Heather A Eicher-Miller1, Saul Gelfand2, Youngha Hwang3, Edward Delp4, Anindya Bhadra5, Jiaqi Guo6. 1. Department of Nutrition Science, Purdue University, 700 West State Street, West Lafayette, IN, 47907, USA. Electronic address: heicherm@purdue.edu. 2. School of Electrical and Computer Engineering, 465 Northwestern Avenue, Purdue University, West Lafayette, IN, 47907, USA. Electronic address: gelfand@ecn.purdue.edu. 3. School of Electrical and Computer Engineering, 465 Northwestern Avenue, Purdue University, West Lafayette, IN, 47907, USA. Electronic address: hwangyo@purdue.edu. 4. School of Electrical and Computer Engineering, 465 Northwestern Avenue, Purdue University, West Lafayette, IN, 47907, USA. Electronic address: ace@ecn.purdue.edu. 5. Department of Statistics, Purdue University, 250 North University Street, West Lafayette, IN, 47907, USA. Electronic address: bhadra@purdue.edu. 6. School of Electrical and Computer Engineering, 465 Northwestern Avenue, Purdue University, West Lafayette, IN, 47907, USA. Electronic address: guo498@purdue.edu.
Abstract
OBJECTIVE: Few attempts to determine dietary patterns have incorporated concepts of time, specifically time and proportion of energy intake consumed throughout a day. A type of modified dynamic time warping (MDTW) was previously developed using an appropriate distance metric for patterning these aspects to determine temporal dietary patterns (TDP). This study further explores dynamic time warping (DTW) distance metrics including unconstrained DTW (UDTW), constrained DTW (CDTW), and MDTW with modern spectral clustering methods to optimize TDP related to dietary quality. MDTW was expected to create TDP with the strongest relationships to dietary quality and distinct visualization among U.S. adults 20-65y of the National Health and Nutrition Examination Survey 1999-2004. METHODS: Proportional energy intake by time of day metrics were optimized to create TDP from complete day-one 24-h dietary recalls using MDTW, UDTW with only a standard local constraint, and CDTW with standard local and global banding constraints, then clustered using spectral clustering. The association between each TDP distance metric clustering and mean dietary quality, as indicated by the 2005 Healthy Eating Index (HEI-2005), were determined using multiple linear regression controlled for potential confounders. Strength of association for each model was compared using adjusted R-squared. The results were also visualized to make qualitative comparisons. RESULTS: Four clusters representing distinct TDP for each distance metric by spectral clustering were generated among participants. MDTW exhibited TDP clusters with strongest associations to HEI compared with the TDP clusters generated from unconstrained and constrained DTW, and visualization of the TDP clusters from MDTW supported the association. IMPLICATION: MDTW paired with spectral clustering is a useful tool for dimension reduction and uncovering temporal patterns with dietary data.
OBJECTIVE: Few attempts to determine dietary patterns have incorporated concepts of time, specifically time and proportion of energy intake consumed throughout a day. A type of modified dynamic time warping (MDTW) was previously developed using an appropriate distance metric for patterning these aspects to determine temporal dietary patterns (TDP). This study further explores dynamic time warping (DTW) distance metrics including unconstrained DTW (UDTW), constrained DTW (CDTW), and MDTW with modern spectral clustering methods to optimize TDP related to dietary quality. MDTW was expected to create TDP with the strongest relationships to dietary quality and distinct visualization among U.S. adults 20-65y of the National Health and Nutrition Examination Survey 1999-2004. METHODS: Proportional energy intake by time of day metrics were optimized to create TDP from complete day-one 24-h dietary recalls using MDTW, UDTW with only a standard local constraint, and CDTW with standard local and global banding constraints, then clustered using spectral clustering. The association between each TDP distance metric clustering and mean dietary quality, as indicated by the 2005 Healthy Eating Index (HEI-2005), were determined using multiple linear regression controlled for potential confounders. Strength of association for each model was compared using adjusted R-squared. The results were also visualized to make qualitative comparisons. RESULTS: Four clusters representing distinct TDP for each distance metric by spectral clustering were generated among participants. MDTW exhibited TDP clusters with strongest associations to HEI compared with the TDP clusters generated from unconstrained and constrained DTW, and visualization of the TDP clusters from MDTW supported the association. IMPLICATION: MDTW paired with spectral clustering is a useful tool for dimension reduction and uncovering temporal patterns with dietary data.
Authors: Susan M Krebs-Smith; TusaRebecca E Pannucci; Amy F Subar; Sharon I Kirkpatrick; Jennifer L Lerman; Janet A Tooze; Magdalena M Wilson; Jill Reedy Journal: J Acad Nutr Diet Date: 2018-09 Impact factor: 4.910
Authors: Sharon I Kirkpatrick; Jill Reedy; Susan M Krebs-Smith; TusaRebecca E Pannucci; Amy F Subar; Magdalena M Wilson; Jennifer L Lerman; Janet A Tooze Journal: J Acad Nutr Diet Date: 2018-09 Impact factor: 4.910
Authors: Dong D Wang; Cindy W Leung; Yanping Li; Eric L Ding; Stephanie E Chiuve; Frank B Hu; Walter C Willett Journal: JAMA Intern Med Date: 2014-10 Impact factor: 21.873