Leveraging OGTT derived metabolic features to detect Binge-eating disorder in individuals with high weight: a “seek out” machine learning approach

doi:10.21203/rs.3.rs-4675042/v1

Download PDF

Article

Leveraging OGTT derived metabolic features to detect Binge-eating disorder in individuals with high weight: a “seek out” machine learning approach

https://doi.org/10.21203/rs.3.rs-4675042/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Binge eating disorder (BED) carries a 6 times higher risk for obesity and accounts for roughly 30% of type 2 diabetes cases. Timely identification of early glycemic disturbances and comprehensive treatment can impact on the likelihood of associated metabolic complications and the overall outcome. Machine learning (ML) techniques were trained with static and dynamic glucose-derived measures to detect BED among 281 individuals with high weight. The pipeline (training, validation, testing) was conducted twice, on two different datasets (2 hours, and 5 hours-long glucose load). After features selection, multiple ML algorithms were trained to classify the sample. The best classifier was then applied to an independent cohort (N = 21). A sensitivity-based analysis was run to investigate the relevance of each feature in the classification. 14 features were selected as relevant, with the support vector machine showing the best performance in classifying BED in both models. The model on the 5 hours-long OGTT exhibited the best metrics (sensitivity = 0.75, specificity = 0.67, F score = 0.71) diagnosing BED in 7 out of 10 cases. Sex, HOMA-IR, HbA1c and plasma glucose in different times, and hypoglycemia events were the most sensitive features for BED diagnosis. This study is the first to use metabolic hallmarks to train ML algorithms for detecting BED in individuals at high risk for metabolic complications. ML techniques applied to objective and reliable glycemic features might prompt the identification of BED among individuals at high risk for metabolic complications, enabling timely and tailored multidisciplinary treatment.

Health sciences/Diseases/Psychiatric disorders/Addiction

Health sciences/Biomarkers/Diagnostic markers

BED

binge-eating

OGTT

hypoglycemia

Binge Eating Disorder (BED) is the most prevalent eating disorder (ED) worldwide (1, 2). It is characterized by recurrent episodes of objective binge eating (i.e., consuming large amounts of food in a short period of time) that occur without the physiological sensation of hunger and that are not followed by compensatory behaviors. Core symptoms of the disorder include secrecy, the feeling of losing control overeating, and intense negative emotions such as guilt, shame, and disgust (3).

Far from being solely an “overeating” issue, authors and a plethora of other studies in the last decades have contributed to edge the binge-eating phenomenon towards a broader frame where personality facets, maladaptive schemes, difficulties in understanding, regulating and coping with emotions, impaired decision making, and mood (4–7) remarkably network and account for its considerable complexity (8–11).

In contrast to bulimia nervosa, individuals with BED do not engage in behaviors aimed at weight control (3). Therefore, the untreated illness naturally progress to obesity, the related medical complications (12–14), and other disabling somatic conditions (15, 16).

Findings suggest that individuals with BED suffer a substantial delay in accessing proper treatment, with the longest duration of untreated illness among all EDs (17). The secretive nature of binge eating, denial or minimizing of symptoms, the perceived stigma, and the self-blame, guilt and shame that individuals experience within the disorder, surely constitute an inherent barrier to both diagnosis and treatment (18). On the other hand, many individuals suffering from BED seek care for weight loss or obesity related medical complications rather than for the eating disorder itself (19). De facto, BED is associated in the long-term with an unfavourable metabolic and inflammatory profile compared to non-BED obesity (22), and preliminary findings support that disturbances in the glycemic homeostasis may be detectable at an early stage in these patients (23). Specifically, symptomatic and severe hypoglycemia is more frequently endorsed by this group in the late stages of the extended glucose load. In these settings, individuals rarely are asked about binge-eating (20), and health practitioners within these services may not have the attitude, the training or expertise in suspecting, diagnosing, or addressing an underlying ED(21).

As such, both patients and clinicians related factors, within a plethora of other barriers, prevent early detection and stage-specific intervention, unfavourably affecting the illness trajectory and outstanding the clinical and public healthcare costs (24).

New tools to aid the diagnostic process, shorten the diagnostic delay and the access to treatment, especially for health providers not specialized in the field of mental health and ED, need to be validated and integrated into clinical practice.

