To determine the relation between exposure to glycaemia over time and the risk of macrovascular or microvascular complications in patients with type 2 diabetes.

Prospective observational study.

23 hospital based clinics in England, Scotland, and Northern Ireland.

4585 white, Asian Indian, and Afro-Caribbean UKPDS patients, whether randomised or not to treatment, were included in analyses of incidence; of these, 3642 were included in analyses of relative risk.

Primary predefined aggregate clinical outcomes: any end point or deaths related to diabetes and all cause mortality. Secondary aggregate outcomes: myocardial infarction, stroke, amputation (including death from peripheral vascular disease), and microvascular disease (predominantly retinal photo-coagulation). Single end points: non-fatal heart failure and cataract extraction. Risk reduction associated with a 1% reduction in updated mean HbA_1c adjusted for possible confounders at diagnosis of diabetes.

The incidence of clinical complications was significantly associated with glycaemia. Each 1% reduction in updated mean HbA_1c was associated with reductions in risk of 21% for any end point related to diabetes (95% confidence interval 17% to 24%, P<0.0001), 21% for deaths related to diabetes (15% to 27%, P<0.0001), 14% for myocardial infarction (8% to 21%, P<0.0001), and 37% for microvascular complications (33% to 41%, P<0.0001). No threshold of risk was observed for any end point.

In patients with type 2 diabetes the risk of diabetic complications was strongly associated with previous hyperglycaemia. Any reduction in HbA_1c is likely to reduce the risk of complications, with the lowest risk being in those with HbA_1c values in the normal range (<6.0%).

The UK prospective diabetes study (UKPDS), a clinical trial of a policy of intensive control of blood glucose after diagnosis of type 2 diabetes, which achieved a median haemoglobin A_1c (HbA_1c) of 7.0% compared with 7.9% in those allocated to conventional treatment over a median 10.0 years of follow up, has shown a substantial reduction in the risk of microvascular complications, with a reduction in the risk of myocardial infarction of borderline significance.¹ Complementary information for estimates of the risk of complications at different levels of glycaemia can be obtained from observational analyses of data during the study.

In patients with type 2 diabetes previous prospective studies have shown an association between the degree of hyperglycaemia and increased risk of microvascular complications,²³ sensory neuropathy,³⁴ myocardial infarction,²⁵⁶ stroke,⁷ macrovascular mortality,^{8 --10} and all cause mortality.^{911 --14} Generally, these studies measured glycaemia as being high or low or assessed glycaemia on a single occasion, whereas repeated measurements of glycaemia over several years would be more informative.

The existence of thresholds of glycaemia ---that is, concentrations above which the risk of complications markedly increases ---has not been studied often in patients with type 2 diabetes. The relative risk for myocardial infarction seems to increase with any increase in glycaemia above the normal range,¹⁵¹⁶ whereas the risk for microvascular disease is thought to occur only with more extreme concentrations of glycaemia.^{17 --19} The diabetes control and complications trial (DCCT) research group showed an association between glycaemia and the progression of microvascular complications in patients with type 1 diabetes for haemoglobin A_1c over the range of 6-11% after a mean of six years of follow up.²⁰ No specific thresholds of glycaemia were identified above which patients were at greater risk of progression of retinopathy, increased urinary albumin excretion, or nephropathy.^{19 --21} Nor has any threshold of fasting plasma glucose concentration been identified for cardiovascular deaths.²²²³

We evaluated the relation between exposure to glycaemia over time and the development of macrovascular and microvascular complications and compared this with the results of the UKPDS trial of a policy of intensive control of blood glucose control.¹

Details are presented in the companion paper (UKPDS 36) published in this issue (see page 412).

Participants in observational analysis. Of 5102 patients, 4585 white, Asian Indian, and Afro-Caribbean patients who had haemoglobin A_1c (HbA_1c) measured three months after the diagnosis of diabetes were included in analyses of incidence rates. Of these, 3642 with complete data for potential confounders were included in analyses of relative risk. Complete data were required for all participants included in the multivariate observational analyses. For this reason there are fewer (3642) participants in these analyses than in the clinical trial, despite the inclusion of patients not randomised in the trial. Their characteristics are presented in table 1.

