Evaluating global and local sequence alignment methods for comparing patient medical records.

Area of Science:

• Bioinformatics
• Health Informatics
• Computational Biology

Background:

• Sequence alignment is crucial for identifying relationships and similarities between data sequences.
• In Electronic Health Records (EHR), sequence alignment aids in identifying patients with similar disease trajectories for improved prognosis, diagnosis, and treatment.
• Evaluating alignment algorithms is key to advancing patient similarity studies.

Purpose of the Study:

• To compare the effectiveness of dynamic time warping (DTW) and Needleman-Wunsch algorithm (NWA), along with their local modifications (DTWL and SWA), for aligning patient medical records.
• To objectively evaluate these sequence alignment algorithms using synthetic EHR data.

Main Methods:

• Tested global sequence alignment methods: dynamic time warping (DTW) and Needleman-Wunsch algorithm (NWA).
• Tested local sequence alignment methods: DTW for Local alignment (DTWL) and Smith-Waterman algorithm (SWA).
• Utilized 4 sets of synthetic patient medical records derived from a large EHR database as gold standard data for evaluation.

Main Results:

• DTW and DTWL generally outperformed NWA and SWA in terms of similarity scores and coverage.
• DTW achieved superior similarity scores in 47/80 global alignments compared to NWA.
• DTWL demonstrated larger coverage and higher similarity scores in 70/80 local alignments compared to SWA.

Conclusions:

• DTW/DTWL algorithms show advantages over NWA/SWA in aligning patient medical records, potentially by better incorporating daily events and identifying more similarities.
• These findings offer insights into the strengths and weaknesses of different sequence alignment methods.
• The results support the development of advanced sequence alignment techniques for patient similarity-based research.