TY - GEN
T1 - Temporal sequence alignment in electronic health records for computable patient representation
AU - Huang, Ming
AU - Zolnoori, Maryam
AU - Shah, Nilay D.
AU - Yao, Lixia
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2019/1/21
Y1 - 2019/1/21
N2 - Constructing patient representation from EHRs has become an emerging hot research topic, as it is widely used for predicting disease prognosis, medication outcomes and mortality, and identifying patients who are similar to a target patient. Sequence alignment methods are able to preserve the temporal sequence information in patient medical records when constructing computable patient representation and thus are worth comprehensive and objective evaluation. In this work, we synthesized patient medical records using a set of synthesis operations on top of real patient medical records from a large real-world EHR database. Then we tested two cutting-edge sequence alignment methods, namely dynamic time warping (DTW) and Needleman-Wunsch algorithm (NWA) for the purpose of patient medical records alignment, in order to understand their strengths and limitations. Our results show that both DTW and NWA outperform the reference alignment. DTW seems to align better than NWA by inserting new daily events and identifying more similarities between patient medical records. By incorporating medical knowledge, we can improve the temporal sequence alignment by these algorithms even better and create more accurate patient representation for predictive models and patient similarity calculation.
AB - Constructing patient representation from EHRs has become an emerging hot research topic, as it is widely used for predicting disease prognosis, medication outcomes and mortality, and identifying patients who are similar to a target patient. Sequence alignment methods are able to preserve the temporal sequence information in patient medical records when constructing computable patient representation and thus are worth comprehensive and objective evaluation. In this work, we synthesized patient medical records using a set of synthesis operations on top of real patient medical records from a large real-world EHR database. Then we tested two cutting-edge sequence alignment methods, namely dynamic time warping (DTW) and Needleman-Wunsch algorithm (NWA) for the purpose of patient medical records alignment, in order to understand their strengths and limitations. Our results show that both DTW and NWA outperform the reference alignment. DTW seems to align better than NWA by inserting new daily events and identifying more similarities between patient medical records. By incorporating medical knowledge, we can improve the temporal sequence alignment by these algorithms even better and create more accurate patient representation for predictive models and patient similarity calculation.
KW - dynamic time warping
KW - electronic health record
KW - needleman-Wunsch algorithm
KW - patient representation
KW - patient similarity
KW - sequence alignment method
KW - temporal sequence
UR - http://www.scopus.com/inward/record.url?scp=85062532392&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85062532392&partnerID=8YFLogxK
U2 - 10.1109/BIBM.2018.8621428
DO - 10.1109/BIBM.2018.8621428
M3 - Conference contribution
AN - SCOPUS:85062532392
T3 - Proceedings - 2018 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2018
SP - 1054
EP - 1061
BT - Proceedings - 2018 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2018
A2 - Schmidt, Harald
A2 - Griol, David
A2 - Wang, Haiying
A2 - Baumbach, Jan
A2 - Zheng, Huiru
A2 - Callejas, Zoraida
A2 - Hu, Xiaohua
A2 - Dickerson, Julie
A2 - Zhang, Le
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2018 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2018
Y2 - 3 December 2018 through 6 December 2018
ER -