Artificial intelligence, increasingly prevalent in the mental health care domain, enables the analysis of big data, assess associations between variables, and identifies the most relevant clinical or instrumental features for the diagnostic and treatment process (25). Machine learning (ML) techniques, a branch of artificial intelligence, have the potential for optimizing prediction of diseases, and have already been suggested for the diagnostic (26–29) and the long-term prediction of outcomes of ED (30, 31), although evidence is still in its infancy (32). Yet, employing algorithms relying solely on psychological variables requires psychometric and/or psychiatric examination, thus confines their use to mental health care settings.

Training ML algorithms with metabolic features known to be altered in BED could allow the prompt detection of at-risk individuals or those already affected by the disorder within a larger user base.

In the present study ML techniques were used to test the ability of glucose metabolism related features to predict BED among individuals seeking care for the treatment of obesity. Currently, no studies have attempted the use of this technique with a similar purpose. We hypothesized that glucose metabolism related features, embedded in the metabolic screening individuals undergo during the diagnostic pathway for obesity, could likely trace individuals at risk or already experiencing BED. ML could constitute an objective, reliable, and reproducible diagnostic strategy for professionals handling obesity and different expertise. Similarly, it could inform screening, identification, and targeted medical management.

Dataset and pre-processing

The dataset for the analysis was gathered from a larger dataset that collects metabolic and psychiatric data from individuals admitted to the diagnostic and therapeutic network of care for obesity (PDTA Obesità) of the University Hospital Renato Dulbecco of Catanzaro (Italy) (eligibility, metabolic and psychiatric examinations in (23)). The PDTA Obesità is a dedicated pathway of care for obesity networked by an interdisciplinary team of specialists (internal medicine, psychiatry, and surgery).

The initial dataset included 313 potential cases registered between May 2017 (start of the recruitment) and June 2023 (date of extraction).

The dataset was filtered for acquisition of plasma glucose levels at multiple time points over a range between 2 and 5 hours after the oral glucose tolerance test (OGTT), no diabetes, and psychiatric examination for BED diagnosis. The study sample included only subjects negative for type 2 diabetes. A set of independent variables known to be altered in BED were chosen to aid classification of group membership. Demographics (age, sex at birth), BMI, OGTT-derived plasma glucose and insulin values, and dynamic measures of insulin secretion/sensitivity/resistance were included in the analysis for classifying participants in BED/non-BED. Additional derived features, such as curve skewness, first maximum peak, and the number of acquisitions above the detectable value have been computed (Table S1). Missing values due to discontinuation of the glucose load (i.e., hypoglycemia) prevented the calculation of OGTT derived measures, resulting in the exclusion of 32 cases.

Data transformation for features scaling and standardization was run with the Standard Scaler algorithm from scikit-learn (33). The final dataset, complete with all the measures for the analysis (Table S1), consisted of 281 cases.

Experimental Pipeline and data analysis

The analysis included three consecutive steps (i.e., training, validation, testing; Fig. 1).

The whole experimental pipeline was firstly conducted on the complete dataset (N = 281, complete acquisition of glucose plasma during the 2 hours long OGTT; Model 1). It was further run considering only cases who completed the extended 5 hours long OGTT (N = 123; Model 2). Accordingly, two ML models were trained, validated, and tested. To guarantee independent and unbiased datasets, the initial dataset was split in three for the training (218 for Model 1, 88 for Model 2), validation (49 for Model 1, 28 for Model 2), and testing steps (14 for Model 1, 7 for Model 2). For the training step, the dataset was further split in the training and testing dataset (80% and 20%, respectively). Feature selection was performed by running the Recursive Feature Elimination method (RFE) (Fig. 1-a3) on 150 different folds. RFE algorithm works by iterative removal of irrelevant or redundant features until a desired number of informative features is reached (34). The set of features that survived the RFE in each fold were recorded to build a global ranking list, from which the 15 most frequent ones were finally selected.

This subset of selected features was then used to train five different ML algorithms, both linear and non-linear, to classify the sample in BED/non-BED. ML algorithms included Decision Tree (DT) (35), Random Forrest Classifier (RFC) (36), Extra Tree (ET) (37), Support Vector Machine Classifier (SVC) (38), Logistic Regression (LR) (39) (more details in Table S2; Fig. 1-a4). Also in this case, to ensure a robust selection of the best classifier, this procedure was repeated 150 times, randomly shuffling the cases included into the training/testing datasets (red box in Fig. 1-A).

