Evaluating AUSDRISK for predicting incident diabetes in an independent sample of women

To the Editor: Chen and colleagues1 published a risk assessment tool for type 2 diabetes (AUSDRISK) based on the Australian Diabetes, Obesity and Lifestyle Study (AusDiab).2 We tested AUSDRISK’s performance in an independent cohort of 1494 women enrolled in the Geelong Osteoporosis Study (1994–1997; 77% participation),3 comprising an age-stratified sample of women randomly selected from the Barwon Statistical Division and followed prospectively over a decade.4

In 2004–2008, of 1015 surviving study participants aged 25 years or older at enrolment, 800 (79%) returned for follow-up assessment. We excluded 261 women who did not have a fasting plasma glucose (FPG) test result at both baseline and follow-up assessments, and 33 with baseline diabetes. The remaining 506 women formed the cohort on which the AUSDRISK tool was tested.

Diabetes was defined by one or more of three criteria: FPG level ≥ 7.0 mmol/L, treatment with insulin or oral hypoglycaemic agents, or self-report. Demographics, ethnicity and lifestyle factors were documented by questionnaire. Participants were described as “active” if they described their mobility as “moves, walks and works energetically, and participates in vigorous activity”; otherwise, they were considered “inactive”. As our baseline questionnaire did not document a history of high glucose levels, we performed two analyses: one assuming no participants had this history, and a second identifying participants with baseline impaired fasting glycaemia (FPG level, 6.1–6.9 mmol/L). The study was approved by the Human Research Ethics Committee, Barwon Health.

Using the final AUSDRISK model,1 we allocated points for baseline characteristics according to sex, age, ethnic background, parental history of diabetes, history of high blood glucose (FPG level, ≥ 6.1 mmol/L), use of antihypertensive medications, current smoker status, physical inactivity, and waist circumference. The predictive power of AUSDRISK was determined using the area under the receiver operating characteristic curve (AROC). Using a total AUSDRISK score ≥ 12 as the criterion for prediction of diabetes, we evaluated the performance of AUSDRISK by calculating its sensitivity, specificity and positive predictive value (PPV) in our cohort. Ninety-eight participants had an AUSDRISK score ≥ 12 (or 106 if those with impaired fasting glycaemia were scored for a history of high blood glucose). Statistical analyses were performed using Stata software, version 9 (StataCorp, College Station, Tex, USA).

Twenty-eight participants (5.6%) developed incident diabetes during the period of follow-up (13 with FPG ≥ 7.0 mmol/L, 14 receiving treatment with insulin or hypoglycaemic agents, and seven self-reporting the condition). If we assumed that none of the participants had a history of high blood glucose levels, the AROC for AUSDRISK in the Geelong cohort (0.78 [95% CI, 0.72–0.85]) was comparable with that in the AusDiab cohort (0.78 [95% CI, 0.76–0.81]).1 In our study, the sensitivity of the AUSDRISK tool was 50.0% (95% CI, 30.6%–69.4%), specificity was 82.4% (95% CI, 78.7%–85.7%) and PPV was 14.3% (95% CI, 8.0%–22.8%). Recognising baseline impaired fasting glycaemia increased AUSDRISK’s predictive power (AROC, 0.81 [95% CI, 0.74–0.88]; sensitivity, 60.7% [95% CI, 40.6%–78.5%]; specificity, 81.4% [95% CI, 77.6%–84.8%]; and PPV, 16.0% [95% CI, 9.6%–24.4%]).

Study limitations were that we only evaluated women, we did not collect data on a history of high blood glucose levels, diabetes was diagnosed in the absence of an oral glucose tolerance test, and criteria for inactivity differed from those used in the AusDiab study. Our population was older than that of the AusDiab study and would probably have had a higher prevalence of diabetes, influencing our PPV result. Not surprisingly, including individuals with impaired fasting glycaemia increased the point estimates for AROC, sensitivity and PPV. In conclusion, our data independently demonstrate the limited predictive value of AUSDRISK for women over a 10-year period.