Skip to main content
Skip main navigationClose Drawer MenuOpen Drawer Menu

Analysis of Large Electronic Health Record Databases Supports Blood Pressure–Incident Diabetes AssociationFree Access

Original Investigation

J Am Coll Cardiol, 66 (14) 1563–1565


Worldwide, an estimated 1.5 million deaths were caused by diabetes mellitus in 2012 (1), up from 1 million in 2000. Diabetes has been shown to be associated with a 2-fold increased risk of all-cause mortality and a 3-fold increased risk of cardiovascular disease mortality relative to age- and sex-matched controls in a contemporary primary care population in the United Kingdom (2). Blood pressure (BP) and diabetes share common mediators, including inflammation and endothelial dysfunction (3). Despite the known relationships between BP and diabetes, evidence supporting a clear population-level association between increased BP and the development of new-onset diabetes has been mixed. Large national databases of electronic medical records and meta-analytical techniques represent 2 possible routes to elucidate the strength of an association, if not a causal relationship, between BP and incident diabetes.

In this issue of the Journal, Emdin et al. (4) evaluate the association between BP and the risk of incident diabetes in a cohort of 4.1 million adults. Making use of electronic medical records, this study included data from patients 30 to 90 years of age without diabetes; exclusion criteria included prevalent renal disease, peripheral vascular disease, heart failure, cerebrovascular disease, ischemic heart disease, and type 1 or type 2 diabetes. Data were collected from the U.K. Clinical Practice Research Datalink, which covers about 9% of the U.K. population, and the Hospital Episode Statistics database, which includes data on hospital admissions, outpatient appointments, and cause-specific mortality for hospital events. BP measurements made between January 1990 and January 2013 were adjusted for regression-dilution bias (a downward bias of the association regression slope resulting from measurement error and unaccounted for variability in BP [5]) by regressing serial BP measurements on a baseline measurement. Hazard ratios for measured BP (adjusted for age, sex, and body mass index [BMI], smoking status within 2 years of baseline, and baseline antihypertensive and lipid-lowering drug use) were multiplied by regression dilution ratios of 2.1 and 2.5 for systolic BP (SBP) and diastolic BP (DBP), respectively, to estimate incident diabetes associations with “usual” BP.

The primary outcome was defined as a diagnosis of type 2 or unspecified diabetes or issuance of a prescription for insulin or an antidiabetic drug. Participants were censored at first occurrence of the primary outcome. A series of 6 sensitivity analyses were performed to evaluate impact of various covariate adjustments including cholesterol, lipid, and hypertension treatments, cohort effects (adjusting for timing of initial BP measurement), diabetes definition (as listed previously vs. excluding those with unspecified diabetes), and timing of events (events in all years vs. excluding events in years 1, 2, and 4).

The authors also conducted a random-effects meta-analysis of prospective observational studies of BP and incident diabetes. This analysis included findings from 30 studies enlisting 285,664 participants with 17,388 incident diabetes events. Regression dilution ratios generated in the main study were applied in the meta-analysis.

In the primary analysis, about 60% of the participants were women, median age was 46 years, median BMI was 25.7 kg/m2, and median follow-up was 6.8 years. A total of 186,698 participants met the definition of the primary outcome. Emdin et al. (4) reported that, overall, a 20 mm Hg increase in SBP and a 10 mm Hg increase in DBP were associated with a 58% (95% confidence interval [CI]: 1.56 to 1.59) and 52% (95% CI: 1.51 to 1.54) increase in the risk of new diabetes, respectively. For SBP, the association was strongest for those with normal or slightly elevated usual pressures with a flattening of the curve below and above the SBP extremes (about 110 and 150 mm Hg, respectively). Diastolic BP was similarly associated with the onset of diabetes but with less evidence for plateaus toward the extremes (70 and 100 mm Hg). The associations were modified by sex (stronger associations observed in men), but not at clinically meaningful differences. Baseline BMI and age more strongly modified the association: lower BMI and younger age exhibited greater relative risk increases with increasing BP (although because of differences in baseline risk, absolute risk increases in the higher BMI and older strata were equal to or higher than increases in absolute risk in lower BMI and younger strata). For example, for those with BMI <25 kg/m2, the hazard ratio for a 20 mm Hg increase in SBP was 1.89 (95% CI: 1.84 to 1.94); for those with BMI >35 kg/m2, the hazard ratio was 1.19 (95% CI: 1.16 to 1.22). In analyses in which subjects on antihypertensive or lipid-lowering drugs were excluded, risk estimates were similar to those made in the full cohort. In the meta-analysis, the pooled relative risk for each 20 mm Hg higher usual SBP was 1.77 (95% CI: 1.53 to 2.06).