The final dataset included 281 cases (133 BED, 148 non-BED), all White. Respectively, mean age was 40.2 ± 13.1 and 44.4 ± 11.7, and females were overrepresented in both groups (85.7%, 67.6%). Non-BED cases exhibited lower BMI than BED cases (41.4 ± 7.7; 39.1 ± 7.4). RFE dropped several items from the models, providing evidence for 15 significant features for both Model 1 and Model 2 (Table 1). Sex, BMI, plasma glucose at the 120’ min, maximum glucose value during the load, hypoglycemia events, Area Under Curve for Insulin and Glucose (AUC Ins-Glu during the 0- to 30-min of the OGTT), skewness of the curve, and three indexes of insulin sensitivity [Gutt’s insulin sensitivity index (Gutt-ISI), Stumvoll index and HOMA-IR] were depicted in both models as significant in discerning BED over non-BED (Table 1).

Table 1

Performance of the best classifier on the validation set. Results for the Model 1 and 2 .
	Best Classifier	Cases	Sensitivity	Specificity	F₁ score	Accuracy	Features
Model 1	SVCP	49	0.5	0.8	0.58	0.67	AUC_Ins−glu 0–30 AUC_glu BMI G₃₀, G₆₀, G₉₀, G₁₂₀ GUTT Hb1AC HOMA-IR Hypo tot Max_Gluc Sex Skewness Stumvoll ISI
Model 2	SVCP	28	0.86	0.43	0.71	0.64	AUC_Ins−glu 0–30 BMI G₆₀, G₁₂₀, G₁₅₀, G₂₁₀, G₂₇₀, G₃₀₀ GUTT HOMA-IR Hypo tot Max_Gluc Sex Skewness Stumvoll ISI
SVCP: support vector machine classifier; BMI: body mass index; HOMA-IR: homeostatic model assessment insulin resistance; ISI: insulin sensitivity index.

Table 2

Testing. Results for the two models.
Model	Cases	Sensitivity	Specificity	F₁ score	Accuracy	TP	TN	FP	FN
Model 1	14	0.35	0.67	0.46	0.5	3	4	2	5
Model 2	7	0.75	0.67	0.67	0.71	2	3	1	1
Model 1 tested on cases with complete 2 hours long OGTT; Model 2 tested on cases with complete 5 hours long OGTT.

Across all five ML algorithms, the SVCP outperformed in classifying the groups. For Model 1 (2 hours long OGTT), the classifier correctly identified half of the cases with BED (sensitivity = 0.5), and 80% of non-BED cases (specificity = 0.8) (Table 1). On the 5 hours long OGTT dataset (Model 2), the classifier was more sensitive (0.86), but less specific (0.43) (Table 1). Accuracy was similar for the two Models (0.67 vs 0.64).

Figures S1-2 show the sensitivity-based analysis conducted on the validation set and for all the classifiers reported in Table S2. Sex, HOMA-IR, Hb1AC, and skewness of the glucose curve scored higher in the Model 1 classification ability. Plasma glucose at the 120’ minute was the most accurate feature of the Model 2. Table 2 reports the performance of the proposed two models tested on the testing cohort. Model 2 exceeded the performance of Model 1 in terms of sensitivity (0.7), specificity (0.67), accuracy (0.71), and the overall ability to predict class membership (0.67).

Present study used a ML approach and data from individuals accessing the care for obesity and no diabetes to identify individual-level metabolic features that could be associated with the diagnosis of BED. It further aimed to investigate the predictive diagnostic accuracy of a pure metabolic based ML approach in screening BED among seekers care for obesity.

Results suggest that sex, BMI, and glucose metabolism-related variables such as glucose levels at specific times of the OGTT, skewness of the glucose load, insulin sensitivity indexes, and hypoglycemia events could be crucial when it comes to identifying those individuals suffering from BED among patients with obesity.

