Proinsulin Concentration Is an Independent Predictor of All-Cause and Cardiovascular Mortality: An 11-year follow-up of the Hoorn Study

The Hoorn Study is a population-based cohort study on type 2 diabetes in the general Dutch population. The study population and research design have been described in detail previously (15). In summary, in 1989, 3,553 men and women, aged 50–75 years, were randomly selected from the population register of the middle-sized Dutch town of Hoorn. Of the 2,540 subjects (71.5%) who agreed to participate, 56 non-Caucasians were excluded. Therefore, the study cohort consisted of 2,484 men and women. All subjects gave written informed consent. The Ethics Committee of the VU University Medical Center approved the study. After exclusion of patients with known diabetes (n = 90), as defined by use of insulin or blood glucose–lowering agents, or a prescribed diet for diabetes, a subgroup of 1,109 participants from the initial study cohort, stratified by age, sex, and 2-h postload glucose values, was invited for a second oral glucose tolerance test. For reasons of efficiency, insulin and proinsulin were measured in duplicate (∼2 weeks apart) in a subsample of 630 participants of the stratified sample of 1,109 participants, including participants with impaired glucose tolerance and all participants with newly detected diabetes (15). Subjects with values below the lower limit of sensitivity of the insulin or proinsulin assay were excluded. For the present analyses, subjects were included if either insulin or proinsulin values were available in duplicate (n = 605). Of these 605 subjects, one subject was lost to follow-up. Thus, the present study population comprised 604 subjects. A total of 602 participants had duplicate insulin values, and proinsulin values were available for 511 participants.

Anthropometric measurements were obtained from all participants. Weight was measured while participants were wearing light clothes only. Participants were not wearing shoes while height was measured. BMI was calculated as weight (in kilograms) divided by the square of height (in meters). Waist circumference (centimeters) was measured following a standardized procedure (15).

Systolic and diastolic blood pressure were determined at the right upper-arm, after 5 min of rest in seated participants, with a random-zero sphygmomanometer (Hawksley-Gelman, Lancing, U.K.). The average of duplicate measurements was used.

All laboratory analyses were performed at the VU University Medical Center in Amsterdam. Fasting plasma glucose concentration and 2-h postload plasma glucose concentration were determined by the glucose dehydrogenase method (Merck, Darmstadt, Germany). Immuno-specific insulin was measured in serum by a double-antibody radio immunoassay (lot SP21; Linco Research, St. Louis, MO) in which proinsulin and 32,33 split proinsulin cross-reacts by <0.2%. Proinsulin was measured by a double-antibody radioimmunoassay (Lilly Laboratory for Clinical Research, Indianapolis, IN) in which 31,32 proinsulin cross-reacts by 63% (18). Triglycerides, total cholesterol, and HDL cholesterol measured on the first visit were determined by enzymatic techniques (Boehringer-Mannheim, Mannheim, Germany).

There is a continuous follow-up to register mortality from the participants of the Hoorn Study, with the municipal register of the city of Hoorn providing information about the vital status of the participants. Information of causes of death was obtained from medical records of general practitioners and from the local hospital. Causes of death were coded according to the ICD-9 (19). CVD mortality was defined as ICD-9 codes 390–459 (diseases of the circulatory system) or 798 (sudden death, cause unknown) because, generally, sudden death is of cardiovascular origin (20). The mortality follow-up was completed by January 2003.

Glucose tolerance status (based on fasting and postload glucose values) and fasting and postload insulin and proinsulin values were based on the mean of two measurements. If only one value was determined, values were based on one measurement. This was done for 1 subject for fasting plasma glucose, 16 subjects for postload plasma glucose, 16 subjects for fasting insulin, and 61 subjects for fasting proinsulin.

The mean fasting and postload glucose values of duplicate oral glucose tolerance tests were used to define the glucose tolerance status of the participants (21). Insulin resistance was estimated by HOMA-IR, calculated as fasting insulin (in microunits per milliliter) × fasting glucose (in millimoles per liter)/22.5. β-Cell function was estimated by HOMA-B, calculated as fasting insulin × 20/fasting glucose (in millimoles per liter) − 3.5 (16).

Due to skewed distributions, baseline characteristics are shown as medians with interquartile ranges. Differences in baseline measurements for groups of glucose tolerance status were examined using linear regression for continuous data and logistic regression for dichotomous data, adjusted for age. Correlations among insulin, proinsulin, HOMA-IR, and HOMA-B were calculated by Spearman correlation coefficients because of skewed distribution of insulin and proinsulin.

