Investigators have measured and related soluble forms of cell adhesion molecules to vascular disease in a variety of study types, with intercellular adhesion molecule 1 (ICAM-1) the most widely studied. Early prospective studies showed significant associations between baseline circulating ICAM-1 levels and incident vascular events [14], but with larger studies and more rigorous adjustment for potential confounders, associations became far less marked [5]. More recently, circulating ICAM-1 levels have been shown to predict diabetes in middle-aged individuals [6, 7] but data in the elderly are lacking. We recently suggested that few, if any, risk factors predict both incident diabetes and cardiovascular disease (CVD) events to similar extents [8]. In this paper we present the first simultaneous assessment of the association of any adhesion molecule with risk of incident diabetes and incident CVD events. Our hypothesis was that elevations in ICAM-1 would predict diabetes more strongly than they would predict vascular events. This is an important question, since ICAM-1 is regularly discussed as a vascular-risk surrogate.


The methods of the Prospective Study of Pravastatin in the Elderly at Risk trial as an ancillary risk-association study have been reported [9]. Between 1997 and 1999, 5,804 men and women aged 70–82 years were recruited if they had either pre-existing vascular disease or raised risk of such disease because of risk factors. The institutional ethics review boards of all centres approved the protocol, and all participants gave written informed consent. The protocol was consistent with the Declaration of Helsinki. The primary outcome was definite or suspect death from CHD, non-fatal myocardial infarction and fatal or non-fatal stroke. New-onset diabetes was classified by self-reported history, or a fasting blood glucose concentration of ≥7.0 mmol/l or a blood glucose measurement of ≥11.1 mmol/l when fasting status was uncertain (baseline and follow-up glucose measurements were fasting in >95% of participants) as detailed before [9]. Participants with known (n = 623) and undiagnosed baseline diabetes (n = 137) were excluded from analyses. All CVD events were validated by a blinded endpoints committee.

Baseline ICAM-1 was measured by ELISA (R&D Systems, Abingdon, UK) on previously unthawed pre-randomisation plasma samples stored at −80°C. The method has an inter-assay CV of <7%. Samples were processed blinded to the identity of samples.

The distribution of ICAM-1 was positively skewed and a logarithmic transformation (loge) was used. Relationships between loge[ICAM-1] and baseline characteristics were assessed using Pearson’s correlation coefficient for continuous variables and two-sample Student’s t tests or ANOVA for categorical variables. The influence of ICAM-1 on the endpoints of interest (diabetes and CVD) was investigated using Cox proportional hazards models, adjusting for randomised treatment and classic and novel risk factors (age, sex, country, LDL- and HDL-cholesterol, triacylglycerol, BP, smoking, BMI, use of antihypertensives, history of vascular disease and C-reactive protein [CRP]), plus glucose for diabetes endpoints only or diabetes for CVD endpoints only. Results are reported as HRs with 95% CIs for change of 1 unit increases in loge[ICAM-1] and corresponding p values.


Baseline analyses in relation to CVD outcomes

ICAM-1 levels were available in 5,685 (97.9%) of the original 5,804 participants. Their baseline characteristics, by endpoint, are shown in Table 1.

Table 1 Comparison of baseline characteristics by CVD endpoint (combined endpoint) and by incident diabetes in all those without diabetes at baseline

Compared with those without follow-up CVD events, those developing a CVD endpoint were slightly older, and more likely to be male and to have a history of diabetes or vascular disease. Total cholesterol and HDL-cholesterol and diastolic BP were lower in those who had a primary event, but fasting glucose after excluding those with diabetes was not higher. Baseline ICAM-1 levels were also not higher in those who had a CVD event. Participants who developed diabetes had a higher fasting glucose, BMI and triacylglycerol and lower HDL-cholesterol, and were more likely to be hypertensive. Baseline ICAM-1 levels were significantly higher among those who developed diabetes.

Correlations of ICAM-1 with other risk variables

At baseline, ICAM-1 levels correlated positively but weakly with triacylglycerol (age-, sex-, smoking status-, BMI- and country-adjusted, r = 0.063, p < 0.0001) and negatively with total cholesterol and HDL- and LDL-cholesterol levels (adjusted r = −0.035, −0.082 and −0.031, respectively, all p ≤ 0.02). ICAM-1 also correlated with CRP (r = 0.133, p < 0.0001) and fasting glucose (r = 0.057, p < 0.0001). ICAM-1 levels were higher in smokers, and in those with known diabetes or a history of peripheral arterial disease (all p < 0.001).

