A final resolution of this dilemma can only be obtained from randomized intervention trials. The purpose of this article is to review the available evidence provided by the only two relevant long-term and large-scale trials that have thus far been completed: the University Group Diabetes Program (UGDP) and the DCCT. The UGDP results generated enormous controversy and were subject to numerous secondary analyses by critics. However, the primary emphasis in this review is on the conclusions that may be derived from the original intention-to-treat analysis plan of the UGDP investigators so as to avoid the bias inherent in post hoc "data dredging."

University Group Diabetes Program

The UGDP was begun in 1961 and completed in 1975 [12-14]. Approximately 1000 patients with NIDDM were randomly assigned to diet treatment plus placebo, a standard insulin regimen, a variable insulin regimen, and two oral hypoglycemic agents. Cardiovascular and microvascular outcomes were studied in all five groups. The standard insulin regimen was a single fixed morning dose of Lente insulin (Lilly, Eli, & Co., Indianapolis, Indiana) not to exceed 20 units. In the variable insulin regimen, the dose of Lente insulin could be altered and other insulins added as deemed necessary by the investigators to keep the fasting plasma glucose level less than 7.2 mmol/L and the 1-hour value after a glucose load less than 13.3 mmol/L. To the consternation of diabetes clinicians and researchers, in 1969 after 7 years of follow-up, the UGDP dropped its tolbutamide arm early because of an unexpected observed increase in all-cause mortality and particularly in cardiovascular mortality [12]. However, the UGDP continued for 6 more years to compare diet and placebo therapy with standard insulin and variable insulin treatments. In the synopsis that follows, the final cardiovascular outcome data in these three treatment groups after a mean follow-up of 12.5 years [13, 14] is reviewed. (See reference 14 for the detailed, complete, and clearly presented UGDP data that form the main basis for this summary.) The discussion then examines pertinent factors in the conduct of the UGDP that help determine the final conclusions that may be drawn from this study.

Table 1 shows the percentage of deaths from all causes and deaths specifically attributable to cardiovascular disease and the percentage of patients who had myocardial infarctions. Comparing all three treatment groups, consisting of approximately 200 patients each, there were no significant differences in any of these major hard outcomes. When further analyzed as event rates per 100 patient years Figure 1, the cumulative incidence of death from all causes and from cardiovascular disease and of myocardial infarctions was also similar in the three treatment groups. The hazard rates for these outcomes were not significantly different. The cumulative incidence of all-cause and cardiovascular mortality was also adjusted for several baseline risk factors, selected from age; sex; race; systolic and diastolic blood pressure measurement; angina pectoris; abnormal electrocardiogram; digitalis use; arterial calcification on radiograph; body weight; and fasting plasma glucose, serum cholesterol, and serum creatinine levels. Using Cox proportional-hazard models, the adjusted cumulative incidence curves for the three treatment groups were even more similar [14].

View larger version (20K): [in this window] [in a new window]   Figure 1. The cumulative incidence of death from all causes (top), cardiovascular death (middle), and myocardial infarction (bottom) in the University Group Diabetes Program. IVAR: insulin variable; ISTD: insulin standard; PLBO: placebo. Reproduced from reference 13 with permission of the American Diabetes Association.

When baseline risk factors were used in logistic regression models that included treatment group assignment, none of the three treatments contributed a significant amount of the variance in outcomes. When the subjects were divided into subgroups according to baseline risk factors, there were no statistically significant trends toward major differences in death rates among the three treatment groups in any particular baseline subgroup, including the subgroup with hypertension as a risk factor at baseline. Not surprisingly, death rates were higher in all treatment groups among those patients who were initially hypertensive. None of the published analyses specifically adjusted outcomes for the subsequent development of hypertension during the trial. However, changes in systolic and diastolic blood pressure at follow-up were relatively small and similar in the three treatment groups. Notably missing from the measured baseline risk factors was the smoking history of the patients.

Therefore, using strict intention-to-treat analyses, it was concluded [13, 14] that no evidence existed that insulin treatment was superior to (or worse than) diet treatment or that more aggressive variable insulin treatment was superior to (or worse than) standard insulin treatment in reducing the risk for cardiovascular disease or mortality. Because of the controversy that was generated by the 7-year findings in the tolbutamide-treated group, an independent Biometric Committee was created to reanalyze the UGDP data at that follow-up time [15]. The report of this committee also indicated that no significant differences were found in cardiovascular outcomes with diet and placebo treatment as compared with standard insulin treatment or variable insulin treatment at this earlier time. Before accepting the formal conclusion of the UGDP that exogenous insulin treatment of patients with NIDDM has no effect on their cardiovascular disease, the patients studied, the treatments used, the degree of adherence to treatment, and post hoc treatment-implemented analyses need to be considered.