Hazard ratios (HRs) and 95% CIs were obtained from multivariate Cox proportional hazards models. HRs are calculated for one increase in interquartile range value, which is calculated for each variable. All models were adjusted for age and sex. We tested for possible interaction between insulin, proinsulin, HOMA-IR, and glucose tolerance status by adding two product terms to the survival models insulin × impaired glucose metabolism and insulin × diabetes. We also tested for possible interaction among insulin, proinsulin, HOMA-IR, and sex because this was found in some studies (2). In the second model, we additionally adjusted for glucose tolerance status as a potential confounding factor. Third, we also adjusted for insulin resistance, calculated as HOMA-IR, to investigate the effect of insulin and proinsulin levels on (CVD) mortality independent of HOMA-IR. To find other potential confounding or mediating factors, we added fasting and postload glucose, BMI, triglycerides, HDL cholesterol, total cholesterol, hypertension, and current smoking one by one to the third model. We adjusted our model for the variables that affected the estimated association. A P value <0.05 was considered statistically significant. Testing for interaction, we considered a P value <0.10 statistically significant. Statistical analyses were performed with the SPSS software package for Windows (version 10.1.4).

The characteristics of the study cohort, stratified for glucose tolerance status, are shown in Table 1. Of the 604 participants, 156 died, of whom 59 died from CVD or sudden death during the 11 years of follow-up. There was an increasing risk of all-cause mortality over the glucose tolerance groups (P = 0.04). For CVD mortality, increase in risk over glucose tolerance groups was not statistically significant (P = 0.10). All other variables, except sex and total cholesterol, showed statistically significant differences (P <0.05) across glucose tolerance groups.

Spearman correlation coefficients among insulin, proinsulin, HOMA-IR, and HOMA-B are shown in Table 2. All shown variables were statistically significant and correlated with each other (P <0.01), except fasting proinsulin with HOMA-B and proinsulin-to-insulin ratio with postload insulin and with HOMA-IR. Proinsulin-to-insulin ratio was statistically significant but inversely related to fasting insulin and HOMA-B.

HRs for fasting insulin, fasting proinsulin, and HOMA-IR, adjusted for age and sex, were calculated for separate strata of glucose tolerance. The HRs for insulin and HOMA-IR were the highest in the normal glucose metabolism group (data not shown). Proinsulin was associated with mortality in all groups of glucose tolerance status. Since the tests for interactions among insulin, proinsulin, HOMA-IR, and glucose tolerance status were not statistically significant (P >0.10), these groups were combined for subsequent analyses.

In Table 3, HRs per increase in interquartile range, calculated from the present study population for both all-cause and CVD mortality, are shown. Fasting insulin, proinsulin-to-insulin ratio, postload insulin and proinsulin levels, and HOMA-B were not significantly associated with all-cause or CVD mortality. Fasting proinsulin, however, was significantly associated with all-cause mortality (HR 1.21 [95% CI 1.04–1.42]) and CVD mortality (1.33 [1.06–1.66]), adjusting for age and sex. After adjustment for glucose tolerance status and additionally for HOMA-IR, the HRs for proinsulin were only slightly attenuated and remained significant. HOMA-IR was associated with all-cause mortality (1.16 [1.01–1.32]) but not with CVD mortality. By including fasting proinsulin in the model of HOMA-IR, the association between HOMA-IR and all-cause mortality disappeared (1.00 [0.83–1.20]). Also, by including proinsulin in the model of insulin, HRs for insulin became smaller (Table 3), whereas HRs for proinsulin did not substantially change (data not shown).

In addition, we tested for other potential confounding risk factors (Table 3) in the relation between proinsulin and (CVD) mortality. Only triglycerides and hypertension slightly attenuated the association between proinsulin and all-cause mortality. Other risk factors, i.e., fasting and postload glucose, BMI, HDL cholesterol, total cholesterol, and HOMA-B, did not change the association between proinsulin and mortality (data not shown). The associations with mortality did not differ between men and women (data not shown).

In the present study, in a subsample of a population-based prospective study, we found that proinsulin was an independent risk factor for both all-cause and CVD mortality, also after adjusting for both insulin resistance and glucose tolerance status.

Fasting proinsulin per increase in interquartile range was associated with a 21% higher all-cause mortality and a 33% higher CVD mortality. The association of proinsulin with mortality risk appeared stronger than that of insulin with mortality risk. This has also been observed in another longitudinal study (9) and in cross-sectional studies (22,23). Moreover, in a model that included both proinsulin and insulin or HOMA-IR, proinsulin remained significantly associated with (CVD) mortality, whereas insulin or HOMA-IR did not remain statistically significant. Only one previous study considered insulin as a covariate (11). In that study, proinsulin remained an independent risk factor for stroke, and the excess risk of insulin disappeared when proinsulin was added to the model.

High proinsulin levels can be a result of insulin resistance or β-cell failure, and both may contribute to the association between proinsulin and CVD. Proinsulin has been shown to be strongly associated with insulin resistance in a recent study (24), which may at least partly explain the association with CVD and mortality (25,26). A strong correlation between proinsulin and HOMA-IR can cause multicollinearity when both variables are included in the same model. In the present study, the Spearman correlation coefficient between proinsulin and HOMA-IR was 0.53, substantially lower than the 0.90 cutoff value, which is often used as a rule of thumb to define multicollinearity. We found that the association between proinsulin and (CVD) mortality was independent of insulin resistance. In contrast, the association between HOMA-IR and CVD mortality disappeared following adjustment for proinsulin. Therefore, high proinsulin levels might be an intermediate factor in the association between insulin resistance and CVD risk, in which high proinsulin levels are more closely related to CVD than insulin resistance.

