Determinants of New-Onset Diabetes Among 19,257 Hypertensive Patients Randomized in the Anglo-Scandinavian Cardiac Outcomes Trial–Blood Pressure Lowering Arm and the Relative Influence of Antihypertensive Medication

  1. Ajay K. Gupta, MD1,
  2. Bjorn Dahlof, MD2,
  3. Joanna Dobson, MSC1,
  4. Peter S. Sever, FRCP1,
  5. Hans Wedel, PHD3,
  6. N.R. Poulter, FRCP1 and
  7. on behalf of the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT) Investigators
  1. 1International Centre for Circulatory Health, National Heart and Lung Institute, Imperial College, London, U.K.
  2. 2Sahlgrenska University Hospital/Ostra, Goteborg, Sweden
  3. 3Nordic School of Public Health, Goteborg, Sweden
  1. Corresponding author: Prof. Neil R. Poulter FRCP, ICCH Building, 59-61 N. Wharf Rd., London W2 1PG, U.K. E-mail: n.poulter{at}


OBJECTIVE—The purpose of this study was to determine the baseline predictors of new-onset diabetes (NOD) in hypertensive patients and to develop a risk score to identify those at high risk of NOD.

RESEARCH DESIGN AND METHODS—Among 19,257 hypertensive patients in the Anglo-Scandinavian Cardiac Outcomes Trial–Blood Pressure Lowering Arm (ASCOT-BPLA) who were randomly assigned to receive one of two antihypertensive regimens (atenolol ± thiazide or amlodipine ± perindopril), 14,120 were at risk of developing diabetes at baseline. Of these, 1,366 (9.7%) subsequently developed NOD during median follow-up of 5.5 years. A multivariate Cox model was developed to identify the independent predictors of NOD and individual risk scores.

RESULTS—NOD was significantly associated with an increase in baseline fasting plasma glucose (FPG), BMI, serum triglycerides, and systolic blood pressure. In contrast, amlodipine ± perindopril in comparison with atenolol ± thiazide treatment (hazard ratio 0.66 [95% CI 0.59–0.74]), high HDL cholesterol, alcohol use, and age >55 years were found to be significantly protective factors. FPG was the most powerful predictor with risk increasing by 5.8 times (95% CI 5.23–6.43) for each millimole per liter rise >5 mmol/l. The risk of NOD increased steadily with increasing quartile of risk score, with a 19-fold increase (95% CI 14.3–25.4) among those in the highest compared with those in the lowest quartile. The model showed excellent internal validity and discriminative ability.

CONCLUSIONS—Baseline FPG >5 mmol/l, BMI, and use of an atenolol ± diuretic regimen were among the major determinants of NOD in hypertensive patients. The model developed from these data allows accurate prediction of NOD among hypertensive subjects.

Observational data suggest that hypertension is a risk factor for type 2 diabetes (1); hence, the two conditions frequently coexist. The increased propensity of the hypertensive population to develop diabetes is variably affected by different classes of antihypertensive medication. Recently, results of a network meta-analysis, using data from 22 clinical trials comprising 143,153 participants who did not have diabetes at randomization, suggested that the association between antihypertensive agents and incident diabetes is lowest for angiotensinogen receptor blockers and ACE inhibitors followed by calcium channel blockers and placebo, with β-blockers and diuretics increasing risk (2). The diabetogenicity of β-blockers and diuretics is consistent with their adverse impact on blood glucose levels, which has been reported for several decades (35). In contrast with the adverse effects of diuretics (3,6,7) and β-blockers (8) on the incidence of new-onset diabetes (NOD) in randomized trials, the bulk of trial evidence suggests that drugs that block the renin-angiotensin system exert a protective role against the development of NOD (2,9,10). These differential effects have influenced recommendations for antihypertensive drug sequencing contained in British guidelines (11 12), whereas the extensive use of β-blockers and diuretics, often in combination, continues worldwide, in part because of controversy regarding whether there is any cardiovascular toll associated with antihypertensive agent–associated incident diabetes (1316).This controversy notwithstanding, little is known about the other baseline predictors of NOD in hypertensive populations and the importance of antihypertensive therapy relative to these variables.