Seventy-one percent of the UGDP study participants were women, 29% were men, 54% were white, 46% were black, and they had a mean age of 53 years. They were within 1 year of formal diagnosis of NIDDM and had to tolerate a run-in period of 4 weeks on only diet treatment without symptomatic hyperglycemia or ketosis. Two thirds of the patients were more than 120% of ideal body weight. Although the mean fasting plasma glucose level on entry was approximately 9.2 mmol/L, only 55% of the patients had values exceeding 7.8 mmol/L. After a standard oral glucose load, the mean plasma glucose level at 2 hours was 14.6 mmol/L, but only 67% of the patients had values exceeding 11.1 mmol/L. Thus, a significant proportion of the patients would now probably be categorized as having impaired glucose tolerance rather than diabetes [16]. In that respect, the UGDP was, in part, a primary prevention study. However, to the extent that the UGDP patients were less hyperglycemic or had shorter periods of undetected hyperglycemia before entering the study, detecting a beneficial effect (or for that matter an adverse effect) from exogenous insulin treatment might have been easier, not harder. About half of the patients had one or more of the following cardiovascular baseline risk factors: hypertension, angina pectoris, abnormal electrocardiogram, digitalis use, a serum cholesterol level greater than 7.8 mmol/L, and arterial calcification. Hypertension was the most common factor. Thus, many participants were poised to enter a period where serious cardiovascular events could be expected to occur. On the other hand, they may have passed the point where these events could any longer be substantially influenced by lowering blood glucose levels with exogenous insulin.

The differences in prescribed insulin doses in the UGDP are shown in Figure 2. The average dose in the standard insulin therapy group remained at approximately 14 units per day throughout the study, whereas in the variable insulin therapy group, the daily dose rose progressively to a plateau of approximately 45 units per day. This difference in exposure to exogenous insulin was substantial. If an adverse effect occurred from more aggressive insulin treatment, one might expect it to become evident over an average follow-up of 12.5 years.

View larger version (37K): [in this window] [in a new window]   Figure 2. Daily insulin doses in the variable insulin and standard insulin treatment groups of the University Group Diabetes Program. Reproduced from reference 13 with permission of the American Diabetes Association.

The hypothesis of the UGDP was the glucose hypothesis, that is, that hyperglycemia is a proximate direct cause of diabetic complications. Hence, the glycemic levels achieved need to be considered. These levels were determined quarterly as fasting and 1-hour post-glucose load values. Figure 3 shows that, in the variable insulin group, fasting blood glucose levels were kept at approximately 6.1 mmol/L (approximately equivalent to plasma levels of 7.2 mmol/L) for the first 5 years but then rose to plasma equivalents of approximately 8.1 mmol/L. In the diet plus placebo and the standard insulin treatment groups, fasting glucose levels rose progressively. In general, the difference in glycemic levels between the variable insulin therapy group and the other two groups was of the order of 1.7 to 2.2 mmol/L. This finding was also true of the 1-hour post-load values (Figure 3). Although these numbers cannot be translated directly into glycohemoglobin equivalents, an estimated absolute difference in glycohemoglobin might be of the order of 1% [17].

View larger version (28K): [in this window] [in a new window]   Figure 3. Mean blood glucose levels determined by fasting (top) and 1 hour after ingestion of glucose (30 g/m2) (bottom) in subjects of the University Group Diabetes Program. IVAR: insulin variable; ISTD: insulin standard; PLBO: placebo. Reproduced from reference 13 with permission of the American Diabetes Association.

Even with intention-to-treat analysis, the outcome data from a randomized intervention trial must also be interpreted in light of how well the treatment regimens being compared were actually implemented. The differences in daily insulin dosage and in glycemic levels achieved can be viewed as one set of markers of adherence to the treatment regimens. However, other measures of adherence were also examined by the UGDP investigators. They defined adherence in terms of the percentage of follow-up periods in which some or all of the assigned treatment was reportedly taken. Using this criterion, 55% of the placebo therapy group, 60% of the standard insulin therapy group, and 70% of the variable insulin therapy group were thought to be taking all or some of their assigned treatment during more than 75% of the quarterly follow-up periods.