Participants in UKPDS blood glucose control study. After a three month dietary run-in period patients were stratified on the basis of fasting plasma glucose concentration and body weight. The 3867 patients who had fasting plasma glucose concentrations between 6.1 and 15.0 mmol/l and no symptoms of hyperglycaemia were randomised to a policy of conventional glucose control, primarily with diet, or to an intensive policy with sulphonylurea or insulin.^{124 --26} The aim in the group allocated to conventional control (n=1138) was to obtain fasting plasma glucose concentration <15 mmol/l, but if concentrations rose to [gt-or-equal, slanted]15 mmol/l or symptoms of hyperglycaemia developed patients were secondarily randomised to non-intensive use of these pharmacological treatments, with the aim of achieving fasting plasma glucose concentrations <15 mmol/l without symptoms. The aim in the group allocated to intensive control (n=2729) was to achieve fasting plasma glucose concentration <6 mmol/l, primarily with a single pharmacological treatment. Details of treatments and their effect on glucose control have been published elsewhere.¹

Biochemical methods have been reported previously.²⁷ Haemoglobin A_1c was measured by high performance liquid chromatography (Biorad Diamat automated glycosylated haemoglobin analyser), the range for people without diabetes being 4.5% to 6.2%.²⁷²⁸ Baseline variables are quoted for measurements after the initial dietary run-in period.

Exposure to glycaemia was measured firstly at baseline as haemoglobin A_1c concentration and secondly over time as an updated mean of annual measurements of haemoglobin A_1c concentration, calculated for each individual from baseline to each year of follow up. For example, at one year the updated mean is the average of the baseline and one year values and at three years is the average of baseline, one year, two year, and three year values.

The clinical end points and their definitions are shown in the box in the companion paper (UKPDS 36) published in this issue (see page 412).

Incidence rates by category of glycaemia. The unadjusted incidence rates were calculated by dividing the number of people with a given complication by the person years of follow up for the given complication within each category of updated mean haemoglobin A_1c concentration and reported as events per 1000 years of follow up.²⁹ The categories were defined (median values in parentheses) as: <6% (5.6%), 6-<7% (6.5%), 7-<8% (7.5%), 8-<9% (8.4%), 9-<10% (9.4%), and [gt-or-equal, slanted]10% (10.6%) over the range of updated mean haemoglobin A_1c of 4.6-11.2% (1st-99th centile). Follow up time was calculated from the end of the initial period of dietary treatment to the first occurrence of that complication or loss to follow up, death from another cause, or to the end of the study on 30 September 1997 for those who did not have that complication. Hence, follow up time is equivalent to duration of diabetes. For myocardial infarction and stroke for participants who had a non-fatal followed by a fatal event, the time to the first event was used. The rates were therefore for single and not recurrent events. The median follow up time for all cause mortality was 10.4 years.

We calculated adjusted incidence rates for each category of updated mean haemoglobin A_1c using a Poisson regression model adjusted for male sex, white ethnic group, age at diagnosis 50-54 years, and duration of diabetes 7.5-12.5 years and expressed in events per 1000 person years of follow up. These parameters were chosen to reflect the median age and duration of diabetes and the modal ethnic group and sex.

