Methods

The NAVIGATOR study design and results have been published. Briefly, 9,306 participants with IGT and cardiovascular disease or risk factors were randomized to nateglinide and/or valsartan in a balanced 2 × 2 factorial design; all participants received a study-specific lifestyle modification program. After randomization, fasting plasma glucose was measured every 6 months for 3 years and annually thereafter. OGTTs were performed annually. Hemoglobin A1c (HbA1c) was measured only at baseline. Progression to diabetes was considered to have occurred if the participant had a fasting plasma glucose level ≥126 mg/dl (≥7.0 mmol/L) or ≥200 mg/dl (≥11.1 mmol/L) 2 hours after a glucose challenge, confirmed by OGTT within the following 12 weeks. The date of diabetes onset was the date of the first elevated glucose value. Among 183 patients, diabetes was diagnosed outside of the study but confirmed by an independent adjudication committee.

Baseline characteristics were summarized by mean and standard deviation for continuous variables and percentages for categorical variables. Characteristics were compared by diabetes status using the Pearson chi-square test for categorical variables and the Wilcoxon rank sum test for continuous variables.

Given the semiannual glycemic assessment schedule, the actual date of transition to diabetes is likely to have occurred before the recorded date of diagnosis. The date of onset was therefore converted to an ordinal category, representing the measurement window in which diabetes was identified (e.g., 0 to 6, 6 to 12, or 12 to 18 months).

A Cox proportional hazards regression model was developed to evaluate the relationship between predictors (risk factors) and the ordinal time to progression to diabetes. Ties were handled by the exact method to reflect the nature of an interval-censored continuous time outcome. Results were similar when missing data were handled by single or multiple imputation using Markov Chain Monte Carlo methods to create a monotone missing pattern, followed by regression methods; we report single imputation here for simplicity. Less than 3% of data were missing for all covariates except glycosylated hemoglobin (HbA1c), which had 15% missing. All variables were checked for linearity and proportional hazards. Linear splines were used to account for nonlinear relationships with outcomes. The proportional hazards assumption was met for all variables in our model.

The full diabetes risk prediction model included 10 established predictors of interest, selected a priori according to clinical judgment rather than statistical significance, and the remaining candidate variables were added by forward selection with a p value <0.05 ( Supplementary Appendix Table 1 ). Randomized treatment assignments were examined as a candidate variable in a sensitivity analysis.

We also examined 5 simplified models that excluded variables unlikely to be available in routine care of IGT patients (hemoglobin, platelet count) and examined the relative importance of glucose tolerance measures (fasting glucose, 2-hour glucose, and HbA1c) in assessing diabetes risk. The resulting models differ from the full model by excluding hemoglobin and platelet count and by including the following glucose tolerance measures: Model A—fasting and 2-hour glucose (an OGTT), HbA1c; Model B—fasting glucose only; Model C—an OGTT only; Model D—HbA1c only; and Model E—fasting glucose and HbA1c only. The predictive capacity of the models is summarized by the C-index, which is a measure of a model’s ability to discriminate risk ranging from 0.5 (poor) to 1 (perfect). Because the C-index is known to be optimistic when calculated in the same data set on which a model is developed, we additionally calculated the optimism-corrected C-index.

We then proposed 3 clinically motivated 5-year risk classes for developing diabetes based on consensus among authors regarding thresholds delineating distinct therapeutic choices. High risk for diabetes progression was defined as approximately 10% per year, based on the annual risk in the placebo group seen in other recent diabetes prevention studies: approximately 13% in the Study To Prevent Non-Insulin Dependent Diabetes Mellitus (STOP-NIDDM), ∼11% in the Diabetes Prevention Program (DPP), approximately 7% in NAVIGATOR, and approximately 6% in the Diabetes REduction Assessment with ramipril and rosiglitazone Medication (DREAM). Modest risk for diabetes progression was defined as 0% to 25% 5-year risk and is assigned a “watchful waiting” clinical approach in which only routine lifestyle advice is given. For those with a moderate risk (>25% to 50%), an intensive lifestyle-based intervention is prescribed, and for high-risk (>50%) participants, we posited a combination of intensive lifestyle and pharmacologic interventions. We created risk classification tables to compare the simplified models to the full NAVIGATOR model, assuming it to be best suited to predict incident diabetes risk in this population.

