• Open Access

Quantifying the duration of pre-diabetes


Correspondence to:
Melanie Y. Bertram, Centre for Burden of Disease and Cost-Effectiveness, School of Population Health, Edith Cavell Building, Herston Rd, Herston QLD 4006. E-mail: m.bertram@sph.uq.edu.au


Objective: Interventions for individuals with pre-diabetes are efficacious, however, identification of people with pre-diabetes does not occur in Australia. This study aims to calculate the duration of pre-diabetes, in order to provide supporting evidence for a screening program.

Methods: We carried out a systematic review and random effects meta-analysis to identify if an increased risk of mortality is present in people with pre-diabetes. The result of this meta-analysis as well as information on prevalence, remission of pre-diabetes and transition to diabetes from an Australian cohort study, were used in the software program DisMod to calculate duration.

Results: From 2,578 articles identified, 11 studies met the inclusion criteria. The pooled relative risk of all-cause mortality was 1.26 (1.17-1.34) with no sign of heterogeneity between the studies. The average duration of pre-diabetes was 8.5 years in males aged 30+ and 10.3 years in females aged 30+.

Conclusions: The duration of pre-diabetes in Australia is long enough to warrant a screening program. The finding is robust to sensitivity testing of very large variations in the epidemiological parameters.

Implications: If the interventions following screening are shown to be cost-effective, a strong rationale for the implementation of a screening program exists.

An important stage in the progression from normal insulin physiology to type 2 diabetes is pre-diabetes. Pre-diabetes is a period prior to the onset of type 2 diabetes when an individual's blood glucose level is no longer ‘normal’ but does not yet fall into the category of diabetes. Pre-diabetes is measured in two ways. The first of these is impaired fasting glucose (IFG), when fasting blood glucose concentration is higher than normal, between 6.1 and 7 mmol/L but does not rise abnormally during an oral glucose tolerance test (OGTT) remaining less than 7.8 mmol/L. The second form is impaired glucose tolerance (IGT), whereby blood glucose concentration during an OGTT rises to more than 7.8 mmol/L but does not exceed 11.1 mmol/L. These two states are mutually exclusive, however both indicate increased risk of type 2 diabetes.1 The term pre-diabetes was first reported in the literature in 1965.2,3 More recently, pre-diabetes has received increasing attention due to the idea that people in this state can be identified and treatment may prevent or delay the onset of diabetes.

Since the mid-1990s there has been a steady increase in published, randomised controlled trials of interventions for the prevention of diabetes in people with pre-diabetes. In 2007, a systematic review and meta-analysis provided a summary of knowledge to date.4 There were two main categories of intervention, lifestyle changes or pharmaceutical therapy. Lifestyle incorporates diet, exercise or combined diet and exercise interventions. The pharmaceutical therapies investigated included acarbose, metformin, orlistat and glipizide. The meta-analysis provided pooled results for diet, exercise, combined diet and exercise and combined pharmaceutical therapies. With clear evidence of a reduction in diabetes incidence, the challenge lies in identifying people in this health state. Screening is not currently implemented in many countries. Hence, cases are not systematically identified.

Strict criteria are followed when considering implementation of a screening program. These criteria were first published by Wilson and Jungner in 1968,5 and have been adopted by the World Health Organization as criteria for screening. The fourth criterion is that there should be a latent or early symptomatic stage. To date, pre-diabetes is recognised as an early stage for diabetes, however the length of this period is unknown.6

This study used existing data to identify a complete set of epidemiological parameters for pre-diabetes using Australia as an example.


A number of methods were used in order to calculate the duration of pre-diabetes. First, we undertook a systematic review and meta-analysis to establish if there is an increased risk of all-cause mortality in those with pre-diabetes. The search was to identify all studies that measured mortality in participants with pre-diabetes at baseline. Pre-diabetes was defined using WHO criteria on impaired glucose tolerance or impaired fasting glucose.1 Given that until recently not much research was undertaken on pre-diabetes, to maximise the number of studies, all types of study were included in the search. Studies were excluded if an alternative definition of pre-diabetes was used or if insufficient data to incorporate into a meta-analysis were available. They were also excluded if the focus was on gestational diabetes or cardiovascular mortality only.

The search terms used for the systematic review were:

  • 1Pre-diabetes OR impaired glucose tolerance OR impaired fasting glucose OR hyperglycaemia.
  • 2Mortality OR death.
  • 31 AND 2.
  • 4Prediabetic state as MESH term with focus on mortality.
  • 5Prediabetic state as MESH term with focus on epidemiology.

The search was undertaken in Medline via Ovid, using the 1966-Present database in August 2007. No restrictions were placed on the years searched. A ‘snowball’ search was then completed. This required searching through the reference lists of each of the identified studies to establish if there were further studies to incorporate. The identified studies were largely the placebo or control groups of cohort studies that reported mortality as an outcome measure. A random-effects meta-analysis was undertaken in Stata 10 (StataCorp, College Station, TX, US).

