Creating personalised recommendations for the management of Type 2 Diabetes Mellitus

Type 2 Diabetes Mellitus (T2DM) is an endocrine disorder, characterised by the inability of the patient to effectively control hyperglycaemia. Despite existing pharmaceutical strategies to manage T2DM, the holistic approach of management using diet and lifestyle has been proven beneficial in the identified literature. The current dietary guidelines published by the National Institute for Health and Care Excellence (NICE) in the UK include recommended consumption of complex carbohydrates and controlling sugar intake. However, these guidelines are generalised, not taking individualisation into consideration. Artificial Intelligence (AI) has been proven as an efficient approach to predict glycaemia in patients; but research into AI to predict Diabetes Status based on diet and lifestyle, which would enable personalisation of T2DM management, is scarce. Therefore, the aims of this PhD-thesis were to investigate compliance of participants in the National Diet and Nutrition Survey (NDNS) 2008 to 2017 cohorts in the UK, to the NICE dietary guidelines for T2DM management; and to explore the capabilities of AI in using dietary and lifestyle profiles to predict an individual with T2DM. For all studies in the PhD-thesis, the NDNS data were sorted to contain adult participants aged 18 years and older, then sorted into four study groups: self-reported Diabetes; participants with potential Diabetes, or Pre-diabetes, based on their blood glucose or Glycohaemoglobin (HbA1c) readings, and participants without Diabetes or Pre-diabetes (Control). For the compliance study, details of recommended intakes of dietary components listed in the NICE guidelines were obtained from the British Nutrition Foundation (BNF), Public Health England (PHE) and Diabetes UK, as the NICE guidelines do not specify the recommended values. For participants achieving the recommendation for each dietary component a score of “1” was applied, then differences in total compliance scores, out of ten, between study groups were compared using Kruskal-Wallis tests. Compliance of participants with self-reported Diabetes was present in free sugars, alcohol and semi skimmed milk intakes, however, there were little differences in the other dietary areas compared to participants with unknown Diabetes, Pre-diabetes and participants without Diabetes. Due to limitations in the cross-sectional nature of the NDNS, and substantial missing data, AI could only be used to predict participants based on macro- and micronutrient profiles, and not lifestyle. AI demonstrated excellent capabilities in prediction, with a ~90% accuracy rate. The influence of lifestyle and demographic factors on Diabetes was investigated further using Analysis of Covariance (ANCOVA), as these factors could not be included in the AI study. In ANCOVA, differences in HbA1c, while controlling for demographic and lifestyle variables, between the four study groups was investigated; and revealed age and qualifications gained to be significant. The lack of compliance to dietary guidelines for T2DM management, scarce research into use of AI for prediction of T2DM based on diet and lifestyle, and the identification of suitability of AI in the present thesis, warrant the need to further research AI with longitudinal data and include AI in development of personalised guidelines for T2DM management.