Effects of soy protein and isoflavones on glycemic control and insulin sensitivity: a 6-mo double-blind, randomized, placebo-controlled trial in postmenopausal Chinese women with prediabetes or untreated early diabetes.

This was a 6-month double-blind, randomized, placebo-controlled trial in 180 postmenopausal Hong Kong Chinese women with prediabetes or untreated early diabetes. Participants were aged 48-70 years, and the three arms were balanced at 60 per group with block randomization; analyses included intention-to-treat and a per-protocol completer set.

Participants received either 15 g soy protein plus 100 mg isoflavones daily for 6 months or 15 g milk protein plus 100 mg isoflavones daily for 6 months, compared with a milk protein placebo. The supplements were provided as one 25 g packet per day, mixed with liquid and taken preferably at breakfast; the soy protein packet naturally contained a small amount of isoflavones and additional isoflavones were added to reach 100 mg total.

Overall, soy protein with isoflavones and isoflavone supplementation did not produce meaningful improvements in glycemic control or insulin sensitivity versus milk protein placebo. Between-group differences were not significant for fasting glucose (P = 0.8523), 2-hour postload glucose (P = 0.1623), glycated serum protein (P = 0.8793), fasting insulin (P = 0.6983), HOMA-IR (P = 0.6413), HOMA-b (P = 0.5733), or QUICKI (P = 0.9203). A baseline-adjusted sensitivity analysis suggested a marginal 2.34% reduction in 2-hour postload glucose at 6 months (P = 0.077), but this did not change the overall negative conclusion. Adverse events were similar across groups; gastrointestinal discomfort was the most common, and mild vaginal bleeding occurred in 3 subjects (2 soy, 1 isoflavone).

The trial was relatively small, lasted only 6 months, and relied on multiple metabolic endpoints and subgroup analyses, increasing the chance of chance findings. The one favorable signal was only marginal after baseline adjustment, and arm-level completer counts were not fully specified, which limits interpretation of robustness and generalizability.