Hazard ratio and risk reduction. To assess potential associations between updated mean haemoglobin A_1c and complications we used proportional hazards regression (Cox) models. Potential confounding risk factors included in all Cox models were sex, age, ethnic group, smoking (current/ever/never) at time of diagnosis of diabetes, and baseline high and low density lipoprotein cholesterol, triglyceride, presence of albuminuria (> 50 mg/l measured in a single morning urine sample) measured after three months' dietary treatment, and systolic blood pressure represented by the mean of measures at two and nine months after diagnosis. The hazard ratio was used to estimate the relative risk. At each event time, the updated mean haemoglobin A_1c value for individuals with an event was compared with the updated value of those who had not had an event by that time. The updated mean value was included as a time dependent covariate to evaluate glucose exposure during follow up.²⁰²⁹³⁰ It was included as a categorical variable in the categories of glycaemia listed above, with the lowest category (<6%) as the reference category assigned a hazard ratio of 1.0 and with the highest category [gt-or-equal, slanted]9%. (This is reflected in the point estimates as shown in figures 3 and 4.) Separate models, with updated mean haemoglobin A_1c as a continuous variable, were used to determine reduction in risk associated with a 1% reduction in haemoglobin A_1c (see regression lines in figures 3 and 4). We evaluated the presence of thresholds by visual inspection. The 95% confidence intervals were calculated on the basis of the floating absolute risk.³¹ Log linear relations are reported by convention.¹³² The risk reduction associated with a reduction of 1% updated mean haemoglobin A_1c was calculated as 100% minus the reciprocal of the hazard ratio expressed as a percentage. The risk reduction from the continuous variable model associated with a 1% reduction in observed haemoglobin A_1c was compared with the risk reduction seen in the UKPDS intervention trial of an intensive versus a conventional policy of blood glucose control, for which no adjustment for potential confounders was required as they were balanced by randomisation.¹

To assess whether the association between mean updated haemoglobin A_1c and complications was independent of randomisation, separate models included mean updated haemoglobin A_1c and randomisation to either intensive or conventional policy, as well as all potential confounders listed above. The model for all end points related to diabetes included 3005 individuals.

Statistical analyses were performed with SAS version 6.12.³³

The risk of each of the microvascular and macrovascular complications of type 2 diabetes and cataract extraction was strongly associated with hyperglycaemia as measured by updated mean haemoglobin A_1c. The incidence rates for any end point related to diabetes, adjusted for age, sex, ethnic group, and duration of diabetes, increased with each higher category of updated mean haemoglobin A_1c, with no evidence of a threshold and with a threefold increase over the range of updated mean haemoglobin A_1c of <6% (median 5.6%) to [gt-or-equal, slanted]10% (median 10.6%) (figs 1 and 2). The unadjusted and adjusted incidence rates are shown in table 2. Figure 2 shows the adjusted incidence rates for myocardial infarction and microvascular end points. The increase in the incidence rate for microvascular end points was greater over the range of increasing glycaemia than was the increase in the incidence rate for myocardial infarction. Thus at near normal concentrations of updated mean haemoglobin A_1c the risk of myocardial infarction was twice to three times that of a microvascular end point, whereas in the highest category of haemoglobin A_1c concentration ([gt-or-equal, slanted]10%) the risks were of the same order.

The estimated hazard ratios associated with different categories of updated mean haemoglobin A_1c concentration, relative to the lowest category, are shown as log linear plots in figures 3 and 4. Mortality related to diabetes and all cause mortality were both strongly associated with glycaemia (P<0.0001). The risk of each of the complications evaluated rose with increasing updated mean haemoglobin A_1c concentration both before and after adjustment for baseline variables including age, sex, ethnic group, lipid concentrations, blood pressure, smoking, and albuminuria. The decrease in risk for each 1% reduction in updated mean haemoglobin A_1c concentration is shown in table 3 and figures 3 and 4. The glycaemia associated reduction in risk for microvascular end points and for amputation or death from peripheral vascular disease was greater (by 37% and 43% per 1% reduction in haemoglobin A_1c concentration, respectively, each P<0.0001) than it was for myocardial infarction, stroke, and heart failure (by 14% (P<0.0001), 12% (P=0.035), and 16% (P=0.021) per 1% haemoglobin A_1c, respectively) (fig 4). In models that included a variable for conventional control of blood glucose or intensive control with either sulphonylurea or insulin, updated mean haemoglobin A_1c remained associated with all complications, although for stroke and heart failure, where the numbers of events were lower than in the previous analyses, these were no longer significant. In these models, treatment of blood glucose per se had no association with any complication beyond that of mean updated haemoglobin A_1c.