Because NOD was a predefined tertiary end point of the Blood Pressure-Lowering Arm of the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT-BPLA) (17,18), its database provides an excellent opportunity to evaluate baseline predictors for development of NOD among a large hypertensive population.


Details of the methods used and the main results of the ASCOT-BPLA trial have been described previously (17,18). In brief, 19,257 hypertensive patients, aged 40–79 years, with ≥3 cardiovascular risk factors but without any previous coronary heart disease were randomly assigned using a prospective randomized open blinded end points design to receive one of two antihypertensive regimens: atenolol with the addition of a thiazide diuretic as required (atenolol-based regimen) or amlodipine with the addition of perindopril as required (amlodipine-based regimen) to reach blood pressure targets. Subsequently, these patients were followed up with fasting blood samples obtained at 6 months, 12 months, and annually thereafter.


For the purposes of these analyses, patients were considered to potentially have “diabetes at baseline” if they satisfied any one of the following three sets of criteria:

  1. Fasting plasma glucose (FPG) ≥7mmol/l or random glucose ≥11.1 mmol/l at randomization or screening visits.

  2. Self-reported history of diabetes and receiving drug or dietary therapy for diabetes.

  3. Presence of both impaired fasting glucose (>6 and <7 mmol/l) and glycosuria at randomization or screening visits. This criterion was chosen to avoid misclassification of such people as nondiabetic at baseline when full investigation with an oral glucose tolerance test may have revealed diabetes.

Individuals included in one or more of these three groups were therefore excluded from the population considered to be at risk of developing NOD in these analyses. Development of diabetes (NOD) during follow-up was determined on the basis of the 1999 World Health Organization criteria (19).

Statistical methods

STATA 9 software was used for all statistical analyses. The impact on the development of NOD of demographic, clinical, and laboratory data at baseline was assessed using the Cox proportional hazards model. Approximately 10% of subjects at baseline had nonfasting values of either triglycerides or glucose or both and were excluded for the purpose of our main analysis.

Baseline characteristics among those who developed NOD were compared with characteristics of those who did not. For each of the baseline characteristics, a univariate Cox model was used to estimate the hazard (risk) ratio (HR) and 95% CI for development of NOD.

Multivariate Cox regression models were developed using forward stepwise selection (P < 0.05 for inclusion) with age, sex, and randomized blood pressure treatment group as prespecified covariates in all models. All baseline variables were considered for inclusion in the multivariate model. Three continuous variables, viz., age, FPG, and BMI showed some evidence of nonlinearity at extreme values. However, these were retained as continuous variables in the model with appropriate cutoff values to account for nonlinearity. Three multivariate Cox models were built: model 1 including all 12,692 patients with known values (1,212 cases of NOD); model 2 including patients with known values who were randomly assigned to the atenolol-based treatment group (n = 6,321, cases = 705); and model 3 including patients with known values who were randomly assigned to the amlodipine-based group (n = 6,371, cases = 507) .

Model 1 was taken forward as the primary model to develop a risk score and to test any prespecified interactions between the treatment groups and other variables. The risk score for each patient was calculated from the primary model by summing the products of the coefficients derived from the primary model and the actual values of the variables in the model. The distribution of risk scores was then divided into quartiles of increasing risk, and calibration of the model was evaluated by comparison of the plots of the actual and predicted outcomes. Bootstrap resampling (100 repetitions) was used to assess the internal validity of the primary model.


Of 19,257 hypertensive patients randomly assigned to ASCOT-BPLA, 14,120 were considered to be at risk of developing NOD at baseline (Fig. 1). Of these, 1,366 subsequently developed NOD during an accumulated follow-up of 73,425 years (median follow-up 5.5 years; incidence rate 18.6 per 1,000 patient-years).