These findings support a longstanding hypothesis (6) regarding the association between BP and the risk of diabetes. This well-designed study illustrates the utility of carefully collected and appropriately analyzed electronic medical records from practice databases. The consistency of findings between the 2 approaches also adds value to the association findings.

Several elements of the Emdin et al. (4) study are worthy of further consideration. For example, the regression dilution ratios of 2.1 (SBP) and 2.5 (DBP), used to derive usual BP in this study, are considerably higher than those in the range of 0.40 to 0.56 employed in a number of large cohort studies in Europe and the United States (7). These differences were not explored, and, if the ratios used in Emdin et al. (4) are elevated, this overcorrection would have inflated hazard estimates of diabetes with increased BP in both the primary study and the meta-analysis.

Emdin et al. (4) reported that the prevalence of antihypertensive drug use during follow-up was considerably higher among those with higher baseline usual BP: 52.2% among those with SBP >136 mm Hg versus 13.0% among those with SBP <127 mm Hg at baseline. The authors, however, made no adjustment for the type of antihypertensive treatment. It is well-documented that thiazide diuretics are associated with a greater incidence of diabetes. For example, in the ALLHAT (Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial) participants without metabolic syndrome, diabetes occurred in 7.7% in the chlorthalidone treatment arm versus 4.2% and 4.7% in the amlodipine and lisinopril arms, respectively (8). Although the sensitivity analyses done by these authors temper concern over confounding by type of antihypertensive treatment, this complication of the clinical picture is worthy of further exploration.

Although Emdin et al. (4) findings may bolster evidence for an existing hypothesis, they are unlikely to motivate any changes in clinical practice. Association studies simply identify clinical correlates; they do not provide insights into the mechanisms that could contribute to the observed findings. Although the authors postulate that increased inflammation is the culprit, other metabolic or biochemical mediators may be the driving force leading to diabetes. Nonetheless, these investigators offer an exceptionally rigorous evaluation of the relation between BP and incident diabetes, including well-powered analyses of subgroups. Their methods make superb use of large clinical databases and their complementary meta-analysis provides further support of the association. This study provides a strong rationale for continued research into the biological basis and pharmacological implications of the observed association.

  • 1. World Health Organization : Global Health Estimates: Deaths by Cause, Age, Sex, and Country, 2000-2012 . Geneva, Switzerland: World Health Organization2014.

    Google Scholar
  • 2. Taylor K.S., Heneghan C.J., Farmer al. : "All-cause and cardiovascular mortality in middle-aged people with type 2 diabetes compared with people without diabetes in a large U.K. primary care database". Diabetes Care 2013; 36: 2366.

    CrossrefMedlineGoogle Scholar
  • 3. Reaven G.M., Lithell H. and Landsberg L. : "Hypertension and associated metabolic abnormalities—the role of insulin resistance and the sympathoadrenal system". N Engl J Med 1996; 334: 374.

    CrossrefMedlineGoogle Scholar
  • 4. Emdin C.A., Anderson S.G., Woodward M. and Rahimi K. : "Usual blood pressure and risk of new-onset diabetes: evidence from 4.1 million adults and a meta-analysis of prospective studies". J Am Coll Cardiol 2015; 66: 1552.

    View ArticleGoogle Scholar
  • 5. MacMahon S., Peto R., Cutler al. : "Blood pressure, stroke, and coronary heart disease. Part 1, Prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias". Lancet 1990; 335: 765.

    CrossrefMedlineGoogle Scholar
  • 6. Freedman P., Moulton R. and Spencer A.G. : "Hypertension and diabetes mellitus". Q J Med 1958; 27: 293.

    MedlineGoogle Scholar
  • 7. Howard S.C. and Rothwell P.M. : "Regression dilution of systolic and diastolic blood pressure in patients with established cerebrovascular disease". J Clin Epidemiol 2003; 56: 1084.

    CrossrefMedlineGoogle Scholar
  • 8. Black H.R., Davis B., Barzilay al. : "Metabolic and clinical outcomes in nondiabetic individuals with the metabolic syndrome assigned to chlorthalidone, amlodipine, or lisinopril as initial treatment for hypertension: a report from the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT)". Diabetes Care 2008; 31: 353.

    MedlineGoogle Scholar


Dr. Arnett has reported that she has no relationships relevant to the contents of this paper to disclose.