Beneficial Effects of Quinoa (Chenopodium Quinoa Willd) in the Prevention of Type 2 Diabetes Mellitus

The study was a cross-over pilot clinical study consisting of two periods. The first period was only an observational and monitoring phase where participants just continued with their regular diet (RD), for this reason all participants initiated this period and wash-out term was no needed. Subsequently, with the data of the first phase obtained, the subjects began the second period in which they had to undergo a nutritional intervention with a quinoa diet (QD).

A total of six visits plus two quinoa products collection days were programmed. After the pre-study visit (V0) which took place a week before start nutritional intervention and where researchers obtained signed informed consent, participants were summoned for a first visit (V1) where they were explained how they should fill in the dietary records and they were applied with the FreeStyle Libre®.

Subjects then began RD, a period of 4 weeks during which only their normal life was monitored. After the first 14 days of this, a second visit (V2) was made where the dietary record was collected that would serve to account for their usual consumption of cereals, flours, tubers and legumes, the FreeStyle Libre® sensor was also collected.

The last day on RD period, on day 28, they were cited (V3) in consultation where blood samples after an 8 hours fast, anthropometrics measurements and blood pressure measure where obtained and for the placement of the new FreeStyle Libre® sensor. Participants were asked about their physical activity and exercise practice during those past 4 weeks by a short questionnaire adapted from the Minnesota Leisure Time Physical Activity Questionnaire for individuals of advanced age (VREM questionnaire) and a new empty 14-day dietary record was given. In addition, the volunteers received the first foods with quinoa to initiate QD the next day. Products were delivered weekly, for conservation reasons but also to ensure that they followed an adequate consumption, they had to go through consultation to pick up the product and gave the researchers the empty packs where quinoa products had been.

On the next visit the day 42 (V4) the Freestyle Libre® sensor was collected and the filled dietary record was collected. Finally, after 28 days of quinoa diet they were summoned for the last visit the day 56 (V5) where all the determinations were repeated identically as V3.

With the premise that the products created replaced not only grains, legumes or tubers, but also farinaceous commonly consumed by the participants and that only the cereal fraction was modified, similar products based on quinoa flour were created. The creation of these products was necessary, after conducting a market search where it was observed that there was not enough food to replace those consumed since these had percentages of quinoa flour not exceeding 20-30%.

Thus, apart from delivering quinoa, quinoa flakes and quinoa flour to the participants, they were given products created with ≥70% quinoa flour and were biscuits, crackers, brioche, sponge cake, baguette bread, sliced bread and pasta. Moreover, a quinoa-based recipe was delivered with eight commonly consumed recipes that replaced the tuber, legume or grain of the recipe. Each subject received the equivalent of what they consumed according to their RD dietary records. Thus, only if the volunteer had indicated that he consumed sponge cake was the quinoa-based product delivered to him.

Descriptive data are presented as the mean and standard deviation (SD) or median and interquartile range (IQR) for continuous variables, and the frequencies and percentages (%) for categorical variables. Anthropometric measurements, blood test variables and dietary intake were compared at different times using the non-parametric Wilcoxon signed rank test because normality and equality of variance could not be assumed due to small sample size (n=9). In order to compare variables related to dietary patterns, mean value for dietary intake, including all meals, was considered for each participant.

The glucose level monitoring sensor takes measurements at discrete time points for each patient. Therefore, firstly the glucose curves have been linearly interpolated in order to have observations for each patient at equal time points. A first sight to the glucose curves over the day shown that they were more homogeneous around breakfast than around other later meals intakes. Therefore, the glucose concentration values corresponding to the breakfast were considered as a function of time in minutes over the interval t= [-30,120], that begins half an hour before the start of breakfast and ends two hours later. Before constructing a functional model, the functional data were time aligned in order to reduce the differences between different patients and/or different days (for instance, some patients could mark the starting time of breakfast systematically before than others, or spend systematically more time in breakfast than the average). The time alignment has been done by warping functions, using the function WFDA in the R package fdapace.

Once the glucose level curves have been synchronized, a functional regression analysis was conducted to model the effect of diet type, patient and nutrient intake on monitored glucose levels. Three different explanatory variables have been considered: diet type with two categories (regular and Quinoa diets), patient indicator (categorical variable with nine levels) and the contents in different nutrients. The breakfast glucose curves are handled as the functional response variable. To study the relationship between these variables function on scalar regression (fosr) models were used.

Firstly, the univariate effect of the diet type on glucose curves have been analyzed and then more complex fosr models with two factors (diet and patient factors) and scalar variables (nutrients) have been constructed. The most complex model including all effects of the independent variables is defined by a specific-created equation. The functional regression models have been fitted by penalized flexible functional regression, as implemented in the function pffr of the R package refund.