Childhood obesity (CO) is a pandemic. The USA&UK are predicted to reach the highest levels of obesity in Europe & USA over the next few decades. CO is a particular concern in Wales; rates being highest of the UK. CO is correlated to unprecedented increases in paediatric metabolic associated fatty liver disease (MAFLD) and type-two diabetes (T2DM). Both produce complications faster in children, and MAFLD in particular, is frequently undetected. CO has also been linked to deprivation, constipation and depression. I aim to investigate the possibility of developing a machine learning model to predict T2DM, MAFLD, constipation and depression, using predictors: obesity, deprivation, year-of-birth, and gender. I linked routinely collected children’s general practitioner records with BMI scores in SAIL database, to the Welsh Index of Multiple Deprivation) using the dataset processing language SQL, before exporting them into the statistical programming language R. I performed statistical tests (unpaired Wilcoxon and chi-squared) to identify relationships between the predictors and comorbidities in 2yrs-6yrs, 7yrs-11yrs, & 11-16yrs; and between age-groups; then logistic regression to predict comorbidity risks within and between the three groups. Children with at least one comorbidity in the eldest group had significantly increased BMI and deprivation scores compared with their peers. My model accurately classified 11-to-16-year-old-children with T2DM and MAFLD, proving an excellent predictor of both (AUCs=0.91 & 0.88 respectively); also predicting T2DM, MAFLD, and depression in 11-16yr-olds using their BMI taken 6-11yrs (AUCs=0.95, 0.86 & 0.70); but failed to classify children with constipation (AUC=0.53-0.59 depending on group). Introduction of further risk factors, for example: ethnicity, family history or clinical observations; and a longitudinal retrospective study of older patients with obesity related and inter-related comorbidities, may identify further predictors to improve the model’s accuracy further, making it useful in diagnosis and a valuable a population screening tool, to target costly interventions.
Code is given as five .txt files in 'Project Code.zip' file
One shows SQL code, the rest show R code
I would like to thank my co-supervisors Dr Pramodh Vallabhaneni for helping me identify significant current health concerns in paediatric medicine and for his support and encouragement throughout the project and Dr Arron Lacey for his for his encouragement and support in using SAIL, as well as submitting the SAIL application. I would also like to thank Hamed Ghanbarialadolat, Ian Farr and Kevin Williams, among the rest of the staff at SAIL for their patience and support as I learned to use the database.
This study makes use of anonymised data held in the Secure Anonymised Information Linkage (SAIL) Databank. I would like to acknowledge all the data providers who make anonymised data available for research.
For full text please email me at whmidgley0@gmail.com