Cross-sectional studies in population with no diabetes have demonstrated that binge-eating associates with fasting hyperglycemia, insulin resistance, and higher frequency of pre-diabetes phenotypes (e.g., impaired glucose tolerance)(23, 42–44). Hypoglycemia events occurring with neurogenic or neuroglycopenic symptoms and typically reverting with carbohydrates intake (i.e., reactive hypoglycemia)(45) have been recently studied and preliminarily associated to binge-eating during an extended laboratory stimulation, suggesting that individuals suffering from BED, obesity, but no diabetes would suffer from more frequent and more severe hypoglycemia events with respect to the counterpart with no ED (23). More consistently, longitudinal studies have showed that binge-eating contributes to higher odds of metabolic syndrome components in the long-term beyond the risk attributable to obesity alone (46, 47).

Present results confirm, but more importantly, leverage the relevance of these metabolic correlates in their potentiality to specifically cluster BED vs non-BED in non-psychiatric settings. Overall, this study suggests that modeling OGTT-related metabolic features together with demographics and anthropometrics could assist the diagnostic process, potentially identifying BED in five to seven out of ten of cases, depending on data collection. Specifically, the Models performed similarly in classifying cases not at risk (67%), but the 5 hours long OGTT based Model outperformed in terms of sensitivity (75%), accuracy (71%), and overall ability to predict class membership (0.67 vs 0.47). The relative weight of hypoglycemia in the model, together with the evidence that individuals with BED would experience more hypoglycemia events during the laboratory test at the latest stages of the glucose load (4th -5th hour), could have contributed to the better performance of Model 2 over Model 1. The diagnostic accuracy of the models is surprisingly encouraging considering the diagnostic accuracy the specifically tailored psychometric instruments exhibit. A recent systematic review compared the performances of the most widely used screening questionnaires in detecting BED in overweight or obesity and found high variability across studies (48). Only moderate accuracy was found for the Eating Disorders Questionnaire (EDE-Q) (sensibility: 0.40–0.87; specificity: 0.62-1), and the Binge Eating Scale (BES) (sensibility: 0.51–0.98; specificity: 0.48–0.76), still considered the most widely used psychometrics tools in the field.

Further, the ML algorithms hereby used to detect BED as a psychiatric disorder were exclusively fed with glycemic/insulinemic and other non-psychological features. Authors may, accordingly, speculate that adding psychometric tools specifically designed for BED and tailored to overweight and obesity (e.g., the Eating Behaviors Assessment for Obesity,(49)) to the training could potentially booster the diagnostic accuracy of the present model.

Other studies have previously tested ML techniques to evaluate the risk, assist the investigation, or predict the outcome of BED (26, 29, 31, 50). Eric Stice and Christopher D. Desjardins in 2018 used a multiple approach-based classification tree analysis to investigate the interactions of risk factors in the prediction of the onset of EDs in a large dataset of individuals at high risk. The onset of BED was accurately predicted in 64% of cases by a complex four-way interaction between body dissatisfaction, overeating, dieting, and thin-ideal internalization (29). In the investigation of Linardon and colleagues, a ML-based decision tree classification analysis captured 70% of binge-eaters by examining eating behaviors and cognitions. The two-way interaction of low intuitive eating and high dichotomous thinking resulted largely associated with being classified with recurrent binge eating (84% incidence)(26). Lastly, Forrest L.N. tested two ML approaches to evaluate treatment outcomes in a sample of individuals suffering from BED randomized to six months behavioural or stepped-care treatments. Authors concluded for a slight advantage of the ML over the traditional models, claiming for further empirical confirmation in the EDs literature (31).

Still recognizing the relevance of ML-based investigations based on psychological measures, they require specific training and specialized settings for the analysis.

BED goes undiagnosed for many years and suffers from the greatest delay to proper treatments of all EDs (17). This is partly due to either patient or clinician-related factors (18, 21), but surely contributes to the great medical burden and overall disability BED associates with. Further, treatment outcomes often deviate from the expectations (51, 52).

The highest prevalence of BED is found in clinical samples attending weight loss programs or bariatric surgery (53, 54). Training ML algorithms with objective measures embedded within the medical screening individuals suffering from BED could undergo in these settings, could enable a prompt detection in non-specialized settings, would generate more targeted interventions and anticipate clinical progression.

To our knowledge, no similar evidence exists so far. The current study is the first that has attempted training ML to diagnose BED basing on objective measures (OGTT-derived features), pioneering the field of metabolic based AI applied to BED. With respect to other studies that used self-report binge episodes to corroborate the diagnosis (26), BED was diagnosed clinically, thereby enhancing the accuracy of the results. Further analyses on larger datasets and considering other valuable biomarkers could enhance modeling performances and translate to clinical practice its use to inform clinical decisions.