Baseline characteristics

Baseline characteristics in the at-risk population were well matched among those randomly assigned to the two blood pressure–lowering regimens (Table 1). In each of the two treatment groups those who developed diabetes were also much more likely to be younger with higher BMI, FPG, pulse rate, diastolic blood pressure, and serum triglyceride levels and lower HDL cholesterol levels compared with those who remained nondiabetic. However, some differences were apparent between those who did and did not develop diabetes in each of the two blood pressure–lowering treatment groups, e.g., prevalence of current smoking.

Risk factors for the development of diabetes

On univariate analysis (online appendix Table A [available at]), patients assigned to the amlodipine-based regimen were 31% less likely to develop NOD than those assigned to the atenolol-based regimen (HR 0.69 [95% CI 0.62–0.77]), and for each unit rise in HDL cholesterol or total cholesterol the risk of NOD fell by 61 and 8%, respectively. In contrast, for each 5-unit rise in BMI or 10 mmHg rise in baseline systolic blood pressure (SBP) the risk of NOD increased by 42 and 6%, respectively. The presence of microalbuminuria, >3 cardiovascular risk factors, and higher serum triglyceride levels, diastolic blood pressure, and heart rate at baseline were among other notable and significant risk factors for NOD. However, the largest impact on risk of NOD was that induced by FPG level, which was associated with a >5-fold increase in risk for each 1 mmol/l increase.

On multivariable analysis (model 1; n = 12,692), higher levels of FPG, BMI, serum triglyceride, SBP, and concomitant use of noncardiovascular medications were found to be significant risk factors for NOD at baseline. In contrast, amlodipine-based treatment (HR 0.66 [95% CI 0.59–0.74]), higher total and HDL cholesterol levels, alcohol use, and age >55 were found to be significantly protective factors. FPG was again the most powerful predictor with risk increasing by 5.8 times (95% CI 5.23–6.43) for each 1 mmol/l rise >5 mmol/l. Risk increased by 49 and 12% for each 5-unit increase in BMI (up to 35 kg/m2) and 1 mmol/l increase in serum triglyceride levels, respectively, whereas randomization to amlodipine-based treatment and an increase in baseline HDL by 1 mmol/l reduced the risk by 34 and 28%, respectively (Table 2).

On multivariable analysis based on treatment allocation, the predictors for NOD among those randomly assigned to atenolol-based (model 2) and amlodipine-based (model 3) treatment groups were essentially similar to those of the primary model; however, some differences were apparent (online appendix Table B). For example, although FPG, BMI, total cholesterol, SBP, and age were significant predictors in both blood pressure treatment groups, a raised serum triglyceride level was a major predictor only among those randomly assigned to the atenolol-based regimen (HR 1.24 [95% CI 1.17–1.32]). Conversely, raised HDL cholesterol (0.55 [0.51–0.75]), alcohol intake, baseline heart rate, and smoking were significant predictors only among those randomly assigned to the amlodipine-based regimen. However, when these differences between the two treatment groups were evaluated in the primary Cox model, there was no strong evidence of a significant interaction between allocated drug treatment and baseline triglyceride (P = 0.09, after excluding an outlier), smoking (P = 0.09), HDL cholesterol (P = 0.75), alcohol intake (P = 0.25), and baseline heart rate (P = 0.13). Of note, among these potential interactions only that between treatment allocation and serum triglyceride was prespecified in the statistical analysis plan.

Risk scores

The β-coefficients, z scores, and P values of each of the baseline variables used in the risk score are shown in Table 2. Larger values of the z score (irrespective of the sign) indicate the strength of the variables as a predictor. Figure 2A illustrates the increasing risk of NOD with an increase in risk quartile using Kaplan-Meier plots. Compared with the lowest risk quartile, patients in the highest quartile had a 19-fold increase in risk of NOD (HR 19.04 [95% CI 14.27–25.41]). There was no evidence of an interaction between risk quartile and antihypertensive treatment group when the risk score was calculated without allocated treatment. Figure 2B shows that allocation to the amlodipine-based regimen reduced the risk to the same extent in each risk quartile. The model had excellent internal validity and reasonably strong discriminative ability (Harrell's c index of 0.80) (appendix Figure A).


