Article Image

ENDO 2021 Preview: Harnessing Data to Help With Diabetes

Op-Med is a collection of original articles contributed by Doximity members.

A talk with Mark A. Clements, MD, PhD about ENDO Abstract Big Data Assessment to Deliver Best Care

What is the central question that your study and/or presentation tries to answer?

Diabetes care generates abundant data. Adults and youth with diabetes use devices like insulin pumps, smart insulin pens, glucometers, and continuous glucose monitors in their daily self-management. Many of these devices stream data passively to cloud data systems on a daily basis; others can be synchronized to those systems by users with little effort. Clinicians, educators, and other staff record information about therapeutic modalities, self-care habits, vital signs and physical measures, comorbid conditions, physical signs and symptoms, laboratory results, detailed descriptions of daily glycemic control, and other clinical observations. Despite the depth and breadth of data produced during diabetes care, few diabetes centers use data to engage in population management. Fewer still use data to forecast outcomes or to quality-improve care. I present two examples in which clinical data are being used to quality-improve diabetes care: the T1D Exchange Quality Improvement Collaborative, a national learning health system, and a new diabetes rapid learning lab initiative at Children’s Mercy Kansas City that seeks to predict clinically important outcomes and pair those predictions to rapid testing of innovative therapies.

What are the highlights that attendees should take away from your presentation?

The T1D Exchange Quality Improvement Collaborative has improved multiple clinical processes, including increasing the rate of screening for depression among adolescents and young adults, increasing the rate of adoption of continuous glucose monitors (which associate with improved glycemic control in many groups), and decreasing the rate of loss to follow-up on the clinic schedule. The diabetes rapid learning lab has used machine learning and deep learning methods to predict a 90-day change in hemoglobin A1c and to predict the 6-month risk for hospitalization for diabetic ketoacidosis, respectively.  These predictions have been deployed routinely in clinic, allowing staff to outreach to families and enroll them in intensified pathways of care that include remote patient monitoring, novel digital health interventions, and care delivery innovations.

How do these findings and/or conclusions potentially impact clinical practice?

We are able to predict near-term clinical outcomes in diabetes with a reasonable positive predictive value and a reasonable sensitivity. There are three important questions that must be answered next: is it possible to improve upon the performance of these predictive models by identifying new sources and types of data to incorporate into the models (e.g., fitness wearable data, new patient-reported outcomes data)? Which interventions seem to move the needle on these novel outcomes? And where multiple options for intervention exist, is it possible to predict which intervention is most likely to be successful for which patient?

What else would you like attendees to know about your presentation?

We are actively seeking innovative interventions to test among patients who are predicted to experience a 90-day rise in hemoglobin A1c (0.3% or greater) or a hospitalization for diabetic ketoacidosis.

Dr. Clements is the CMO for Glooko. He has consulted for Medtronic and Eli Lily, and he receives research support from Abbott Diabetes Care and Dexcom.

Illustration by April Brust

All opinions published on Op-Med are the author’s and do not reflect the official position of Doximity or its editors. Op-Med is a safe space for free expression and diverse perspectives. For more information, or to submit your own opinion, please see our submission guidelines or email

More from Op-Med