Online Program

Prediction equations for the resting metabolic rate of obese patients based on digital innovations and machine learning

Tuesday, November 3, 2015 : 4:50 p.m. - 5:10 p.m.

Matthew Harding, PhD, Duke-UNC USDA Center for Behavioral Economics and Healthy Eating, Sanford School of Public Policy, Duke University, Duke University, Durham, NC
Terry Hartman, MPH, MS, Duke-UNC USDA Center for Behavioral Economics and Healthy Eating, Sanford School of Public Policy, Duke University, Duke University, Durham, NC
Christine Tenekjian, MPH, Duke Diet & Fitness Center, Duke University Health System, Durham, NC
Elisabetta Politi, MPH, Duke Diet & Fitness Center, Duke University Health System
Ronald Sha, MD, PhD, Duke Diet & Fitness Center, Duke University health System

With a large number of people in the United States being overweight or obese, there is an increasing need for nutritional interventions to alter lifestyles.  In order to lose weight, individuals must consume fewer calories than expended.  Therefore, an accurate prediction of caloric requirements and resting metabolic rate (RMR) is needed to determine appropriate caloric deficiency.

In clinical practice, prediction equations have traditionally been used to estimate RMR and help set nutritional goals.  However, previous research demonstrates significant error when these equations are applied to subpopulations, such as the overweight or obese.  Recent changes in health information technology (HIT) have allowed clinicians to measure RMR to improve nutritional interventions for subpopulations.


The objective of this study was to utilize machine learning to develop better predictive equations to improve patient care for obese and overweight patients.


Currently, the Duke Diet and Fitness Center (DFC) has over 1,500 patients with a measured RMR by handheld indirect calorimetry that is easily and inexpensively administered in the outpatient clinical setting. These data were compiled into a database and machine-learning techniques were utilized to develop predictive clinical equations.


We compared the performance of several leading predictive clinical equations and found that most suffer from significant bias in severely obese patients. Additionally, existing equations suffer from overdispersion so that each patient is described with a large degree of variance.  We developed new equations using advanced machine learning techniques, which can be applied in a clinical setting using a user interface.


The data in this subject population are unique and collected with point of care HIT accessible and relevant to today’s clinical practice model more than alternatives.  The incorporation of predictive models for the obese will have long-lasting implications on public health by allowing clinicians to better treat patients and improve the obesity rate in the US.

Learning Areas:

Clinical medicine applied in public health
Conduct evaluation related to programs, research, and other areas of practice
Public health or related research

Learning Objectives:
Describe current disparities in prediction equations for obese patients. Evaluate the utility of digital innovations in dietary and fitness interventions. Compare effectiveness of predictive equations on clinical outcomes.

Keyword(s): Nutrition, Obesity

Presenting author's disclosure statement:

Qualified on the content I am responsible for because: I have graduate training in public health and biomedical informatics. Additionally, I have worked in clinical and translational research for over 5 years. I'm interested in using informatics to improve patient care and health policy.
Any relevant financial relationships? No

I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.