Recognizing that existing risk prediction tools, such as FINRISK and San Antonio Heart models, both developed in populations unselected for glucose tolerance, might also be used in our population, we evaluated their ability to discriminate risk in our population. The FINRISK score, developed from a random population sample of Finnish adults without diabetes at baseline, includes categorical variables for age (45 to 54 and 55 to 64 years), body mass index (BMI; >25 to 30 and >30 kg/m ² ), waist circumference (94 to <102 and ≥102 cm [men]; 80 to <88 and ≥88 cm [women]), use of blood pressure medication, and history of high blood glucose. The San Antonio Heart model, developed from a random sample of adults in San Antonio, Texas, without diabetes at baseline, includes continuous variables for age, gender, ethnicity, fasting glucose level, systolic blood pressure, high-density lipoprotein cholesterol, BMI, and family history of diabetes. These studies predicted long-term diabetes at 10 and 7.5 years, respectively, via a complete case logistic regression, excluding loss to follow-up and mortality. Because of the differences in population and follow-up, these models cannot be used to predict absolute 5-year event rates in our population. However, relative risk is less sensitive to these differences, and these models can still be used to discriminate risk in our population. Therefore, we have treated the linear predictions as scores and computed the C-index.

Results

Over a median of 5 years of follow-up, 3,254 (35%) NAVIGATOR participants developed diabetes. Progressors were slightly younger, more likely to have a family history of diabetes, and had a higher baseline BMI ( Table 1 ).They also had higher baseline fasting and 2-hour postchallenge glucose values and levels of HbA1c. Progressors were more likely to receive aspirin, angiotensin-converting enzyme inhibitors, calcium channel blockers, and lipid-lowering drugs at baseline, but no differences were seen in use of beta blockers or diuretics.

The full model shown in Table 2 includes both those variables forced in and those selected based on statistical significance. The full model C-index is 0.70 (optimism corrected C-index of 0.70, reflecting minimal optimism). No statistically significant interactions were identified among age, gender, BMI, and fasting glucose values. Sensitivity analyses including randomized treatments showed that only valsartan made a statistically significant contribution, but this did not change the C-index or alter the hazard ratios for the other variables ( Supplementary Appendix Table 2 ).

The simplified models are shown in Table 3 . Equations for the full and simplified models are included in Supplementary Appendix Table 3 . All had C statistics ranging from 0.67 to 0.70 except for Model D (C = 0.63), which used HbA1c as the sole glucose tolerance measure and was excluded from further analyses. Comparison of the C statistics for the full and simplified models with FINRISK (C = 0.55) and San Antonio Heart scores (C = 0.63) shows that both had less discriminative ability than the remaining NAVIGATOR-derived models.

