Abstract
Background
Comorbidities are common in bariatric patients. However, the degree of comorbidities is hard to quantify objectively as they develop gradually and do not independently reflect the continuum of metabolic syndrome.
Introduction
In the Catharina Hospital bariatric patients are monitored with extensive laboratory panels prior and after surgery. Besides detecting nutrient deficiencies, these parameters offer the opportunity to search for objective markers to describe the health of bariatric patients.
Objectives
The bariatric health index (BHI) is developed enabling quantification and classification of comorbidity in bariatric patients.
Methods
Machine learning is applied to comprehensive laboratory data, collected from 2367 patients containing both pre- and post-surgical data (6, 12 and 24 months). Since comorbidities of interest (diabetes, hypertension, and dyslipidemia), were correlated, an ordinal output variable was defined, stating presence as ‘none’, ‘one’, or ‘multiple’ comorbidities. Different ordinal logistic regression models were fit to the data and compared by AUC.
Results
Next to gender and age at surgery, blood marker levels of HbA1c, triglycerides, urea, potassium, and estimated GFR (CKD-EPI) appeared descriptive for the degree of comorbidity. For the classes ‘none’, ‘one’, or ‘multiple’ the best performing BHI model had an AUC (SE) of 0.82 (0.01), 0.66 (0.01), and 0.88 (0.01), respectively.
Conclusion
A model has been developed by mining bariatric laboratory data that enables quantification and classification of presence of comorbidity. The BHI provides the basis for a tool that predicts the evolution of bariatric health state and may be used to personalize the patient’s monitoring plan.
Comorbidities are common in bariatric patients. However, the degree of comorbidities is hard to quantify objectively as they develop gradually and do not independently reflect the continuum of metabolic syndrome.
Introduction
In the Catharina Hospital bariatric patients are monitored with extensive laboratory panels prior and after surgery. Besides detecting nutrient deficiencies, these parameters offer the opportunity to search for objective markers to describe the health of bariatric patients.
Objectives
The bariatric health index (BHI) is developed enabling quantification and classification of comorbidity in bariatric patients.
Methods
Machine learning is applied to comprehensive laboratory data, collected from 2367 patients containing both pre- and post-surgical data (6, 12 and 24 months). Since comorbidities of interest (diabetes, hypertension, and dyslipidemia), were correlated, an ordinal output variable was defined, stating presence as ‘none’, ‘one’, or ‘multiple’ comorbidities. Different ordinal logistic regression models were fit to the data and compared by AUC.
Results
Next to gender and age at surgery, blood marker levels of HbA1c, triglycerides, urea, potassium, and estimated GFR (CKD-EPI) appeared descriptive for the degree of comorbidity. For the classes ‘none’, ‘one’, or ‘multiple’ the best performing BHI model had an AUC (SE) of 0.82 (0.01), 0.66 (0.01), and 0.88 (0.01), respectively.
Conclusion
A model has been developed by mining bariatric laboratory data that enables quantification and classification of presence of comorbidity. The BHI provides the basis for a tool that predicts the evolution of bariatric health state and may be used to personalize the patient’s monitoring plan.
| Original language | English |
|---|---|
| Pages (from-to) | 652 |
| Number of pages | 1 |
| Journal | Obesity Surgery |
| Volume | 27 |
| Publication status | Published - Jul 2017 |
| Event | 22nd World Congress of the International Federation for the Surgery of Obesity and Metabolic Disorders (IFSO 2017) - London, United Kingdom Duration: 29 Aug 2017 → 2 Sept 2017 |