Within these strengths, some limitations should be addressed. Embedded within the methodology used, no information can be extracted about the nature of associations between the relevant features and BED class (e.g., higher Hb1Ac in BED vs non-BED). Further, authors acknowledge that the cross-sectional design prevent to infer any causal or temporal association between variables. Although distinct methods were applied to avoid overfitting (i.e., scaling, splitting, features selection with the RFE), an inflation in model accuracy due to the sample size cannot be excluded. The sample was all White, and not gender balanced. Accordingly, results may not be generalizable to samples with other characteristics.

Finally, authors selected for the analysis the most informative metabolic features so far associated with BED and known to constitute early markers of late metabolic disruption (e.g., diabetes, metabolic syndrome); still, we recognize that the use of a different set of features or more advanced statistical methodology could produce distinct and perhaps more significant results.

Competing interests

The authors declare that they have no competing interests.

Ethics approval and consent to participate

Eligible patients were fully informed about the aim and the procedures of the study, and that participation was voluntary and free from any compensation. All participants gave written informed consent. The study was approved by the institutional review boards (Local Ethical Committee, n. 53/2013), and all the procedures were performed in accordance with the principles of the Declaration of Helsinki.

Funding

No fundings were received for this study.

Authors' contributions

statement (CRediT)

Acknowledgements

Not applicable.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Qian J, Wu Y, Liu F, Zhu Y, Jin H, Zhang H, Wan Y, Li C, Yu D. An update on the prevalence of eating disorders in the general population: a systematic review and meta-analysis. Eat Weight Disord (2022) 27:415–428. doi:10.1007/S40519-021-01162-Z
Kessler RC, Berglund PA, Chiu WT, Deitz AC, Hudson JI, Shahly V, Aguilar-Gaxiola S, Alonso J, Angermeyer MC, Benjet C, et al. The prevalence and correlates of binge eating disorder in the World Health Organization World Mental Health Surveys. Biol Psychiatry (2013) 73:904–914. doi:10.1016/j.biopsych.2012.11.020
American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders (DSM-5®). American Psychiatric Pub (2013).
Aloi M, Rania M, Caroleo M, Carbone EA, Fazia G, Calabrò G, Segura-Garcia C. How are early maladaptive schemas and DSM-5 personality traits associated with the severity of binge eating? J Clin Psychol (2020) 76: doi:10.1002/jclp.22900
Rania M, Aloi M, Caroleo M, Carbone EA, Fazia G, Calabrò G, de Filippis R, Staltari F, Segura-Garcia C. ‘Impaired Autonomy and Performance’ predicts binge eating disorder among obese patients. Eat Weight Disord (2020) 25: doi:10.1007/s40519-019-00747-z
Arexis M, Feron G, Brindisi MC, Billot PÉ, Chambaron S. A scoping review of emotion regulation and inhibition in emotional eating and binge-eating disorder: what about a continuum? J Eat Disord (2023) 11:197. doi:10.1186/S40337-023-00916-7
Aloi M, Rania M, Caroleo M, De Fazio P, Segura-García C. Social Cognition and Emotional Functioning in Patients with Binge Eating Disorder. Eur Eat Disord Rev (2017) 25: doi:10.1002/erv.2504
Aloi M, Rania M, Carbone EA, Calabrò G, Caroleo M, Carcione A, Nicolò G, Semerari A, Segura-Garcia C. The role of self-monitoring metacognition sub-function and negative urgency related to binge severity. Eur Eat Disord Rev (2020) 28:580–586. doi:10.1002/erv.2742
