Effects of Wheat Flour Milling Method on Glycemic Response and Gastrointestinal Symptoms

Wheat is processed today almost exclusively by roller milling, a method introduced in the late 19th century that has largely displaced traditional stone milling. This technological shift has altered the physical structure of flour to a much greater extent than existing research has examined: because intact plant structure limits enzymatic access to starch, coarser, less fractionated flour is expected to be digested more slowly and produce a smaller postprandial glucose rise than finely milled flour. Stone milling has been found to produce measurably coarser, less fully fractionated flour than roller milling, and one prior human trial - conducted in adults with type 2 diabetes, using a single-occasion glycemic index design - found a significantly lower postprandial glucose response to bread made from stone-ground versus roller-milled wholegrain wheat flour. Whether this effect is present and reproducible in healthy adults, using repeated within-condition measurement rather than a single test occasion, has not been established. Separately, cereal-based foods are a common source of gastrointestinal symptoms even without diagnosed gastrointestinal disease, and milling method could plausibly influence gastrointestinal tolerance through its effects on particle size and fermentable substrate delivery to the colon, independent of its effects on glycemic response.

This trial is a randomized, crossover study designed to isolate the effect of milling method - stone milling versus roller milling - while holding grain source, whole-grain fraction, fermentation method, and carbohydrate dose constant across conditions. Both study breads are produced from a single batch of wheat grain in one dedicated milling run per method, are matched on 100% whole-grain composition (all milling streams recombined after roller milling), and are sourdough-fermented using the same starter and protocol. Flour particle-size distribution and damaged starch - the two properties through which milling method is expected to act - are measured and reported but deliberately not standardized between conditions, since doing so would remove the mechanism under study. Each bread-containing meal delivers a fixed, directly analyzed dose of available carbohydrate rather than a fixed number of slices, unless the two breads are shown before enrollment to match within 5% per slice.

Each participant is randomized (stratified by sex, in permuted blocks, by a person independent of trial conduct) to one of two treatment sequences and consumes each bread type for a 3-day period, separated by a 4-day washout of standardized yeast-baked bread that also aligns each period to the same weekdays across the two halves of the study and attenuates carryover from the preceding intervention. A continuous glucose monitor is worn continuously for the full 14-day study, beginning with a 1-day sensor calibration period; breakfast and lunch bread meals are eaten at fixed times of day at least 4 hours apart, and each participant fixes their own topping and beverage pattern at the outset and repeats it identically throughout, holding non-bread dietary components constant within, though not between, participants. Both study staff and participants are blinded to which bread corresponds to which milling method: breads are prepared and blinded by the research institute producing them (Nofima), and treatment allocation is withheld from the analysis dataset until after data lock. Throughout the study, participants record all food and beverage intake, including photographs of meals, and higher-intensity physical activity, via a dedicated digital application that timestamps bread-meal entries at the point of consumption.

The primary analysis uses a design-driven mixed-effects model with fixed effects for treatment and period, a variance structure allowing period-specific residual variance to reflect known changes in CGM sensor accuracy across the wear period, and baseline handled as recommended by Kenward and Roger to avoid cross-level bias - reflecting that single-occasion postprandial measurements have poor within-participant reliability, which is why the trial averages responses over three days per condition rather than relying on a single test day. No formal statistical test for carryover is performed, since testing for carryover and conditioning the primary analysis on that test is known to inflate the type I error rate in two-period crossover designs; carryover is instead addressed through prespecified sensitivity analyses, including a model restricted to the sole strictly pre-exposure baseline and a descriptive comparison of period-2 baseline values by sequence. Additional prespecified analyses examine the time course of the treatment effect across the three days of each period, differences between breakfast and lunch, and day-to-day response variability. The planned sample size was informed by a power calculation transferring effect-size and reliability estimates from the only prior comparator trial, adjusted for this trial's repeated-measures design.

The trial is conducted at Haukeland University Hospital (Bergen, Norway) as an investigator-initiated study, in collaboration with Nofima (production and blinding of study breads) and two Norwegian mills providing grain and milling in-kind. Findings are expected to apply specifically to the wheat batch, mills, fermentation protocol, carbohydrate dose, and population studied; given the trial's short, acute design in healthy young adults, results cannot support inference about long-term glycemic control, insulin sensitivity, body weight, or cardiometabolic disease risk.