There was no indication of a threshold for any complication below which risk no longer decreased nor a level above which risk no longer increased. The updated mean haemoglobin A_1c showed steeper relations than did baseline haemoglobin A_1c (table 3), and when both glycaemic variables were included in a model for all complications of diabetes only updated mean haemoglobin A_1c reached significance (P<0.0001).

This observational analysis shows highly significant associations between the development of each of the complications of diabetes, including mortality, across the wide range of exposure to glycaemia that occurs in patients with type 2 diabetes. This association remained after adjustment for other known risk factors, including age at diagnosis, sex, ethnic group, systolic blood pressure, lipid concentrations, smoking, and albuminuria. Each 1% reduction in haemoglobin A_1c was associated with a 37% decrease in risk for microvascular complications and a 21% decrease in the risk of any end point or death related to diabetes. The association with glycaemia was less steep for stroke and heart failure, for which blood pressure is a major contributing factor.³²³⁴³⁵ In patients within the lowest category of updated mean haemoglobin A_1c the incidence of myocardial infarction was higher than that of microvascular disease.⁵ These results suggest that, in these people, the effect of hyperglycaemia itself may account for at least part of the excess cardiovascular risk observed in diabetic compared with non-diabetic people beyond that explained by the conventional risk factors of dyslipidaemia, hypertension, and smoking.³⁶ The rate of increase of relative risk for microvascular disease with hyperglycaemia was greater than that for myocardial infarction, which emphasises the crucial role of hyperglycaemia in the aetiology of small vessel disease and may explain the greater rate of microvascular complications seen in populations with less satisfactory control of glycaemia.

This observational analysis provides an estimate of the reduction in risk that might be achieved by the therapeutic lowering of haemoglobin A_1c by 1.0%, but it is important to realise that epidemiological associations cannot necessarily be transferred to clinical practice. Tissue damage from previous hyperglycaemia may not promptly be overcome, but the results are not inconsistent with those achieved by the policy of intensive glucose control in the clinical trial.¹ This suggests that the reduction in glycaemia obtained over a median 10 years of follow up of the trial, comparing median haemoglobin A_1c 7.0% with 7.9%, provided much of the benefit that could be expected from that degree of improved glycaemic control. Our results suggest that intensive treatment with sulphonylurea or insulin does not have an effect beyond that of lowering blood glucose concentration with respect to altering risk. The 16% risk reduction (P=0.052) in myocardial infarction in the clinical trial in the group allocated to a policy of intensive blood glucose control (associated with a 0.9% difference in haemoglobin A_1c) was similar to the 14% risk reduction seen in the epidemiological analysis, which was associated with a 1% reduction in concentration of updated mean haemoglobin A_1c. The UKPDS clinical trial evaluated a policy of intensive glucose control based primarily on single pharmacological treatments to enable evaluation of the individual treatments. Now that the UKPDS has shown that improved glucose control reduces the risk of complications and that the treaments used are safe in clinical practice, a larger reduction in haemoglobin A_1c might be achieved by the earlier use of combination treatments or by the use of newer treatments, which could further reduce the risk of myocardial infarction.

The observational analysis extends the range of hyperglycaemia studied in the UKPDS by including participants who, throughout the study, had near normal glucose concentrations on dietary treatment alone and participants who could never be treated by dietary treatment alone.³⁷ The UKPDS population was likely to be at lower risk of complications than other diabetic populations. Hence, the incidence rates we report are perhaps lower than might be observed in other diabetic populations as the cohort was newly diagnosed with diabetes, excluded old or ill patients, and contained a small proportion (6%) of participants with impaired fasting glycaemia.³⁸ None the less, the decrease in relative risk is unlikely to be different from other diabetic populations.

