Home  |  About  |  Contact  |  Join MCACS  Donate to MCACS
Massachusetts Chapter of the American College of Surgeons Massachusetts Chapter of the American College of Surgeons Massachusetts Chapter of the American College of Surgeons Massachusetts Chapter of the American College of Surgeons Massachusetts Chapter of the American College of Surgeons
57th Annual Meeting Abstracts


Abstract 18

Abstract Title

Application of Artificial Neural Networks and Decision Trees to Predict Weight Loss following Laparoscopic Adjustable Gastric Banding

 

Author Block

Ali Ardestani, MD, David B. Lautz, MD, Ali Tavakkolizadeh, MD

Brigham and Women's Hospital/Harvard Medical School, Boston, MA

 

Abstract Body

BACKGROUND: Laparoscopic Adjustable Gastric Banding (LAGB) has become a popular surgical approach for weight loss.  Although most patients do well with the procedure, some fail to achieve the desired weight loss. Identifying these patients early, may help physicians develop targeted strategies to help them.  Such predictive algorithms currently do not exist. We set out to develop such a predictive algorithm, using two different models, one based on neural networks, and other on decision trees.

METHODS: We collected demographic, weight loss data, as well as clinic attendance patterns of patients who had LAGB from 2004 to 2007.  Weight loss outcomes were categorized into Good, Average and Poor (EBWL>50%; 50%>=EBWL>30%; and 30%>=EBWL, respectively). Two different sets of models were created using artificial neural networks and decision trees to help categorize patients based on EBWL at 1-year and 2-years.

RESULTS: 325 patients were identified. Using the 6-months data, the decision tree and neural network models were able to predict 1-year EBWL with 74.6% and 72.4% accuracy, respectively. The one year follow-up data was then used to predict EBWL at 2-years. The decision tree and neural network models predicted 2-year EBWL classification with 83.1% and 76.9% accuracy, respectively.

CONCLUSIONS: Our results demonstrate that mathematical modeling based on artificial intelligence can have high predictive capabilities. Since these models are dynamic, they can be integrated into applications (e.g. bariatric databases) to facilitate the decision making process for clinicians during the follow-up. Such algorithms can be developed to predict surgical outcomes in other clinical settings.

 

 

 Home | About | Contact