High proinsulin levels may also be attributed to β-cell dysfunction or an increased demand on the β-cells (14). In the current study, disproportionately elevated proinsulin-to-insulin ratios or HOMA-B (16) were not associated with an increased risk of CVD mortality, suggesting that proinsulin per se may contribute to the atherogenic or thrombogenic process. Indeed, a clinical study in which human proinsulin was administered for at least 1 year was stopped when six patients treated with human proinsulin had adverse cardiovascular events (13), whereas subjects treated with insulin had no adverse events.

In the IRAS (Insulin Resistance Atherosclerosis Study), plasminogen activator inhibitor-1 (PAI-1) was suggested to play a role in the association between proinsulin and CVD (22). Increased PAI-1 activity promotes both fibrosis and thrombosis and is associated with CVD (27). It was already shown that PAI-1 activity increased after insulin or proinsulin was administered in vitro (28) and in vivo (29). Unfortunately, in the Hoorn Study, data on PAI-1 are not available. PAI-1 levels are elevated in subjects with insulin resistance, impaired glucose metabolism, and the metabolic syndrome (27). Therefore, the metabolic syndrome, which is defined by high triglycerides, impaired glucose metabolism, hypertension, abdominal obesity, and/or low HDL (30), is another potential confounding factor in the relation between proinsulin and CVD risk. However, in our data, triglyceride concentration, glucose tolerance status, and hypertension explained only a small part of the association between proinsulin and CVD risk.

Hyperinsulinemia has been shown to predict all-cause (1) and CVD mortality (2,4,5,8) in the general population. We did find an association between insulin and CVD mortality, similar to the findings in the meta-analysis of the DECODE (Diabetes Epidemiology: Collaborative Analysis of Diagnostic Criteria in Europe) Insulin Study Group. However, in the present study, the association between insulin and CVD mortality was not significant, probably because of the relatively small number of case subjects.

In a number of studies, the association between insulin and coronary heart disease was J-shaped (3) or U-shaped (31). In the present subpopulation of the Hoorn Study, the number of cases was too small to explore this association. When we analyzed the entire Hoorn Study population, baseline insulin levels were available for 1,806 participants with normal glucose metabolism. Indeed, the HRs for fasting-specific insulin of the lowest quintile against the middle three quintiles, adjusted for age and sex, were 1.10 (95% CI 0.81–1.50) for all-cause mortality and 1.07 (0.65–1.75) for CVD mortality (unpublished data). For the present analysis, we did not consider this a relevant deviation of a linear association.

In some studies, postload insulin levels, 2 h after a 75-g glucose load, have been associated with coronary artery disease (7,8). In the present study, postload insulin levels were not significantly associated with mortality, and the association was weaker than between fasting insulin and mortality. These results are consistent with findings from a recent meta-analysis (2). The weaker association between postload insulin and mortality was explained by the larger biological variation of postload insulin levels. In our study, biological variation was reduced by measuring insulin levels in duplicate, but this could still play a role in attenuating the association between postload insulin and mortality.

We did not use gold standard techniques for measuring insulin sensitivity (i.e., the hyperinsulinemic-euglycemic clamp) and for β-cell function (i.e., the hyperglycemic clamp) (32). Instead, HOMA-IR was used as an estimate of insulin resistance (16). In previous studies, HOMA-IR was found to be a predictor of CVD in diabetic patients (25,33). However, HOMA-IR cannot be used to estimate insulin resistance in diabetic patients (34). Therefore, patients with known diabetes were excluded from the present study. Our observation that insulin levels and HOMA-IR were the strongest predictors of all-cause and CVD mortality in subjects with a normal glucose tolerance status might be explained by the fact that insulin and HOMA-IR are less accurate indexes of insulin resistance in patients with impaired glucose metabolism or newly detected type 2 diabetes.

A strength of our study is that levels of insulin, proinsulin, and glucose were measured twice to improve the precision. We included participants with newly diagnosed diabetes and with impaired glucose metabolism to enlarge the range of distribution of insulin, proinsulin, HOMA-IR, and HOMA-B. Also, these subjects are especially important to include because of their enhanced CVD risk.

In our study, proinsulin concentration was associated with both all-cause and CVD mortality, independent of glucose tolerance status and insulin resistance. In addition, we showed that the association between HOMA-IR and all-cause mortality disappeared when proinsulin was entered into the model. High proinsulin levels might therefore play a role in the atherothrombotic process, possibly as an intermediate in the association between insulin resistance and CVD risk.

M.A. was supported by the Dutch Diabetes Research Foundation (DFN grant no. 2001.00.052).

We gratefully acknowledge Corry Popp-Snijders for performing analyses of insulin-like molecules and Manon van Hecke for her work on the follow-up of the cohort.