The special Biometric Committee analyzed adherence in terms of patient-years on assigned treatment [15]. By this criterion, in the standard insulin therapy group, only 30% of the assigned patient-years on this regimen were received as such by the patients. In 55% of the assigned patient-years of treatment in the standard insulin therapy group, insulin dose adjustment was carried out, and in 15%, diet or no treatment was received. However, because the average daily dose in the standard insulin therapy group remained approximately 14 units, most of the insulin adjustments received by this group probably consisted of only small and temporary increments in dosage. Of greater importance, during 85% of assigned patient-years in the variable insulin treatment group and during 97% of assigned patient-years on diet plus placebo treatment, the correct treatment was thought to have been received.

Although some of these estimates of adherence are reassuring, indices of compliance with treatment did not correlate well with indices of blood glucose level control achieved. Among patients thought to have exhibited good adherence to treatment, only 29% of standard insulin and 38% of variable insulin therapy patients were classified as being in good control by the UGDP investigators. Furthermore, of patients who were defined as exhibiting poor adherence, approximately these same percentages were thought to be in good blood glucose level control.

In their final report [14], the UGDP study group presented analyses of mortality and cardiovascular outcomes in subgroups of patients categorized by levels of blood glucose control (Table 2). In patients defined as being in good control, total mortality was 17% with variable insulin treatment compared with 21% with standard insulin treatment and 35% with diet plus placebo treatment. In patients defined as being in fair or poor control, the percentage of deaths was somewhat higher in both groups treated with insulin but not in the placebo group. The same differences were observed when cardiovascular mortality was specifically analyzed. These post hoc subgroup analyses do not provide definitive answers to the major question of whether glycemic control influences the development of cardiovascular disease. They do, however, suggest a trend toward better cardiovascular outcomes with lower levels of blood glucose on insulin treatment (Table 2). None of these analyses suggest a significantly increased risk for cardiovascular disease with the more intensive variable insulin regimen.

As noted above, assessments of adherence to treatment did not correlate particularly well with quarterly clinic glucose measurements. The latter could be taken as a better measure of compliance in the UGDP. Because "compliers" may have had better outcomes than "noncompliers" in the active treatment groups, Table 2 also suggests that compliance per se may be another confounder in interpreting the results.

Critics of the UGDP argued that adherence to monitoring and to treatment assignment was so compromised that 42% of all insulin and placebo recipients were either off the assigned treatment or were unavailable for follow-up for half of the study's duration [18, 19]. In addition, these critics thought that a disproportionate number of patients who died of cardiovascular causes in the variable insulin therapy group had baseline risk factors that made them unsuitable for study. After pruning the three treatment groups to about 35% of their original cohort sizes, the critics calculated that the percentage of deaths caused by cardiovascular disease was approximately 4% for variable insulin as compared with 15% for standard insulin and 17% for placebo. They then suggested that this analysis supported a beneficial effect of the variable insulin regimen. It was not unreasonable for skeptics to look for quirks or deficiencies in the UGDP data to explain why no beneficial treatment effect was observed on intention-to-treat analysis. But it was not reasonable or legitimate to use such a tailored post hoc subgroup analysis to imply that the UGDP actually showed that insulin treatment and control of blood glucose levels do prevent or reduce cardiovascular mortality [19].

In summary, a review of the UGDP trial suggests that in many ways it was a well-designed, pioneering effort to answer a question still being posed 25 years later: Does intensive insulin therapy reduce (or increase) the risk for cardiovascular disease in diabetic patients? Much was learned that was of benefit to subsequent investigators, but for several reasons, the UGDP has not provided us with a definitive answer. First, the study was probably underpowered. The investigators stated that they had only an 80% chance of detecting an approximately 50% reduction in event rates (presumably aggregated) in the variable insulin treatment group compared with the placebo group. Furthermore, event rates in the placebo group turned out to be lower than had been originally predicted. Second, the entry criteria may have resulted in selection of patients with cardiovascular disease that was too far advanced. Third, smoking history was unavailable, and differences in this or other unmeasured baseline risk factors could have confounded any real treatment effect. Fourth, the treatment regimens were probably incompletely implemented by comparison with a trial such as the DCCT [6, 20]. Fifth, the chronic glycemic levels that were maintained were uncertain. The investigators depended on two blood glucose levels measured at quarterly visits. Patients in all treatment groups may have altered their treatment and their glucose levels for a few days before these examinations. Sixth, if glycemic levels do, in fact, influence the development of cardiovascular disease, the blood glucose level differences obtained between the variable insulin therapy group and either the placebo or standard insulin therapy group may have been too small to allow detection of any associated differences in cardiovascular events.