Aloi M, Rania M, Carbone EA, Caroleo M, Calabrò G, Zaffino P, Nicolò G, Carcione A, Coco G Lo, Cosentino C, et al. Metacognition and emotion regulation as treatment targets in binge eating disorder: a network analysis study. J Eat Disord (2021) 9: doi:10.1186/S40337-021-00376-X
Dingemans AE, Son GE, Vanhaelen CB, Furth EF. Depressive symptoms rather than executive functioning predict group cognitive behavioural therapy outcome in binge eating disorder. Eur Eat Disord Rev (2020) 28:620–632. doi:10.1002/erv.2768
Rania M, Aloi M, de Filippis R, Carbone EA, Caroleo M, De Fazio P, Segura-Garcia C. Executive functions and depressive symptoms interplay in binge eating disorder: A structural equation model analysis. Eur Eat Disord Rev (2021) 29:811–819. doi:10.1002/ERV.2854
Raevuori A, Suokas J, Haukka J, Gissler M, Linna M, Grainger M, Suvisaari J. Highly increased risk of type 2 diabetes in patients with binge eating disorder and bulimia nervosa. Int J Eat Disord (2015) 48:555–562. doi:10.1002/EAT.22334
Hudson JI, Lalonde JK, Coit CE, Tsuang MT, McElroy SL, Crow SJ, Bulik CM, Hudson MS, Yanovski JA, Rosenthal NR, et al. Longitudinal study of the diagnosis of components of the metabolic syndrome in individuals with binge-eating disorder. Am J Clin Nutr (2010) doi:10.3945/ajcn.2010.29203
Solmi F, Moreno AB, Lewis G, Angélica Nunes M, de Jesus Mendes da Fonseca M, Harter Griep R. Longitudinal association between binge eating and metabolic syndrome in adults: Findings from the ELSA-Brasil cohort. Acta Psychiatr Scand (2021) 144:464–474. doi:10.1111/ACPS.13356
Thornton LM, Watson HJ, Jangmo A, Welch E, Wiklund C, von Hausswolff-Juhlin Y, Norring C, Herman BK, Larsson H, Bulik CM. Binge-eating disorder in the Swedish national registers: Somatic comorbidity. Int J Eat Disord (2017) 50:58–65. doi:10.1002/EAT.22624
Wassenaar E, Friedman J, Mehler PS. Medical Complications of Binge Eating Disorder. Psychiatr Clin North Am (2019) 42:275–286. doi:10.1016/J.PSC.2019.01.010
Austin A, Flynn M, Richards K, Hodsoll J, Duarte TA, Robinson P, Kelly J, Schmidt U. Duration of untreated eating disorder and relationship to outcomes: A systematic review of the literature. Eur Eat Disord Rev (2021) 29:329–345. doi:10.1002/ERV.2745
Bremer MF, Garnweidner-Holme L, Nesse L, Molin M. Experiences of living with binge eating disorder and facilitators of recovery processes: a qualitative study. J Eat Disord (2023) 11:1–8. doi:10.1186/S40337-023-00929-2/FIGURES/1
Kornstein SG. Epidemiology and Recognition of Binge-Eating Disorder in Psychiatry and Primary Care. J Clin Psychiatry (2017) 78 Suppl 1:3–8. doi:10.4088/JCP.SH16003SU1C.01
Mond JM, Myers TC, Crosby RD, Hay PJ, Mitchell JE. Bulimic eating disorders in primary care: hidden morbidity still? J Clin Psychol Med Settings (2010) 17:56–63. doi:10.1007/S10880-009-9180-9
Mills R, Hyam L, Schmidt U. A Narrative Review of Early Intervention for Eating Disorders: Barriers and Facilitators. Adolesc Health Med Ther (2023) 14:217–235. doi:10.2147/AHMT.S415698
Giel KE, Bulik CM, Fernandez-Aranda F, Hay P, Keski-Rahkonen A, Schag K, Schmidt U, Zipfel S. Binge eating disorder. Nat Rev Dis Prim (2022) 8:16. doi:10.1038/S41572-022-00344-Y
Rania M, Caroleo M, Carbone EA, Ricchio M, Pelle MC, Zaffina I, Condoleo F, de Filippis R, Aloi M, De Fazio P, et al. Reactive hypoglycemia in binge eating disorder, food addiction, and the comorbid phenotype: unravelling the metabolic drive to disordered eating behaviours. J Eat Disord (2023) 11: doi:10.1186/S40337-023-00891-Z
Watson HJ, Jangmo A, Smith T, Thornton LM, von Hausswolff-Juhlin Y, Madhoo M, Norring C, Welch E, Wiklund C, Larsson H, et al. A register-based case-control study of health care utilization and costs in binge-eating disorder. J Psychosom Res (2018) 108:47–53. doi:10.1016/J.JPSYCHORES.2018.02.011