We observed no thresholds of glycaemia for any type of complication of diabetes. This suggests that there is no specific target value of haemoglobin A_1c for which one should aim but that the nearer to normal the haemoglobin A_1c concentration the better. In reality, it is difficult to obtain and maintain near normal concentrations of haemoglobin A_1c in patients with type 2 diabetes, particularly in those with a high concentration of haemoglobin A_1c at diagnosis of diabetes.³⁷ Intensification of treatment by adding insulin to improve the relatively modest reduction in glycaemia achieved with oral hypoglycaemic treatments can be constrained by reluctance from patients and providers because, in part, of side effects such as hypoglycaemia or weight gain. These observational analyses, together with the results of the clinical trial, however, indicate that any improvement in a raised haemoglobin A_1c concentration is likely to reduce the risk of diabetic complications.

The magnitude of the risk reduction associated with a 1% reduction in haemoglobin A_1c concentration for myocardial infarction and microvascular disease (mostly retinopathy) was consistent with that observed in a cohort of patients from Wisconsin.² As in this analysis, a stronger association with haemoglobin A_1c concentration was observed for amputation than for ischaemic heart disease, possibly because glycaemia increases the risk of microvascular disease, neuropathy, and peripheral arterial disease, each of which increases the risk of amputation.^{481839 --41} The estimated 14% decrease in all cause mortality per 1% reduction in haemoglobin A_1c concentration was similar to that seen in other studies that have assessed glycaemia as haemoglobin A_1c as a continuous variable (per 1% change) in multivariate proportional hazards models.⁹

Both the observational and clinical trial analyses of an intensive glucose control policy suggest that even a modest reduction in glycaemia has the potential to prevent deaths from complications related to diabetes as cardiovascular and cerebrovascular disease account for 50-60% of all mortality in this and other diabetic populations.^{842 --47} Individuals with very high concentrations of glycaemia would be most likely to benefit from reduction of glycaemia as they are particularly at risk from the complications of type 2 diabetes, but the data suggest that any improvement in glycaemic control across the diabetic range is likely to reduce the risk of diabetic complications.

The cooperation of the patients and many NHS and non-NHS staff at the centres is much appreciated. We thank Mr Dick Jelfs for the measurement of haemoglobin A_1c. Details of participating centres can be found on the BMJ 's website.

Editorial by Tuomilehto

Figure 1 Incidence rate and 95% confidence intervals for any end point related to diabetes by category of updated mean haemoglobin A1c concentration, adjusted for age, sex, and ethnic group, expressed for white men aged 50-54 years at diagnosis and with (more ...)

Figure 2 Incidence rates and 95% confidence intervals for myocardial infarction and microvascular complications by category of updated mean haemoglobin A1c concentration, adjusted for age, sex, and ethnic group, expressed for white men aged 50-54 years (more ...)

Figure 3 Hazard ratios, with 95% confidence intervals as floating absolute risks, as estimate of association between category of updated mean haemoglobin A1c concentration and any end point or deaths related to diabetes and all cause mortality. Reference (more ...)

Figure 4 Hazard ratios, with 95% confidence intervals as floating absolute risks, as estimate of association between category of updated mean haemoglobin A1c concentration and myocardial infarction, stroke, microvascular end points, cataract extraction, (more ...)

Table 1 Characteristics of patients included in proportional hazards model measured after three month dietary run-in after diagnosis of diabetes and those included in UKPDS glucose control study.1 Figures are means (SD) unless stated otherwise

Table 2 Incidence of complications in patients with type 2 diabetes by category of updated mean haemoglobin A1c concentration (%). Rates per 1000 person years' follow up adjusted in Poisson regression model to white men aged 50 to 54 years at diagnosis (more ...)

Table 3 Observational analysis of relation between glycaemic exposure and complications of diabetes as estimated by decrease in risk for 1% reduction in haemoglobin A1c (HbA1c) concentration, measured at baseline and as updated mean, controlled for age (more ...)