Second, we applied this information in a computer program DisMod that uses a series of mathematical equations to estimate a consistent set of disease parameters.7 Duration is one of the outputs. DisMod requires three out of five disease parameters as inputs: incidence, prevalence, remission, case-fatality or RR mortality. The prevalence of pre-diabetes was reported in the Australian Diabetes, Obesity and Lifestyle study, a large cohort study undertaken in Australia in 1999/2000.8 The same individuals were followed up in 2004/05, and remission to normal glucose status as well as transition to diabetes were reported.9 We use the standard errors of prevalence and remission from AusDiab and the standard error of the RR mortality meta-analysis to estimate uncertainty around duration estimates using the sensitivity function in the DisMod software.


The authors identified 2,578 articles, of which 11 met the inclusion criteria and these were used in the meta-analysis of the relative risk of mortality in people with pre-diabetes (Figure 1). The titles and abstracts of the 2,578 articles were reviewed by MB. For the studies for which full text review was undertaken, if inclusion or exclusion was not immediately obvious TV was consulted. Of these 11 studies, three measured pre-diabetes using IFG only10–12 and three used IGT only.13–15 Two studies16,20 had two sub-groups of patients, those with IGT and those with IFG, and were included as separate studies in the meta-analysis. Three studies provided results for pre-diabetes overall.17–19 As pre-diabetes can be defined using either IGT or IFG, meta-analysis was performed across all 13 cohorts. Although the studies were carried out in different settings, the definition of pre-diabetes was the same for each cohort, and each cohort included was representative of the population of the country it is performed in.

Figure 1.

Publications identified, excluded and selected.

Using a random effects model, the pooled increased RR of mortality of people with pre-diabetes was estimated at 1.26 (1.17-1.34) (Figure 2). No evidence for heterogeneity between the trials was found (Q statistic=16.2; p=0.24)

Figure 2.

Forest plot of all-cause mortality in people with pre-diabetes.

Overall duration in males was 7.8 years (95% confidence interval 7.6-8.2 years), ranging from 12.7 (12.4 – 13.4) years at ages 25-34 to 2.7 (2.6-2.7) years at ages 85 and above. In females average duration was 9.8 years (9.6-10.3), ranging from 15.9 (15.6-16.9) years in those aged 25-34 to 3.4 (3.3-3.5) years in those aged 85 and above (Tables 2 and 3).

Table 2.  Inputs and outputs of male Dismod model for pre-diabetes epidemiology.
Age GroupPrevalenceRemissionIncidence of DiabetesIncidence of pre-diabetesDuration of pre-diabetes
All ages0.1830.0590.0520.0207.8
Table 3.  Inputs and outputs of female Dismod model for pre-diabetes epidemiology
Age GroupPrevalenceRemissionIncidence of DiabetesIncidence of pre-diabetesDuration of pre-diabetes
All ages0.1550.0430.0550.0219.8


This study is the first to estimate a consistent set of all parameters (incidence, prevalence, remission, RR of mortality and duration) for pre-diabetes. Prevalence, transition rate to diabetes and rate of remission to normal glucose status were derived from a single, representative study in Australia. Potential increases in mortality have only occasionally been reported. We undertook a systematic review and meta-analysis and found a small but significant increase in mortality risk.

The estimated duration of pre-diabetes varies is sufficiently long to allow for screening of pre-diabetes. The input data used in DisMod were from appropriate data sources – AusDiab and a meta-analysis of mortality in pre-diabetic people – which gives confidence the output for duration is plausible.

The epidemiology of pre-diabetes between countries and the duration calculated in this study may not be immediately applicable to other settings. However, it would take a very large change in input parameters to reduce the duration estimates to such an extent that a different conclusion would be drawn on the opportunity to intervene by screening for pre-diabetes. Sensitivity analyses indicated that incidence of diabetes would need to increase four-fold or remission to normal glucose tolerance increase three-fold to cause a halving of pre-diabetes duration.

Further research should focus on a screening program that identifies people enabling treatment commencement in a cost-effective manner. Efficacious interventions have already been developed, indicating reductions in the risk of diabetes progression of up to 57%.4 Little information is available regarding cost-effectiveness or potential of subpopulations to benefit from screening. Those at high risk of type 2 diabetes through either their age, weight or a family history, or Indigenous people, would be the most obvious candidates for an initial screening program. This has the potential to reduce the burden of disease due to diabetes, as well as preventing downstream sequelae such as cardiovascular disease and renal failure.


This work was supported by an Australian National Health and Medical Research Council Health Services Research Grant [grant number 331558]. MB was supported by a University of Queensland Post Graduate Research Scholarship.