Iyortsuun NK, Kim SH, Jhon M, Yang HJ, Pant S. A Review of Machine Learning and Deep Learning Approaches on Mental Health Diagnosis. Healthcare (2023) 11: doi:10.3390/HEALTHCARE11030285
Linardon J, Messer M, Helms ER, McLean C, Incerti L, Fuller-Tyszkiewicz M. Interactions between different eating patterns on recurrent binge-eating behavior: A machine learning approach. Int J Eat Disord (2020) 53:533–540. doi:10.1002/EAT.23232
Levinson CA, Trombley CM, Brosof LC, Williams BM, Hunt RA. Binge Eating, Purging, and Restriction Symptoms: Increasing Accuracy of Prediction Using Machine Learning. Behav Ther (2023) 54:247–259. doi:10.1016/J.BETH.2022.08.006
Ren Y, Lu C, Yang H, Ma Q, Barnhart WR, Zhou J, He J. Using machine learning to explore core risk factors associated with the risk of eating disorders among non-clinical young women in China: A decision-tree classification analysis. J Eat Disord (2022) 10: doi:10.1186/S40337-022-00545-6
Stice E, Desjardins CD. Interactions between risk factors in the prediction of onset of eating disorders: Exploratory hypothesis generating analyses. Behav Res Ther (2018) 105:52–62. doi:10.1016/J.BRAT.2018.03.005
Haynos AF, Wang SB, Lipson S, Peterson CB, Mitchell JE, Halmi KA, Agras WS, Crow SJ. Machine learning enhances prediction of illness course: a longitudinal study in eating disorders. Psychol Med (2021) 51:1392. doi:10.1017/S0033291720000227
Forrest LN, Ivezaj V, Grilo CM. Machine learning v. traditional regression models predicting treatment outcomes for binge-eating disorder from a randomized controlled trial. Psychol Med (2023) 53:2777–2788. doi:10.1017/S0033291721004748
Fardouly J, Crosby RD, Sukunesan S. Potential benefits and limitations of machine learning in the field of eating disorders: current research and future directions. J Eat Disord (2022) 10:66. doi:10.1186/S40337-022-00581-2
Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Müller A, Nothman J, Louppe G, et al. Scikit-learn: Machine Learning in Python. (2012) Available at: http://arxiv.org/abs/1201.0490 [Accessed February 28, 2024]
Chen X, Jeong JC. Enhanced recursive feature elimination. in Sixth International Conference on Machine Learning and Applications (ICMLA 2007), 429–435. doi:10.1109/ICMLA.2007.35
Rokach L, Maimon O. Data Mining with Decision Trees: Theory and Applications, 2nd Edition. Data Min with Decis Trees Theory Appl 2nd Ed (2014) 81:1–305. doi:10.1142/9097/SUPPL_FILE/9097_CHAP01.PDF
Pal M. Random forest classifier for remote sensing classification. Int J Remote Sens (2005) 26:217–222. doi:10.1080/01431160412331269698
Abhishek L. Optical Character Recognition using Ensemble of SVM, MLP and Extra Trees Classifier. in 2020 International Conference for Emerging Technology (INCET), 1–4. doi:10.1109/INCET49848.2020.9154050
Suthaharan S. Machine Learning Models and Algorithms for Big Data Classification. (2016) 36: doi:10.1007/978-1-4899-7641-3
Nusinovici S, Tham YC, Chak Yan MY, Wei Ting DS, Li J, Sabanayagam C, Wong TY, Cheng CY. Logistic regression was as good as machine learning for predicting major chronic diseases. J Clin Epidemiol (2020) 122:56–69. doi:10.1016/J.JCLINEPI.2020.03.002
van Rossum G. Python reference manual. (1995)
McKinney W. Python for Data Analysis: Data Wrangling with Pandas, NumPy, and IPython. (2012)466. Available at: https://books.google.com/books?id=v3n4_AK8vu0C&pgis=1 [Accessed February 28, 2024]
Mehler PS, Frank GKW, Mitchell JE. Medical comorbidity and medical complications associated with binge-eating disorder. Int J Eat Disord (2016) 49:319–323. doi:10.1002/EAT.22452