Associations of elevated ICAM-1 with risk of incident vascular events and diabetes

There was also no significant interaction in the association of ICAM-1 with CVD events with statin allocation (data not shown). ICAM-1 was not significantly related to risk of either primary CVD endpoint or, taken separately, either coronary events or stroke events (Table 2). This observation held regardless of a history of vascular disease at baseline or not (p > 0.15). The same was also true in fully adjusted analyses.

Table 2 HRs (95% CIs) for a 1 unit increase in log[ICAM-1] vs risk of incident CVD events and incident diabetes

Among 4,945 participants without baseline diabetes, 292 developed new-onset diabetes (248 on the basis of serial glucose measurements, and the remaining 44 on the basis of new prescriptions for oral diabetes medications). In keeping with expectations, predictors for new-onset diabetes were fasting glucose (HR 3.90, p < 0.001 per mmol/l increases), BMI (HR 1.09, p < 0.001, per BMI unit) and triacylglycerol (HR 1.68, p = 0.0027 per unit log[triacylglycerol]). For ICAM-1 the HR in treatment-adjusted analyses was 1.60 (95% CI 1.14–2.25, p = 0.0065), with little attenuation of the association upon adjustment for traditional ‘CVD risk factors’ (HR 1.82, 95% CI 1.27–2.63, p = 0.0013) (Table 2). Further adjustment for fasting glucose did not attenuate the association (HR 1.85, 95% CI 1.27–2.70, p = 0.0012), nor did additional adjustment for log[CRP] (HR 1.84, 95% CI 1.26–2.69, p = 0.0015).

Finally, when we examined associations using 1 SD change in ICAM-1, the HR was 1.01 (95% CI 0.94–1.08) for CVD events and 1.22 (95% CI 1.08–1.39) for incident diabetes.


This study in the elderly adds additional strong support for an authentic association of elevated ICAM-1 with risk of incident diabetes, despite no association with incident CVD. This is, as far as we are aware, the first study to test simultaneously the association of baseline ICAM-1 with both incident diabetes and CVD events.

A meta-analysis has reported univariate, but not multivariate, associations of ICAM-1 with CHD [5]. In the present elderly population of men and women, even univariate associations with vascular events were absent, and this was true for participants with or without a history of vascular disease. This lack of association may reflect a general age-related weakening of risk factor associations with vascular events. Accordingly, it is even more remarkable that high ICAM-1 was significantly associated with diabetes risk, even after adjustment for traditional risk factors and fasting glucose and CRP. These findings suggest ICAM-1 levels, in as much as they may reflect endothelial activation, are more strongly related to diabetes risk in the elderly than they are to cardiovascular risk. These data extend existing data in middle-aged individuals, since, in general, reported associations of circulating ICAM-1 with diabetes risk [3, 4] appear to be of greater magnitude than their associations with CVD risk [5].

Why ICAM-1 should better predict incident diabetes is unknown. Speculatively, since there is much more microvascular than large artery endothelium, one could assume that concentrations of ICAM-1 are determined more by microcirculatory function. This is relevant since endothelial perturbations in small vessels influence nutrient delivery and thus glucose metabolism. Alternatively, since ICAM-1 levels are not specific for endothelial activation, perturbances in other pathways may account for associations of elevated ICAM-1 with diabetes. The search for responsible mechanisms is important and may yield novel therapeutic mechanisms to treat or prevent diabetes.

Strengths and limitations

The endpoint numbers in this study are large by most standards. The study was conducted in the context of a randomised trial of statins [9]; however, baseline measures were taken before randomisation, and there was no significant interaction by randomised treatment. The study had an average 3.2 years of follow-up, although short follow-up may be appropriate in the more elderly because of the high event risk, and others have shown no difference in the association of ICAM-1 with vascular events by longer follow-up [4]. We acknowledge that the lack of OGTTs means that some individuals may have developed type 2 diabetes not detectable by changes in fasting glucose alone. However, lack of OGTTs would mean missed cases of diabetes in controls, but assuming similar baseline elevations in ICAM-1 in this latter group, then this would have biased the study against a link between ICAM-1 and incident diabetes. Moreover, the strongest predictors for incident diabetes in this population (fasting glucose, BMI and triacylglycerol) concur with expectations. Finally, whilst we accept that since many risk variables lessen in their association with vascular events in the elderly, most events, both diabetes and CVD events, occur increasingly in these populations, so the results are of interest.


We have shown for the first time that elevated ICAM-1 levels are associated with risk of diabetes but not CVD events (CHD or stroke combined or separately) in the elderly. We propose that endothelial dysfunction, at least as ‘estimated’ by elevated circulating ICAM-1 levels may be aligned more to diabetes than to vascular risk. The aetiological mechanisms underpinning these data require future investigation.