A prediction model for individual glycaemic responses to food

This study used a wide array of clinical, anthropometric, dietary and microbial data to develop a model to predict individual postprandial glycaemic responses in people with type 1 diabetes. Participants who were using insulin pump therapy supported with continuous glucose monitoring were instructed to log, in real time, data including food intake, sleep times, physical activity and non-insulin medications, using a proprietary smartphone app (www.personalnutrition.org). Insulin doses were logged by the insulin pump. Over a two-week period, participants logged their usual meals, as well as a number of pre-prepared meals with specific energy, carbohydrate, fat and protein levels. They also provided a sample for evaluation of their gut microbiota. The final model also incorporated postprandial data from 900 individuals without diabetes to improve accuracy.

A total of 121 people with type 1 diabetes (75 adults and 46 children), with an average diabetes duration of around 10 years and generally good glycaemic control, provided glucose data in response to a total of 6377 meals. Compared with a model based on standard carbohydrate counting (incorporating meal carb content and pre-meal blood glucose level), the authors’ model was significantly more accurate, with a correlation (Pearson’s R) of 0.59 between predicted and observed postprandial glucose levels, compared with the simpler model’s R of 0.40. Gut microbial composition had a significant impact on an individual’s postprandial blood glucose levels.

The authors conclude that their model enables more accurate prediction of an individual’s glycaemic response to meals. They state that it could potentially be used to improve algorithms in closed-loop insulin delivery systems, bolus calculators and glucose alarm systems. It could also lead to development of personally tailored nutritional interventions for people with diabetes.

Click here to read the study in full.