These analyses of baseline measures among >14,000 hypertensive patients considered to be free of diabetes at the start of the ASCOT-BPLA trial (17) indicate that randomization to antihypertensive treatment, low HDL cholesterol, and raised BMI, serum triglycerides, SBP, and particularly FPG are important determinants of NOD. The relative importance of each of the determinants of incident diabetes is implied by an increase in z score regardless of its sign (Table 2). The risk model thus developed allows the accurate prediction of NOD over a 5-year period for an individual.

The >5-fold increase in risk of NOD for each 1 mmol/l rise in FPG reported in this article is larger than that observed in most (810) but not all (20) earlier reports. In contrast with some earlier studies (8), the exclusive use of fasting glucose and unambiguous, robust definitions may have contributed to the large effect size observed. The putative effects of FPG were linear and apparent from 5 mmol/l onward, a threshold for incremental risk that has previously been identified (21). The risk attenuated progressively through the trial with the effect reducing from a HR of 9.72 (95% CI 8.06–11.72) during the first year to 1.88 (95% CI 1.25–2.83) after ≥5 years of follow-up. This trend may reflect the attrition of subjects susceptible to development of NOD.

These results are consistent with most other analyses of trials using antihypertensive agents in finding that a regimen based on a calcium channel blocker to which an ACE inhibitor was added was associated with significantly less NOD than a regimen based on a β-blocker to which a diuretic was usually added (39). Indeed, the randomization to amlodipine-based regimen emerged as the strongest protective factor of the variables evaluated. The finding that the differential risk of NOD between the two antihypertensive regimens remained the same irrespective of baseline risk (Fig. 2B), contrasts with results in the Captopril Prevention Project (CAPPP) Trial (10) but is in keeping with findings in the Losartan Intervention For Endpoint (LIFE) trial (8).

Increasing age was an independent protective factor for the development of diabetes that contrasts with some (10) but not all trial results (22) and is consistent with several observational studies (23,24). These studies have shown that although the prevalence continues to increase with age, the incidence of diabetes plateaus in elderly individuals.

Our study is consistent with several other observational studies in finding alcohol intake to be protective (25,26) and increased triglycerides to be a putative risk factor for NOD (27,28). Somewhat counterintuitively, raised total cholesterol appeared to be protective in these analyses, although this finding too has been reported in other trials (810). The increased risk associated with concomitant use of ≥1 noncardiovascular medications, including some that are known to be diabetogenic, may reflect or be a marker of chronic ill health.

The performance of several previous analyses relating to NOD has been subject to methodological criticism (9,29), such as being post hoc analyses, using different definitions of NOD, and using nonfasting and/or whole blood glucose values, but most of these criticisms do not apply to the current study design and analyses. This study demonstrates the relative importance of antihypertensive medications, after FPG and BMI, and suggests that their judicious use will benefit all regardless of risk category. These analyses allowed the development of a relatively simple risk score for predicting NOD. This score appears to have excellent internal validity and pending further external validation, could be potentially useful in routine clinical practice to guide not just prescribing of antihypertensive medication but other interventions aimed to prevent NOD.

Given evidence from previous trials (2,9,30), it seems likely that the differential effect of the two antihypertensive regimens used in ASCOT-BPLA on NOD is a composite of the adverse effects on risk produced by atenolol and thiazide, plus the protective effects of perindopril, with amlodipine probably playing a neutral role. However, recent analyses of the Diabetes Reduction Assessment with Ramipril and Rosiglitazone Medication (DREAM) trial (31) in contrast to the Heart Outcomes Prevention Evaluation (HOPE) trial (32) did not show a significant protective effect of ramipril against NOD. Nevertheless, 2-h glucose levels were significantly improved among those taking ramipril in the DREAM trial, which, allied with the 9% nonsignificant reduction in NOD, suggests that the apparently “negative” findings in the DREAM trial may reflect inadequate power to detect an effect in too short a time.

