Regional variations in medical trainee diet and nutrition counseling competencies: Machine learning-augmented propensity score analysis of a prospective multi-site cohort study

Anish Patnaik; Justin Tran; John W McWhorter; Helen Burks; Alexandra Ngo; Tu Dan Nguyen; Avni Mody; Laura Moore; Deanna M Hoelscher; Amber Dyer; Leah Sarris; Timothy Harlan; C Mark Chassay; Dominique Monlezun

doi:10.1007/s40670-020-00973-6

Regional variations in medical trainee diet and nutrition counseling competencies: Machine learning-augmented propensity score analysis of a prospective multi-site cohort study

Med Sci Educ. 2020 May 20;30(2):911-915. doi: 10.1007/s40670-020-00973-6. eCollection 2020 Jun.

Authors

Anish Patnaik¹, Justin Tran¹, John W McWhorter^{2

3}, Helen Burks¹, Alexandra Ngo¹, Tu Dan Nguyen¹, Avni Mody¹, Laura Moore^{2

3}, Deanna M Hoelscher^{2

3}, Amber Dyer⁴, Leah Sarris⁴, Timothy Harlan⁴, C Mark Chassay¹, Dominique Monlezun^{1

4}

Affiliations

¹ McGovern Medical School, University of Texas Health Sciences Center at Houston (UTHealth), Houston, TX USA.
² School of Public Health, University of Texas Health Sciences Center at Houston (UTHealth), Houston, TX USA.
³ Michael & Susan Dell Center for Healthy Living, Austin, TX USA.
⁴ The Goldring Center for Culinary Medicine, Tulane University School of Medicine, New Orleans, LA USA.

Abstract

Background: Medical professionals and students are inadequately trained to respond to rising global obesity and nutrition-related chronic disease epidemics, primarily focusing on cardiovascular disease. Yet, there are no multi-site studies testing evidence-based nutrition education for medical students in preventive cardiology, let alone establishing student dietary and competency patterns.

Methods: Cooking for Health Optimization with Patients (CHOP; NIH NCT03443635) was the first multi-national cohort study using hands-on cooking and nutrition education as preventive cardiology, monitoring and improving student diets and competencies in patient nutrition education. Propensity-score adjusted multivariable regression was augmented by 43 supervised machine learning algorithms to assess students outcomes from UT Health versus the remaining study sites.

Results: 3,248 medical trainees from 20 medical centers and colleges met study criteria from 1 August 2012 to 31 December 2017 with 60 (1.49%) being from UTHealth. Compared to the other study sites, trainees from UTHealth were more likely to consume vegetables daily (OR 1.82, 95%CI 1.04-3.17, p=0.035), strongly agree that nutrition assessment should be routine clinical practice (OR 2.43, 95%CI 1.45-4.05, p=0.001), and that providers can improve patients' health with nutrition education (OR 1.73, 95%CI 1.03-2.91, p=0.038). UTHealth trainees were more likely to have mastered 12 of the 25 competency topics, with the top three being moderate alcohol intake (OR 1.74, 95%CI 0.97-3.11, p=0.062), dietary fats (OR 1.26, 95%CI 0.57-2.80, p=0.568), and calories (OR 1.26, 95%CI 0.70-2.28, p=0.446).

Conclusion: This machine learning-augmented causal inference analysis provides the first results that compare medical students nationally in their diets and competencies in nutrition education, highlighting the results from UTHealth. Additional studies are required to determine which factors in the hands-on cooking and nutrition curriculum for UTHealth and other sites produce optimal student - and, eventually, preventive cardiology - outcomes when they educate patients in those classes.

Keywords: Machine learning; medical education; medical student; nutrition; public health.