To be able to draw a random sample of the diabetic population in the municipality of Horsens, Denmark, the insulin-treated part of the diabetic population was delineated by the prescription method. From 1 November 1992 to 1 July 1993, 328 users of insulin were registered. The municipality had, at that point in time, 55 265 inhabitants, giving a prevalence of insulin-treated diabetes mellitus of 0.6%.
A randomly drawn age- and gender-stratified sample of 120 users of insulin in the age group of 40–75 years was asked to participate in the study. The diagnosis of diabetes mellitus according to the criteria of WHO  was verified in each participant. If fasting C-peptide was below 0.30 nmol L−1, persons were registered as having Type 1 diabetes mellitus; the remainder had insulin-treated Type 2 diabetes mellitus.
The autonomic nervous system function was assessed by the deep breathing test. This test ‘approaches the optimal test for the cardiovagal function’ according to a recent expert committee on non-invasive tests of the autonomic function . The examination was carried out in the morning after a light meal. Each participant was instructed to breath deeply whilst sitting up , each inspiration and expiration lasting 5 s. During the procedure, a six-lead electrocardiogram (ECG) was obtained at 50 mm s−1, and the beginning of each inspiration and expiration marked on the ECG with a pencil. The ECG was blinded by a random number, before it was interpreted. In each respiratory cycle the shortest R–R interval during inspiration and the longest during expiration were selected using a pair of compasses, and the R–R intervals were measured by a ruler to the nearest 0.5 mm. The ECGs were examined by two different people, and if a reading differed by more than 0.5 mm, the ECG was measured once more to reach agreement. The ratio between the longest and the shortest R–R interval in each respiratory cycle, the E/I ratio, was calculated and averaged over three consecutive respiratory cycles. The variability of the R–R interval was expressed as the E/I ratio because this reading, as opposed to the difference between maximal and minimal heart rates, is independent of the resting heart rate in normals [12, 13]. The E/I ratio was defined as normal if the reading was higher than the age-adjusted 5th percentile in healthy non-diabetics. As previously reported, values of E/I ratio > 1 + exp(−1.12 − 0.0198 × age [years]) can be regarded as normal .
A 12-lead ECG taken at rest was blindly interpreted according to the Minnesota code  by two experienced technicians. Evidence of left ventricular hypertrophy was taken as suggested by the World Health Organization .
The participants were asked about myocardial infarction (MI), smoking habits, angina pectoris (Rose questionnaire ) and lower leg oedema or pulmonary congestion. From pre-coded forms, the questions were read aloud to each participant. A history of MI was only accepted for the analysis if verified by hospital records.
To avoid observer-induced bias, all blood pressure (BP) values were recorded with a Hawksley random zero manometer . The BP was measured twice or more until the difference between two consecutive readings was ≤ 4 mmHg, the mean of the two was used in the calculations.
A fasting blood sample was taken at 08.00 h in the morning. Blood glucose was measured enzymatically (Cobas Mira, Roche), C-peptide was analysed radioimmunologically (antibody K6, Novo Nordisk), cholesterol was measured enzymatically (CHOD-PAP) and haemoglobin A1c was quantified photometrically (Waters HPLC). A morning clean-catch midstream urine specimen was tested for nitrite and leucocyte esterase activity with a dipstick (Boehringer–Mannheim, LN). If any of the two reagent squares responded, the urine was cultured and, in case of significant bacteriuria, antibiotics were given. As soon as the urine was dipstick-negative, three overnight urines were collected. The urine albumin excretion rate (UAE) was evaluated by turbidimetry (Cobas Mira).
The 10-year risk of future CHD was estimated by an algorithm developed in the Framingham Heart Study  based on gender, age, systolic BP, total/HDL cholesterol, left ventricular hypertrophy, diabetes and smoking status. The algorithm using systolic BP is recommended because systolic BP is more accurately determined, has a wider range and stronger predictive power compared with diastolic BP .
A maximal symptom-limited bicycle ergometer exercise test was accomplished. The workload was initially 25 watts (W), but increased by 25 W every 2 min. Prior to and each minute during the test, a standard 12-lead ECG was recorded, and the participants questioned about possible chest pain. If present, the pain was quantified according to the Borg scale . Blood pressure was measured manually before, every 2 min during and 1, 2, 4 and 6 min after exercise. Horizontal or descending ST-segment depression of at least 0.1 mV measured 80 ms after the J-point in three consecutive cycles was considered a significant sign of ischaemia. If ST depression was present on the resting ECG, an additional 0.1 mV was required. The ECGs were blinded by a random number before interpretation.
From the results of the exercise test, the Veterans Affairs (VA) Prognostic Score was calculated . This score was developed by Froelicher, who examined a consecutive series of routine clinical exercise tests in 3134 men. In the study by Froelicher, cardiovascular death or non-fatal MI was independently predicted by the exercise-induced change in systolic BP, ST depression, exercise capacity and the presence of congestive heart failure/use of digoxin. A score including these four parameters was constructed:
5 × CHF/digoxin [yes = 1]+ (ST depression [mm]) + systolic BP change score [0 > 40 mmHg; 1: 31 − 40 mmHg; 2: 21 − 30 mmHg; 3: 11 − 20 mmHg; 4: 0 − 10 mmHg; 5: < 0 mmHg]– METs
A low score (less than − 2) was consistent with an annual cardiovascular mortality below 2%.
Angina pectoris was present if the Rose questionnaire  was positive or chest pain was registered in the presence of ST depression during the exercise test.
The study complies with the Declaration of Helsinki II, and the study protocol was approved by the local ethics committee. Written informed consent was obtained from each participant after verbal and written information was given. The results of each individual examination were forwarded to the participant as well as the general practitioner, and in cases with abnormal results, the participant was offered a consultation at the hospital.
The associations between the E/I ratio and each coronary risk factor were expressed by Spearman's coefficient of correlation. Differences between two groups were tested using the Mann–Whitney test in cases of continuous variables and Fisher's exact test in cases of proportions. Divergence between more than two groups was examined by the Kruskal–Wallis test. Tests results with a two-sided P-value below 0.05 were taken as statistically significant.