traditional risk markers. Frequent PVCs on ECGs have been demonstrated in population-based studies to constitute predictors of cardiovascular mortality (24), as was also seen in this cohort (12). This finding may reflect myocardial electrical instability or irritability secondary to underlying silent coronary heart disease, possibly related to the increased risk of sudden arrhythmic death associated with diabetes and impaired glucose tolerance (25). QTc interval prolongation in diabetes has been associated with several unfavorable conditions, such as cardiac dysautonomia (26), underlying coronary heart disease (27), increased SBP and left ventricular mass (27), and insulin resistance (28). Therefore, it is not unexpected that QTc interval prolongation, probably reflecting the conjunction of these disadvantageous conditions, constituted a risk marker for all-cause and diabetes-related mortality, as we have previously found for stroke (13) and cardiovascular events (12). The prognostic value of QTc interval prolongation for overall mortality risk stratification is supported by two other reports (20,22). Unlike one of these studies (22), we were unable to demonstrate any predictive value of QTd for all-cause mortality.

The use of β-blocker drugs as a protective factor for mortality was rather unexpected. β-Blockers have been showed to decrease mortality in patients with heart failure and coronary heart disease, especially after myocardial infarction. Hence, β-blockers are probably the antihypertensive drug of choice in diabetic patients with established cardiovascular disease (29). In the present cohort, only 36 patients (7.6%) were using β-blockers and only 2 of them died. Because of this small number of patients, the protective effect of this class of drug on mortality in diabetic patients should be interpreted cautiously and further assessed in future studies.

Two generally accepted mortality predictors in patients with diabetes, SBP levels (3,5,18,21) and glycemic control (1,3,5,19,21), were not selected in this study. SBP was a predictor of diabetes-related mortality in univariate Cox analysis, but was not chosen as an independent predictor in multivariate analysis. Possibly the prognostic information given by blood pressure levels was incorporated by the QTc_max, given the strong relation between QTc duration and SBP that has been demonstrated (27). An alternative explanation is the influence of antihypertensive treatment on blood pressure levels, as nearly all the hypertensive patients were on drug treatment and blood pressure levels were mean values obtained during the first year of follow-up. With regard to glycemic control, neither fasting glycemia nor serum fructosamine showed any prognostic value for mortality. Because HbA_1c levels were not available in this study, it is possible that metabolic control status was not adequately evaluated. Another explanation may be that the median follow-up period of nearly 5 years could have been too short to demonstrate the effects of poor glycemic control on survival, as the occurrence and development of micro- and macrovascular complications that ultimately lead to death depends not only on the quality of metabolic control, but also on time for their progression.

Some limitations of this study must be pointed out. Besides the absence of information about HbA_1c levels previously discussed, other potentially important prognostic variables were also missing, such as microalbuminuria and smoking status. Microalbuminuria is an established risk factor for mortality in diabetic patients (1,6,19), but its prognostic information could possibly have been replaced by 24-h proteinuria. Subjects’ smoking statuses were considered to be nonreliable data in this cohort and were not included. Smoking status has been reported as a mortality risk factor in some investigations (3,18), but not in others (6,19,21), so its prognostic importance is debatable. Another possible flaw was the study patients’ selection. Because our cohort was hospital based, the diabetic patients in this investigation may not have been representative of the general type 2 diabetic population of Rio de Janeiro. Nevertheless, the excess mortality ratios observed, as previously discussed, were comparable with those reported in population-based studies from other countries. Finally, the different methods used to ascertain the causes of death might have introduced some variability into the accuracy of the information, although the main source was death certificates (available in more than 90% of the deceased patients), and the other methods were mainly confirmatory.

In conclusion, this prospective study with follow-up for up to 7 years provided evidence that Brazilian type 2 diabetic patients have a more than threefold excess mortality than the background population; this excess mortality was most important in subjects younger than 69 years and was partially explained by an increased cardiovascular mortality. Furthermore, the independent predictors of mortality were older age, preexisting vascular disease, increased 24-h proteinuria, decreased serum HDL cholesterol, and frequent PVCs and QTc interval prolongation on baseline ECGs. The use of β-blockers appeared to be protective. These characteristics could help identify high-risk diabetic patients. Additional investigations with multifactorial interventions are needed to verify if these risk markers can be modified and hence used to reduce the burden of mortality in type 2 diabetic patients.