Succurro E, Segura-Garcia C, Ruffo M, Caroleo M, Rania M, Aloi M, De Fazio P, Sesti G, Arturi F. Obese patients with a binge eating disorder have an unfavorable metabolic and inflammatory profile. Med (United States) (2015) 94: doi:10.1097/MD.0000000000002098
Chao A, Grey M, Whittemore R, Reuning-Scherer J, Grilo CM, Sinha R. Examining the mediating roles of binge eating and emotional eating in the relationships between stress and metabolic abnormalities. J Behav Med (2016) 39:320. doi:10.1007/S10865-015-9699-1
Kittah NE, Vella A. MANAGEMENT OF ENDOCRINE DISEASE: Pathogenesis and management of hypoglycemia. Eur J Endocrinol (2017) 177:R37–R47. doi:10.1530/EJE-16-1062
Hudson JI, Lalonde JK, Coit CE, Tsuang MT, McElroy SL, Crow SJ, Bulik CM, Hudson MS, Yanovski JA, Rosenthal NR, et al. Longitudinal study of the diagnosis of components of the metabolic syndrome in individuals with binge-eating disorder. Am J Clin Nutr (2010) 91:1568–1573. doi:10.3945/ajcn.2010.29203
Abraham TM, Massaro JM, Hoffmann U, Yanovski JA, Fox CS. Metabolic characterization of adults with binge eating in the general population: the Framingham Heart Study. Obesity (Silver Spring) (2014) 22:2441–9. doi:10.1002/oby.20867
House ET, Lister NB, Seidler AL, Li H, Ong WY, McMaster CM, Paxton SJ, Jebeile H. Identifying eating disorders in adolescents and adults with overweight or obesity: A systematic review of screening questionnaires. Int J Eat Disord (2022) 55:1171–1193. doi:10.1002/EAT.23769
Segura-Garcia C, Aloi M, Rania M, de Filippis R, Carbone EA, Taverna S, Papaianni MC, Liuzza MT, De Fazio P. Development, validation and clinical use of the Eating Behaviors Assessment for Obesity (EBA-O). Eat Weight Disord (2022) 27:2143–2154. doi:10.1007/s40519-022-01363-0
Raab D, Baumgartl H, Buettner R. Machine Learning Based Diagnosis of Binge Eating Disorder Using EEG Recordings.2020.
Bardone-Cone AM, Alvarez A, Gorlick J, Koller KA, Thompson KA, Miller AJ. Longitudinal follow-up of a comprehensive operationalization of eating disorder recovery: Concurrent and predictive validity. Int J Eat Disord (2019) 52:1052–1057. doi:10.1002/EAT.23128
Salvia MG, Ritholz MD, Craigen KLE, Quatromoni PA. Women’s perceptions of weight stigma and experiences of weight-neutral treatment for binge eating disorder: A qualitative study. EClinicalMedicine (2022) 56: doi:10.1016/J.ECLINM.2022.101811
De Zwaan M. Binge eating disorder and obesity. Int J Obes (2001) 25:S51–S55. doi:10.1038/sj.ijo.0801699
Mitchell JE, King WC, Pories W, Wolfe B, Flum DR, Spaniolas K, Bessler M, Devlin M, Marcus MD, Kalarchian M, et al. Binge eating disorder and medical comorbidities in bariatric surgery candidates. Int J Eat Disord (2015) 48:471–476. doi:10.1002/EAT.22389

The authors have declared there is NO conflict of interest to disclose

Supplementarymaterial.docx

Download PDF

Review #1 received at journal
30 Sep, 2024
Review #2 received at journal
27 Sep, 2024
Reviewer #2 agreed at journal
25 Sep, 2024
Reviewer #1 agreed at journal
09 Sep, 2024
Reviewers invited by journal
04 Sep, 2024
Submission checks completed at journal
05 Jul, 2024
First submitted to journal
04 Jul, 2024
Unknown event
03 Jul, 2024
Editor assigned by journal
02 Jul, 2024

You are reading this latest preprint version

Leveraging OGTT derived metabolic features to detect Binge-eating disorder in individuals with high weight: a “seek out” machine learning approach

Status:

Version 1

Abstract

Introduction

Methods and Materials

Dataset and pre-processing

Experimental Pipeline and data analysis

Results

Discussion

Declarations

Competing interests

Ethics approval and consent to participate

Funding

Authors' contributions

Acknowledgements

Availability of data and materials

References

Additional Declarations

Supplementary Files

Status:

Version 1