Table 3

Reduced proportional hazards model for progression to diabetes

Characteristic	Model A		Model B		Model C		Model D		Model E
	HR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value
Age (per 10 yrs) ∗	0.88 (0.84–0.93)	<0.0001	0.93 (0.88–0.98)	0.008	0.92 (0.87–0.97)	0.002	0.88 (0.83–0.93)	<0.0001	0.89 (0.84–0.94)	<0.0001
Female ∗	1.04 (0.97–1.13)	0.29	1.11 (1.02–1.19)	0.01	1.09 (1.01–1.17)	0.04	0.97 (0.90–1.05)	0.44	1.06 (0.98–1.15)	0.14
Region (vs North America) †
Asia	1.83 (1.11–3.00)	0.02	1.60 (0.99–2.58)	0.06	1.59 (0.98–2.58)	0.06	1.67 (1.03–2.71)	0.04	1.84 (1.13–3.00)	0.02
Europe	0.84 (0.77–0.93)	0.0004	0.82 (0.75–0.91)	<0.0001	0.84 (0.76–0.92)	0.0003	0.92 (0.83–1.01)	0.07	0.82 (0.75–0.91)	<0.0001
Latin America	0.95 (0.84–1.07)	0.36	0.96 (0.85–1.08)	0.45	0.95 (0.84–1.07)	0.40	1.01 (0.90–1.14)	0.86	0.95 (0.84–1.07)	0.36
Other	0.79 (0.64–0.98)	0.04	0.85 (0.69–1.06)	0.14	0.86 (0.70–1.07)	0.17	0.76 (0.61–0.95)	0.01	0.77 (0.62–0.96)	0.02
Race (vs White) ∗
Other	0.99 (0.86–1.15)	0.91	1.02 (0.89–1.18)	0.77	1.04 (0.90–1.20)	0.63	0.86 (0.75–1.00)	0.04	0.97 (0.84–1.12)	0.71
Black	0.79 (0.62–1.00)	0.05	1.02 (0.81–1.28)	0.89	1.03 (0.82–1.30)	0.80	0.67 (0.53–0.85)	0.0008	0.76 (0.60–0.96)	0.02
Asian	0.50 (0.31–0.81)	0.005	0.59 (0.37–0.94)	0.03	0.58 (0.37–0.93)	0.02	0.53 (0.33–0.84)	0.008	0.51 (0.32–0.81)	0.005
Family history of type 2 diabetes mellitus ∗	1.12 (1.04–1.20)	0.002	1.15 (1.07–1.23)	0.0002	1.12 (1.05–1.21)	0.001	1.14 (1.06–1.23)	0.0003	1.14 (1.06–1.23)	0.0003
Previous history of cardiovascular disease ∗,‡	1.04 (0.96–1.12)	0.31	1.12 (1.04–1.21)	0.005	1.11 (1.03–1.20)	0.006	1.01 (0.94–1.10)	0.72	1.04 (0.96–1.12)	0.32
BMI (kg/m 2 ) ∗	1.01 (1.00–1.02)	0.002	1.01 (1.01–1.02)	<0.0001	1.01 (1.01–1.02)	<0.0001	1.02 (1.01–1.02)	<0.0001	1.01 (1.00–1.02)	0.002
Systolic blood pressure (per 10 mm Hg) ∗	1.02 (1.00–1.04)	0.08	1.02 (0.99–1.04)	0.17	1.01 (0.99–1.03)	0.28	1.03 (1.01–1.05)	0.01	1.02 (1.00–1.04)	0.04
Fasting glucose (per 10 mg/dl) ∗	1.70 (1.63–1.78)	<0.0001	1.87 (1.79–1.96)	<0.0001	1.80 (1.72–1.89)	<0.0001	_	_	1.75 (1.68–1.84)	<0.0001
2-hour glucose (per 10 mg/dl) ∗	1.16 (1.14–1.19)	<0.0001	_	_	1.17 (1.15–1.20)	<0.0001	_	_	_	_
Hemoglobin A1c (%) ∗	1.94 (1.78–2.10)	<0.0001	_	_	_	_	2.41 (2.22–2.61)	<0.0001	2.01 (1.85–2.19)	<0.0001
Low-density lipoprotein (per 10 mg/dl)	0.98 (0.97–0.99)	<0.0001	0.98 (0.97–0.99)	<0.0001	0.98 (0.97–0.99)	<0.0001	0.97 (0.96–0.98)	<0.0001	0.97 (0.96–0.98)	<0.0001
High-density lipoprotein (per 10 mg/dl)	0.93 (0.91–0.96)	<0.0001	0.93 (0.91–0.96)	<0.0001	0.94 (0.91–0.97)	<0.0001	0.92 (0.89–0.94)	<0.0001	0.93 (0.90–0.95)	<0.0001
Platelet (10 3 /μl) (per 10 increase)	_	_	_	_	_	_	_	_	_	_
Hemoglobin (per 10 g/L)	_	_	_	_	_	_	_	_	_	_
C-index	0.70		0.67		0.68		0.63		0.69

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

1. Impact of Age on Long-Term Outcome After Primary Angioplasty With Bare-Metal or Drug-Eluting Stent (from the DESERT Cooperation)
2. Cardiovascular Risk Profile and Lifestyle Habits in a Cohort of Italian Cardiologists (from the SOCRATES Survey)
3. Usefulness of Speckle Tracking Echocardiography in Hypertensive Crisis and the Effect of Medical Treatment
4. Procedural Complications May Affect Serum Bilirubin Levels in Patients With ST-Elevation Myocardial Infarction

Stay updated, free articles. Join our Telegram channel

Join