Finally, any relation between glycemia and cardiovascular disease may be such that the UGDP was operating at an experimentally difficult portion of such a curve. For example, in the DCCT, the relation between the risk for developing retinopathy and the mean HbA_1c level was found to be exponential [6]. The risk gradient was such that the median HbA_1c value of 7.2% in the intensive treatment group as compared with the median HbA_1c of 9% in the conventional treatment group was associated with a substantial difference in retinopathy event rates. This difference was readily detected by the DCCT because of the number of patients studied and the duration of follow-up. But the UGDP may have been operating at lower HbA_1c levels than the DCCT in all treatment groups and with smaller differences between them. This possibility resulted in a failure to detect a small difference in cardiovascular event rates if the putative relation between glycemia and cardiovascular disease were similar to that for glycemia and retinopathy (which may not be the case). Persons who have only impaired glucose tolerance are also at increased risk for cardiovascular disease [21, 22]. Moreover, such persons, as well as those with even subtler degrees of hyperglycemia, have ultrasonographic evidence of carotid artery atherosclerosis and electrocardiographic abnormalities [23]. This finding suggests the opposite theoretical possibility, namely, that most of the risk may be conferred by minimal but longstanding hyperglycemia. If so, the glycemic levels in the UGDP treatment groups may have already been at or near the upper plateau of the risk gradient, which would again make it difficult to show significant differences in outcome.

Diabetes Control and Complications Trial

In contrast to the UGDP, the DCCT was not primarily designed to test the hypothesis that blood glucose control would influence the risk for cardiovascular disease. The patients with IDDM who were studied were too young and it was too early in the course of their diabetes to expect significant cardiovascular event rates during the projected follow-up period [6]. Nonetheless, 17 initial major cardiovascular events were recorded: 14 in the conventionally treated group as compared with 3 in the intensively treated group [24]. Total major cardiovascular and peripheral vascular events were 40 in the conventional group as compared with 23 in the intensive group [21]. With intensive treatment, the risk for cardiac events was reduced by 78%, and the risk for combined cardiac and peripheral vascular events, by 42%, but these differences in risk were not statistically significant [24]. Intensive insulin treatment also significantly reduced low-density lipoprotein cholesterol levels but, conversely, significantly increased body mass index [6, 20]. A large percentage of the 12 patients who had myocardial infarction or sudden death had a parental history of myocardial infarction and smoked; thus, they were at higher risk for such events than the average risk of the DCCT cohort [24].

At this point, the results of the DCCT in IDDM suggest a possible trend toward benefit, and certainly provide no evidence of an adverse effect, of intensive insulin treatment on cardiovascular disease. Further long-term follow-up of the DCCT cohort over the next 10 years might provide additional evidence on this issue if there is a carryover effect from the preceding period of glycemic difference between the two treatment groups. However, the average total daily insulin dose in the intensive treatment group of the DCCT was less than 10% higher than in the conventional treatment group [20]. Unless acceleration of atherosclerosis and cardiovascular diseases is exquisitely sensitive to a small difference in exogenous insulin dose or unless more physiologic insulin replacement with short-acting regular insulin before all meals is atherogenic, it is unlikely that follow-up of the DCCT cohort will ever show an adverse effect of exogenous insulin on cardiovascular disease.

Conclusions

The evidence from the largest and longest randomized intervention trials thus far completed does not resolve the issue of whether insulin treatment prevents or delays (or accelerates) the development of cardiovascular disease in NIDDM or IDDM. Exogenous insulin possibly has both a beneficial effect by reducing hyperglycemia and an adverse effect by promoting atherogenesis or thrombogenesis. If so, these effects could neutralize each other so that no net effect of intensive insulin treatment on the risk for cardiovascular disease in either type of diabetes will ever be apparent. Currently, the long-running United Kingdom Prospective Diabetes Study [25] is comparing diet, several oral hypoglycemic agents, and insulin therapies in the treatment of patients with NIDDM. This study is projected to have a 91% power to detect 20% treatment group differences in the first cardiovascular outcome. We hope that at its anticipated conclusion in 1997, the United Kingdom study will resolve this major issue.