ESPE2021 ePoster Category 2 Fat, metabolism and obesity (59 abstracts)
Machine Learning Quest for Predictive Markers of Lifestyle Modification Outcomes in Pediatric Obesity Treatment
Aneta Gawlik 1 , Michael Shmoish 2 , Abdullah Bereket 3 , Malgorzata Wasniewska 4 , Aleksandra Antosz 1 , Tarık Kırkgoz 3 , Serap Turan 3 , Tulay Guran 3 , Tommaso Aversa 4 , Domenico Corica 4 , Stefan A. Wudy 5 , Michaela F. Hartmann 5 , Katarzyna Gruszczynska 6 & Ze’ev Hochberg 7
1Department of Pediatrics and Pediatric Endocrinology, Faculty of Medical Sciences, Medical University of Silesia, Katowice, Poland; 2Bioinformatics Knowledge Unit, Lorry I. Lokey Interdisciplinary Center for Life Sciences and Engineering, Technion-Israel Institute of Technology, Haifa, Israel; 3Division of Pediatric Endocrinology, Marmara University, School of Medicine, Istanbul, Turkey; 4Department of Human Pathology of Adulthood and Childhood University of Messina, Messina, Italy; 5Steroid Research and Mass Spectrometry Unit, Division of Pediatric Endocrinology and Diabetology, Center of Child and Adolescent Medicine, Justus Liebig University, Giessen, Germany; 6Department of Diagnostic Imaging, School of Medicine in Katowice, Medical University of Silesia, Katowice, Poland; 7Faculty of Medicine, Technion- Israel Institute of Technology, Haifa, Israel
Context: The traditional approach to childhood obesity management is lifestyle modification/LSM. Nevertheless, the response rate is variable and difficult to predict.
Aim: A systematic search for markers to predict outcomes of simple LSM in pediatric obesity management.
Patients/Methods: Out of 240 children with obesity (BMI>97%), recruited to a prospective ‘multi-OMICS’ study granted by ESPE Research Unit, 159 subjects (age 8-17 yrs, median 12.8 yrs; 45% females) finished twelve-months of LSM obesity management at three clinical centers in three counties. Their baseline (V0) phenotype was precisely described with more than 180 clinical and laboratory features grouped as markers of general description, body composition/BC, family and patient’s history, lifestyle/LS and socioeconomic status/SES, insulin resistance/IR, liver diseases/LD, metabolic syndrome/MetS, steroid metabolome and gut microbiome. Additional 150 features were measured at V3/V6/V12 months. Machine learning technique/CART as implemented in ’rpart’ R-package was applied to build decision trees to automatically identify the combination of markers and their cut-offs with the strongest correlation to a “success” of LSM, defined as a decrease in z-score BMIV12-V0. Results: 118 out of 159 (74.2%) participants were classified as responders to LMS. When built on the IR & LD features, a decision tree pointed to a strongest role of the following parameters: acanthosis nigricans/AN, resistin levels, glucose 120’, NAFLDUSG and insulin/glucose ratio. The AN feature was significantly associated with the response to LSM (OR 2.75; P = 0.0106), where the lack/presence of AN predicted success in 84%/65% cases, respectively. When the lack of AN was observed simultaneously with resistin value <16 ng/ml the response rate grew to 91% (OR 9.05; P = 0.0026), while the presence of AN with glucose 120’≥136 mg/dl predicted 93% of the successful outcomes (OR 8.51; P = 0.0281). On the other extreme, high insulin/glucose ratio ≥0.34 (with the presence of AN together with NAFLDUSG, and with glucose 120’<136 mg/dl), decreased the success rate to 23% (OR 0.20; P = 0.0452). Out of SES&LS features at V0, a small (<6.6hrs) number of sleep hours on schooldays /high frequency of sweet beverages were found to be associated with the lowest chance for success (OR 0.15/0.34, P = 0.0023/0.013, respectively).
Conclusions: Insulin resistance features, history of inappropriate sleep or beverages consumption before intervention are significantly associated with failure of LSM in childhood obesity. Pending the validation on an independent cohort, our findings suggest the predictive role of these markers.
Article tools
My recent searches
My recently viewed abstracts
Authors
ESPE Abstracts
© Bioscientifica 2024 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more