Comment on Tay et al. A Very Low-Carbohydrate, Low–Saturated Fat Diet for Type 2 Diabetes Management: A Randomized Trial. Diabetes Care 2014;37:2909–2918

We read with interest the article by Tay et al. (1) that compared the effects of a very low-carbohydrate (LC) diet with those of a high–unrefined carbohydrate (HC) diet on glycemic control and cardiovascular disease risk factors in type 2 diabetes. The best result of this randomized trial was body weight reduction, which averaged 12 kg at 24 months in both groups. This result is important considering that the diets were moderately energy restricted (about 1,500 kcal/day), the patients with diabetes were overweight (few) or obese (most), and about 16% of the obese patients lost about 18 kg (beyond 1 upper standard deviation). While both diets, incorporated as part of a lifestyle program, achieved similar weight loss in overweight and obese adults with type 2 diabetes, the LC diet was associated with greater improvement of glycemic control. As clearly stated in Research Design and Methods section, the primary outcome was HbA_1c; surprisingly, there was no mention of what happened to the primary outcome in the whole population after 24 months of intervention. The reported scatterplot and regression lines of HbA_1c at 24 weeks against week 0 (see Fig. 2) do not help the average reader to understand the relative merit of each diet on the primary outcome. In order to dissolve the impression that the regression slopes were used on a posteriori basis, after the acquisition of the main 24-month results, the authors should have specified that a significant effect of the LC diet, as compared with the HC diet, on the primary outcome (HbA_1c) was not met (as we believe) in the whole population. Subgroup analyses may be important, but it is never prevalent on the primary, prespecified outcome as subgroups analyses are based on smaller numbers, as in the case of the study by Tay et al. Only 9 out of 46 patients on the LC diet (19.6%) and 14 out of the 47 patients on the HC diet (29.8%) were analyzed for difference in HbA_1c at 24 weeks because they presented a starting HbA_1c value >7.8% (62 mmol/mol).

One merit of the trial is to show that the relation between starting HbA_1c and its change after treatment also applies to dietary interventions. Previous analyses between trials have noted that baseline HbA_1c level is a strong predictor of response to glucose-lowering agents, including insulin (2): baseline HbA_1c level accounts for 36% of the improvement in HbA_1c levels when starting therapy with dipeptidyl peptidase-4 inhibitors (3) and 50% or more with insulin analogs (4,5).

Although the trial by Tay et al. (1) has some merit, it failed to reach its primary end point and the results must be smoothed accordingly: both diets are good for patients with type 2 diabetes to lose weight (important), with some additional effect for the LC diet on HbA_1c level but only for that minority of patients with a baseline HbA_1c level >7.8% (62 mmol/mol).