Understanding and Predicting the Effect of Environmental Factors on People with Type 2 Diabetes

Kailas Vodrahalli, Gregory D Lyng, Brian L Hill, Kimmo Karkkainen, Jeffrey Hertzberg, James Zou, Eran Halperin
Proceedings of the Conference on Health, Inference, and Learning, PMLR 209:545-555, 2023.

Abstract

Type 2 diabetes mellitus (T2D) affects over 530 million people globally and is often difficult to manage leading to serious health complications. Continuous glucose monitoring (CGM) can help people with T2D to monitor and manage the disease. CGM devices sample an individual’s glucose level at frequent intervals enabling sophisticated characterization of an individual’s health. In this work, we leverage a large dataset of CGM data (5,447 individuals and 940,663 days of data) paired with health records and activity data to investigate how glucose levels in people with T2D are affected by external factors like weather conditions, extreme weather events, and temporal events including local holidays. We find temperature (p=$2.37\times10^{-8}$, n=3561), holidays (p=$2.23\times10^{-46}$, n=4079), and weekends (p=$7.64\times10^{-124}$, n=5429) each have a significant effect on standard glycemic metrics at a population level. Moreover, we show that we can predict whether an individual will be significantly affected by a (potentially unobserved) external event using only demographic information and a few days of CGM and activity data. Using random forest classifiers, we can predict whether an individual will be more negatively affected than a typical individual with T2D by a given external factor with respect to a given glycemic metric. We find performance (measured as ROC-AUC) is consistently above chance (across classifiers, median ROC-AUC=0.63). Performance is highest for classifiers predicting the effect of time-in-range (median ROC-AUC=0.70). These are important findings because they may enable better patient care management with day-to-day risk assessments based on external factors as well as improve algorithm development by reducing train- and test-time bias due to external factors.

Cite this Paper

BibTeX
@InProceedings{pmlr-v209-vodrahalli23a, title = {Understanding and Predicting the Effect of Environmental Factors on People with Type 2 Diabetes}, author = {Vodrahalli, Kailas and Lyng, Gregory D and Hill, Brian L and Karkkainen, Kimmo and Hertzberg, Jeffrey and Zou, James and Halperin, Eran}, booktitle = {Proceedings of the Conference on Health, Inference, and Learning}, pages = {545--555}, year = {2023}, editor = {Mortazavi, Bobak J. and Sarker, Tasmie and Beam, Andrew and Ho, Joyce C.}, volume = {209}, series = {Proceedings of Machine Learning Research}, month = {22 Jun--24 Jun}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v209/vodrahalli23a/vodrahalli23a.pdf}, url = {https://proceedings.mlr.press/v209/vodrahalli23a.html}, abstract = {Type 2 diabetes mellitus (T2D) affects over 530 million people globally and is often difficult to manage leading to serious health complications. Continuous glucose monitoring (CGM) can help people with T2D to monitor and manage the disease. CGM devices sample an individual’s glucose level at frequent intervals enabling sophisticated characterization of an individual’s health. In this work, we leverage a large dataset of CGM data (5,447 individuals and 940,663 days of data) paired with health records and activity data to investigate how glucose levels in people with T2D are affected by external factors like weather conditions, extreme weather events, and temporal events including local holidays. We find temperature (p=$2.37\times10^{-8}$, n=3561), holidays (p=$2.23\times10^{-46}$, n=4079), and weekends (p=$7.64\times10^{-124}$, n=5429) each have a significant effect on standard glycemic metrics at a population level. Moreover, we show that we can predict whether an individual will be significantly affected by a (potentially unobserved) external event using only demographic information and a few days of CGM and activity data. Using random forest classifiers, we can predict whether an individual will be more negatively affected than a typical individual with T2D by a given external factor with respect to a given glycemic metric. We find performance (measured as ROC-AUC) is consistently above chance (across classifiers, median ROC-AUC=0.63). Performance is highest for classifiers predicting the effect of time-in-range (median ROC-AUC=0.70). These are important findings because they may enable better patient care management with day-to-day risk assessments based on external factors as well as improve algorithm development by reducing train- and test-time bias due to external factors.} }
Endnote
%0 Conference Paper %T Understanding and Predicting the Effect of Environmental Factors on People with Type 2 Diabetes %A Kailas Vodrahalli %A Gregory D Lyng %A Brian L Hill %A Kimmo Karkkainen %A Jeffrey Hertzberg %A James Zou %A Eran Halperin %B Proceedings of the Conference on Health, Inference, and Learning %C Proceedings of Machine Learning Research %D 2023 %E Bobak J. Mortazavi %E Tasmie Sarker %E Andrew Beam %E Joyce C. Ho %F pmlr-v209-vodrahalli23a %I PMLR %P 545--555 %U https://proceedings.mlr.press/v209/vodrahalli23a.html %V 209 %X Type 2 diabetes mellitus (T2D) affects over 530 million people globally and is often difficult to manage leading to serious health complications. Continuous glucose monitoring (CGM) can help people with T2D to monitor and manage the disease. CGM devices sample an individual’s glucose level at frequent intervals enabling sophisticated characterization of an individual’s health. In this work, we leverage a large dataset of CGM data (5,447 individuals and 940,663 days of data) paired with health records and activity data to investigate how glucose levels in people with T2D are affected by external factors like weather conditions, extreme weather events, and temporal events including local holidays. We find temperature (p=$2.37\times10^{-8}$, n=3561), holidays (p=$2.23\times10^{-46}$, n=4079), and weekends (p=$7.64\times10^{-124}$, n=5429) each have a significant effect on standard glycemic metrics at a population level. Moreover, we show that we can predict whether an individual will be significantly affected by a (potentially unobserved) external event using only demographic information and a few days of CGM and activity data. Using random forest classifiers, we can predict whether an individual will be more negatively affected than a typical individual with T2D by a given external factor with respect to a given glycemic metric. We find performance (measured as ROC-AUC) is consistently above chance (across classifiers, median ROC-AUC=0.63). Performance is highest for classifiers predicting the effect of time-in-range (median ROC-AUC=0.70). These are important findings because they may enable better patient care management with day-to-day risk assessments based on external factors as well as improve algorithm development by reducing train- and test-time bias due to external factors.
APA
Vodrahalli, K., Lyng, G.D., Hill, B.L., Karkkainen, K., Hertzberg, J., Zou, J. & Halperin, E.. (2023). Understanding and Predicting the Effect of Environmental Factors on People with Type 2 Diabetes. Proceedings of the Conference on Health, Inference, and Learning, in Proceedings of Machine Learning Research 209:545-555 Available from https://proceedings.mlr.press/v209/vodrahalli23a.html.

Related Material