In this prospective study of middle-aged men who were followed for 35 years or until death, we found that even moderately increased SBP predicted the subsequent development of diabetes. The association with SBP was graded, and persisted after adjustment for a variety of potential confounders, with the greatest risk of subsequent development of diabetes in the highest blood pressure group. The finding that blood pressure within the higher normal blood pressure range increases the risk of diabetes is consistent with the findings from earlier studies
[9–13], with the added information that the increased risk remains through an extended follow-up until old age. Our median follow-up was 28 years which is considerably longer than in previous studies which were 8.9
 or a mean follow-up of 8.3 ±1.0
 years. Findings for DBP were similar, albeit less pronounced, with only DBP above 90 mm Hg indicating a later risk of diabetes. This is consistent with some studies
 which did not find DBP to be a risk factor for incident diabetes whereas other studies did
Compared to other studies
[9–13] we have a low proportion of diabetes cases during the first 10 years and this is probably explained by the fact that diabetes is often managed in primary care during the first years after diagnosis. We have a detection delay in the study since we did not have access to primary care data and were able to identify diabetes only as a hospital discharge diagnosis.
We found an inverse relation between smoking and blood pressure, probably due to the fact that smokers weigh less, which has also been described elsewhere
. Body weight is a strong determinant for diabetes. The increased risk of diabetes in the higher blood pressure categories was strongly attenuated when adjusting for BMI. This is in coherence with what other studies have found
[9, 13] and indicates that BMI is the factor with the greatest influence on diabetes risk.
At baseline the participants were asked about antihypertensive medication and the analyses were adjusted for this. Data on new antihypertensive treatment during follow-up was not available. This could be a bias since some antihypertensive drugs are known to increase the risk of developing diabetes
. Other studies have indicated that the greatest increase in risk of later development of type 2 diabetes in a hypertensive patient is due to hypertension itself
 and that the increased risk of diabetes remains after adjusting for specific antihypertensive treatments
. Therefore it does not seem plausible that medication could explain more than part of the total effect.
The benefits of considering high-normal blood pressures as a predisease have been debated
. Whether this should be labeled predisease or not may not be the issue here, however, but what this and other studies
[10–13] show is that the risk of developing diabetes is already increased at blood pressure levels below the limits generally used in considering treatment for hypertension. The clinical impact of these findings should be further analysed in appropriate studies.
The pathophysiological mechanisms underlying the association between high blood pressure and type 2 diabetes are not yet completely identified. Hypertension has been shown to induce microvascular changes
 and these changes may result in a reduced capacity for insulin mediated glucose uptake in the tissue
. Impaired microcirculation in tissues may thus be the connecting link between hypertension and insulin resistance
. Microcirculation has also been shown to be impaired by obesity; microcirculation has therefore emerged as a potential common denominator for several risk factors for metabolic syndrome
[28, 29]. Hypertension is also associated with endothelial dysfunction
 and markers of endothelial dysfunction have in turn been demonstrated to precede the development of diabetes
. Another possible linking factor between blood pressure and diabetes is inflammation. Hypertension
 and diabetes
 are both known to be associated with an increase in inflammatory markers.
Strengths and limitations
The strengths of the present study include a large number of unselected participants from the general population, prospective longitudinal design, extended follow-up, and hence the identification of a large number of diabetes cases (956 cases or 13%) compared to other studies
[10–13]. Even so, there are also a number of limitations to be considered. Firstly, participants who reported no history of diabetes during the screening examination were considered to be free from diabetes but neither blood glucose analyses nor an oral glucose tolerance test were performed. Therefore some of the participants might have had undiagnosed diabetes at inclusion. However, the majority of diabetes cases were identified at least a decade after the screening, so it is unlikely that the few cases that might have been included could significantly affect our results. We have also performed an analysis in which all the diabetes cases that are identified during the first 7.5 years were excluded which did not alter our results. Secondly, diabetes was defined as a discharge from hospital with a primary or secondary diagnosis code or a death certificate diagnosis of diabetes as an underlying cause. Some of the participants who developed diabetes but did not visit hospital during the follow-up period might not have been identified. Even so, the majority of them did attend hospital at some point (94.2%) and most of them several times, particularly in the last few years of the follow-up. Additionally, in our research group we have carefully followed another cohort of men, the men born in 1913 in Gothenburg, for whom we also have data from primary health care centres as well as diagnosis data from the registers. After 30 years of follow-up 13% of these men had developed type 2 diabetes (unpublished data) which is a similar figure to the proportion of men who developed diabetes in the Primary Prevention Study. Therefore, we believe that most of the diabetes cases were actually detected in the current analyses. Thirdly, the older ICD versions (ICD 8 and ICD 9) do not distinguish between type 1 and type 2 diabetes; however, given the age of the population, few, if any cases would have been type 1 diabetes. Fourthly, a remaining issue could be whether people reacting with high blood pressure at the screening were more prone to develop diabetes. This is something we are unable to adjust for in the study and, as we see it, a separate issue that must be analysed in another study. Fifthly, information on important risk factors other than blood pressure may be considered to be somewhat crude; therefore our results might have been affected by residual confounding. For instance, we have only information about BMI and not hip-waist ratio. One study investigating the relationship between blood pressure and diabetes incidence
 had both BMI and hip-waist measurement and there was a very minor difference in diabetes risk when using BMI instead of hip-waist ratio. For physical activity, no figures for hours per week are available in our study, just classification into 3 groups (sedentary, moderate and active). In their study, Conen et al.
 had information regarding the number of hours per week of physical activity, and even after controlling for this, the relationship between blood pressure and diabetes remained. In our study no information on smoking duration is available, and nor do we have dietary information. Diet pattern has been shown to affect diabetes incidence in a previous study
 but to our knowledge there are no studies concerning the relationship between blood pressure and diabetes that have adjusted for dietary pattern. Nevertheless, it seems unlikely that information about smoking duration and dietary pattern would eliminate the significance in our findings. Moreover, the results from our sensitivity analysis showed that the problem of residual confounding is likely to be negligible. Finally, we have only baseline information on all of the covariates. Many of the covariates are time dependent and fluctuations occur. Therefore we might not have captured the total influence of the covariates on the association between blood pressure levels and diabetes incidence.