Poster Presentation Australian and New Zealand Obesity Society Annual Scientific Conference 2024

AI-driven approaches to identifying baseline characteristics of weight loss in a multidisciplinary metabolic weight management program for people with Class 3 obesity (#221)

Wilson KM Wong 1 2 , Ritesh Chimoriya 2 3 , Pamela Acosta 1 , Pooja S Kunte 1 , Hrishikesh P Hardikar 1 , Nick Kormas 3 , Mugdha V Joglekar 1 , Milan Piya 3 4 , Anandwardhan Hardikar 1 4
  1. School of Medicine, Western Sydney University, Campbelltown, NSW, Australia
  2. Equal , contributors
  3. South Western Sydney Metabolic Rehabilitation and Bariatric Program, Campbelltown and Camden Hospitals, Camden, NSW, 2570
  4. Equal senior and , corresponding authors

Background: It is well-accepted that different components of an individual’s constitution and environment play a role in weight management/loss. This study aimed to use machine learning workflow to identify baseline variables that can be predictive of weight loss over 12 months in a medical weight management program (WMP).

 

Methods: A prospective cohort study was conducted in a publicly funded multidisciplinary WMP in Sydney. A total of n=148 individuals (≥18 years, BMI ≥40 kg/m2), with at least one obesity-related comorbidity and 24 different baseline variables (e.g. age, health record, medication usage, social data and questionnaires), with weight loss outcome at 12 months, were included in this study. Among these participants, n=36 lost >10% weight, while n=112 lost <10% weight at 12 months. Participants between these two groups were randomly split into a train set (n=104; n=25 >10% weight loss) and a test set (remaining participants), and Random Forest machine learning workflow was implemented to identify components predictive of >10% weight loss (on the test set) at 12 months.

 

Results: Baseline mental and physical component scores of quality of life (SF36), age, socio-economic status and excess body weight were amongst the top five baseline variables predictive of >10% weight loss at 12 months, with a receiver operating characteristic AUC of 0.81 (recall =0.96 and F1 score =0.89) on the validation test set. Assessment on an independent validation set is in process.

 

Conclusion: This study identified key baseline components predictive of weight loss at 12 months. The assessment of 24 baseline components, despite variations in clinical records, highlights the potential for broader applicability and exploration in diverse clinical settings.  Future studies to validate these findings in other larger cohorts and assess the value of genetic/molecular biomarkers (e.g. microRNAs) on weight loss prediction are warranted.