Although individuals studied in the ASCOT trial were more representative of the general hypertensive population than those in several other recent trials (68,20,30,32), the population was largely Caucasian and male and from the U.K., Ireland, and Nordic countries. Whether and to what extent the findings relate to other ethnic groups requires evaluation in other studies. Furthermore, given the large sample size and hence power of these analyses, the clinical relevance of some of the less significant relationships needs to be considered when the results are interpreted.

Further analyses evaluating the effects of changes in baseline variables (e.g., body weight, blood pressure, and others) throughout the trial are in progress, as are analyses evaluating whether worsening dysglycemia and NOD are associated with worsening cardiovascular outcomes. Although these analyses may inform policy decisions on prescribing for hypertensive patients, the limited power of such analyses with relatively limited follow-up hampers appropriate interpretation of such data and possibly explains earlier conflicting results (1416).

In summary, the present analyses provide robust evidence that treating hypertensive patients with a regimen based on amlodipine and perindopril compared with a regimen based on atenolol and a thiazide diuretic significantly reduces the risk of NOD such that the number needed to treat 30 patients for just >5 years is required to prevent 1 case of NOD (95% CI 23–42). They further describe a robust, discriminative model, which helps to determine accurately the risk of NOD in hypertensive patients and highlights the relative importance of various other independent predictors such as FPG, BMI, SBP, serum HDL cholesterol, and triglycerides in the development of NOD. Pending further definitive evidence related to cardiovascular morbidity and mortality with antihypertensive-associated incident diabetes, it seems at best unwise, except where compelling indications apply, to use β-blockers and diuretics in combination in preference to other combinations such as a calcium channel blocker plus an ACE inhibitor, particularly because the latter agents have been shown to be more cost-effective (11).

Figure 1—

NOD development according to the randomized treatment group in ASCOT-BPLA Trial.

Figure 2—

A: Kaplan-Meier graphs of incidence of NOD stratified by quartiles of risk score (*) with the uppermost quartile divided equally into two as 4a and 4b (cutoff at 5-year follow-up). *HRs (95% CI) for each of the risk quartile with the first quartile as the reference group: second quartile 2.5 (1.8–3.5); third quartile 5.0 (3.7–6.9); and fourth quartile 19.0 (14.3–25.4) [4b quartile 9.72 (7.14–13.25) and 4a quartile 30.31 (22.64–40.57)]. Corresponding risk scores for each of the quartile groups are as follows: first quartile <10.26; second quartile 10.26–10.85; third quartile 10.86–11.62; and fourth quartile ≥11.63 (4b quartile 11.63–12.29 and 4a quartile ≥12.30). B: Kaplan-Meier graph of incidence of NOD stratified by quartile of risk score and the blood pressure–lowering treatment (cutoff at 5-year follow-up). ——, atenolol-based treatment; –––, amlodipine-based treatment.

Table 1—

Baseline characteristics in the at-risk population by treatment group and development of NOD

Table 2—

Primary multivariate Cox proportional hazard regression model for the development of NOD (model 1)


The ASCOT trial and analyses have received funding from Pfizer Inc.

Parts of this study were presented in abstract form at the World Congress of Cardiology 2006, Barcelona, Spain, 2–5 September 2006 and the 21st Scientific Meeting of the International Society of Hypertension, Fukuoka, Japan, 15–19 October 2006.


  • Published ahead of print at on 11 February 2008. DOI: 10.2337/dc07-1768.

    Additional information for this article can be found in an online appendix at

    A.K.G. has received support for travel to meetings from Pfizer. B.D., P.S., H.W., and N.P. have received travel expenses, payment for speaking at meetings, and funding for research from Pfizer and Servier to cover administrative and staffing costs of the ASCOT trial and travel, and from Pfizer to cover costs related to the present analyses.

    The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

    • Accepted January 18, 2008.
    • Received September 7, 2007.


| Table of Contents

This Article

  1. Diabetes Care vol. 31 no. 5 982-988
  1. Online-Only Appendix
  2. All Versions of this Article:
    1. dc07-1768v1
    2. dc07-1768v2
    3. 31/5/982 most recent