The flow chart for the study was published in the primary paper . Participants (n = 162) were randomised from January 2002 to the end of October, 2003; for this paper, 25 participants were excluded because of insufficient data as follows: (1) OGTT at baseline (n = 11) or baseline and 3 months (n = 10) only; and (2) indices not calculable because blood samples were only available for fasting and 2 h (n = 4). This left 137 participants with an OGTT at 6 and/or 12 months. Two participants were excluded because of outlying values for muscle insulin sensitivity index. One missing baseline OGTT was imputed as the mean of the three other values. Values missing at 3 or 6 months (n = 21) were imputed as the mean of the values before and after. Values missing at 12 months (n = 25) were replaced by the 6 month values (last value carried forward).
At baseline, there were no significant differences in glucose, insulin or indices of insulin sensitivity or secretion among diet groups; however, compared with women, men had significantly higher waist circumference, muscle insulin sensitivity index and MIS × II, and lower 2 h plasma insulin and insulinogenic index (Table 1). Records of concomitant medications used during the trial were available for 126 of the 135 participants included in this paper (93%). Since all participants had type 2 diabetes treated by diet alone, none were taking any oral hypoglycaemic or anti-hyperglycaemic drug or insulin at baseline or during the trial. There were no significant differences among diet groups in the number of participants taking medications at baseline nor in the number of participants who changed dose during the trial for any of the 18 categories of concomitant medications (Electronic supplementary material [ESM] Table 1). Steroids used during the trial were all short courses as follows: (1) cortisone injection into a joint on one (n = 2) or two (n = 1) occasions; (2) topical prednisone for 2 to 3 weeks for dermatological conditions (n = 2); (3) inhaled prednisone for bronchitis for 6 weeks (n = 1); and (4) systemic prednisone for bronchitis for 6 (n = 1) or 9 days (n = 1). Participants taking aspirin had lower HOMA-r at baseline than those who did not, i.e. 2.5 (95% CI 2.1, 3.0) vs 3.6 (3.0, 4.1; p = 0.007). Participants who took non-steroidal anti-inflammatory drugs (NSAIDs) had higher insulinogenic index, i.e. 33 (26, 40) vs 25 (21, 29; p = 0.044) pmol/mmol, higher muscle insulin sensitivity index, 0.022 (0.010, 0.033) vs 0.010 (0.007, 0.013; p = 0.006) mmol pmol−1 h−1 and higher disposition index, 0.34 (0.23, 0.45) vs 0.19 (0.14, 0.25; p = 0.009) h−1 at baseline than those who did not. Participants taking nutritional supplements had higher insulinogenic index at baseline than those who did not, i.e. 31 (25, 36) vs 23 (19, 27; p = 0.041) pmol/mmol. Participants on statins had higher insulinogenic index at baseline than those who did not, i.e. 31 (26, 36) vs 22 (18, 26; p = 0.009) pmol/mmol.
Energy and protein intakes did not change significantly on the High-GI, Low-GI and Low-CHO diets (Table 2). Total fat intake decreased from baseline on the High-GI and Low-GI diets and increased on Low-CHO; about 66% of the increase in fat on Low-CHO was accounted for by MUFA and the remainder was accounted for by polyunsaturated fat. Intake of saturated fat fell slightly but significantly on all three diets. Carbohydrate intake increased from baseline on the High-GI and Low-GI diets and decreased on Low-CHO. Glycaemic index increased from baseline on High-GI, decreased on Low-GI and did not change on Low-CHO. Thus, glycaemic load increased on High-GI and decreased on Low-CHO, but did not change significantly on Low-GI. Dietary fibre intake did not change on High-GI or Low-CHO, but increased significantly on Low-GI.
Adverse events, body weight and HbA1c
There were no important adverse effects associated with the treatments. As reported previously , body weight fell by ∼0.5 kg over the first 8 weeks, then rose steadily to ∼0.8 kg higher than at baseline by 12 months (p = 0.0005). Mean weight on Low-GI was 0.4 kg less than that on High-GI and Low-CHO (p = 0.062). The primary endpoint, HbA1c, rose from ∼0.061 at baseline to ∼0.063 after 12 months (p < 0.0001), but there was no significant difference among diets .
Glucose and insulin during OGTT
There was a significant time × diet × visit interaction (p = 0.033) for plasma glucose during the OGTT indicating that the pattern of responses to the different diets varied significantly over the course of the study. At 3 months unadjusted 2hPCG did not differ significantly from baseline on any diet, but the change on Low-CHO was significantly smaller than that on High-GI by 1.26 (0.22, 2.30) mmol/l (p < 0.05) (ESM Fig. 1). However, by 12 months 2hPCG had risen significantly from baseline on both the High-GI and Low-CHO diets by 1.41 (0.37, 2.45) and 1.33 (0.29, 2.37) mmol/l, respectively (p < 0.05). The increase in 2hPCG at 12 months on Low-GI, 0.37 (−0.63, 1.37) mmol/l, was not significant from baseline and was 1.06 (0.02, 2.10) mmol/l less than that on High-GI (p < 0.05) (ESM Fig. 1). There were no significant differences among diets for plasma insulin.
Insulin sensitivity, insulin secretion and disposition index
There was no significant effect of time (3–12 months), diet, or time × diet interaction, respectively, for the residuals of HOMA-r (p = 0.25, p = 0.72, p = 0.83), HOMA-b (p = 0.40, p = 0.91, p = 0.96), HIR (p = 0.71, p = 0.59, p = 0.95), insulinogenic index (p = 0.78, p = 0.77, p = 0.94) or muscle insulin sensitivity index (p = 0.37, p = 0.67, p = 0.38) (Fig. 1). However, a time × diet interaction was seen for MIS × II (p = 0.036). Initially, the overall mean for MIS × II tended to be higher on Low-CHO than on Low-GI by 0.07 (−0.04, 0.18) h−1, but by the end of the study, the reverse was seen and MIS × II was 0.12 (0.01, 0.23) h−1 higher on Low-GI than on Low-CHO (p < 0.05, pooled baseline value 0.23 h−1), with the value for High-GI being intermediate (Fig. 1).
The use of aspirin, NSAIDs, nutritional supplements or statins did not significantly affect the response of MIS × II to the Low-GI compared with the Low-CHO diet. In the 126 participants for whom data on drug use was available, the significance of the time × diet interaction for MIS × II, adjusted for baseline and sex, was p = 0.013; this p value was unchanged after additional adjustment for use of aspirin, NSAID, nutritional supplements or statins during the trial. Regardless of whether participants were taking these drugs or not, the residuals for MIS × II after 12 months on Low-GI always tended to be greater than those on Low-CHO (ESM Table 2).
Changes (Δ) in 2hPCG were not significantly related to ΔHOMA-r, ΔHOMA-b or ΔHIR at any time. Δ2hPCG was related to Δmuscle insulin sensitivity index at 3, 6 and 12 months (r = −0.463, −0.466 and −0.502 respectively, p < 0.0001 for all). Δ2hPCG was significantly related to Δinsulinogenic index at 6 (r = −0.204, p = 0.017) and 12 months (r = −0.194, p = 0.024), but not at 3 months (r = −0.092). Δ2hPCG was related to ΔMIS × II at 3, 6 and 12 months (r = −0.577, −0.592 and −0.579 respectively, p < 0.0001). The correlations between Δ2hPCG and ΔMIS × II were greater than those between Δ2hPCG and Δinsulinogenic index (p < 0.05), and were non-significantly greater than those between Δ2hPCG and Δmuscle insulin sensitivity index (Fig. 2).