Limits...
Navigating longitudinal clinical notes with an automated method for detecting new information.

Zhang R, Pakhomov S, Lee JT, Melton GB - Stud Health Technol Inform (2013)

Bottom Line: The new information proportion (NIP) in target notes decreased logarithmically with increasing numbers of previous notes to create the language model.Higher NIP scores correlated with notes having more new information often with clinically significant events, and lower NIP scores indicated notes with less new information.Our analysis also revealed "copying and pasting" to be widely used in generating clinical notes by copying information from the most recent historical clinical notes forward.

View Article: PubMed Central - PubMed

Affiliation: Institute for Health Informatics, University of Minnesota, Minneapolis, MN, USA.

ABSTRACT
Automated methods to detect new information in clinical notes may be valuable for navigating and using information in these documents for patient care. Statistical language models were evaluated as a means to quantify new information over longitudinal clinical notes for a given patient. The new information proportion (NIP) in target notes decreased logarithmically with increasing numbers of previous notes to create the language model. For a given patient, the amount of new information had cyclic patterns. Higher NIP scores correlated with notes having more new information often with clinically significant events, and lower NIP scores indicated notes with less new information. Our analysis also revealed "copying and pasting" to be widely used in generating clinical notes by copying information from the most recent historical clinical notes forward. These methods can potentially aid clinicians in finding notes with more clinically relevant new information and in reviewing notes more purposefully which may increase the efficiency of clinicians in delivering patient care.

Show MeSH

Related in: MedlinePlus

(A) statistical language model development; (B) longitudinal data set; (C) score matrix of new information proportion (NIP). Build a language model (A) to calculate the NIP of note k (B) and generate the corresponding cell in the matrix (C).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4495914&req=5

Figure 1: (A) statistical language model development; (B) longitudinal data set; (C) score matrix of new information proportion (NIP). Build a language model (A) to calculate the NIP of note k (B) and generate the corresponding cell in the matrix (C).

Mentions: EHR notes were retrieved from University of Minnesota Medical Center affiliated Fairview Health Services. For this study, we selected patients with multiple co-morbidities, allowing for relatively large numbers of longitudinal records in the outpatient clinic setting. These notes were extracted in text format from the Epicâ„¢ EHR system 1 during a six-year period (06/2005 to 06/2011). To simplify the study, we limited the notes to office visit notes (Fig. 1, see part B). Each note was indexed based on chronological order (e.g., note A1 indicates the 1st note of patient A). Institutional review board approval was obtained and informed consent waived for this minimal risk study.


Navigating longitudinal clinical notes with an automated method for detecting new information.

Zhang R, Pakhomov S, Lee JT, Melton GB - Stud Health Technol Inform (2013)

(A) statistical language model development; (B) longitudinal data set; (C) score matrix of new information proportion (NIP). Build a language model (A) to calculate the NIP of note k (B) and generate the corresponding cell in the matrix (C).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4495914&req=5

Figure 1: (A) statistical language model development; (B) longitudinal data set; (C) score matrix of new information proportion (NIP). Build a language model (A) to calculate the NIP of note k (B) and generate the corresponding cell in the matrix (C).
Mentions: EHR notes were retrieved from University of Minnesota Medical Center affiliated Fairview Health Services. For this study, we selected patients with multiple co-morbidities, allowing for relatively large numbers of longitudinal records in the outpatient clinic setting. These notes were extracted in text format from the Epicâ„¢ EHR system 1 during a six-year period (06/2005 to 06/2011). To simplify the study, we limited the notes to office visit notes (Fig. 1, see part B). Each note was indexed based on chronological order (e.g., note A1 indicates the 1st note of patient A). Institutional review board approval was obtained and informed consent waived for this minimal risk study.

Bottom Line: The new information proportion (NIP) in target notes decreased logarithmically with increasing numbers of previous notes to create the language model.Higher NIP scores correlated with notes having more new information often with clinically significant events, and lower NIP scores indicated notes with less new information.Our analysis also revealed "copying and pasting" to be widely used in generating clinical notes by copying information from the most recent historical clinical notes forward.

View Article: PubMed Central - PubMed

Affiliation: Institute for Health Informatics, University of Minnesota, Minneapolis, MN, USA.

ABSTRACT
Automated methods to detect new information in clinical notes may be valuable for navigating and using information in these documents for patient care. Statistical language models were evaluated as a means to quantify new information over longitudinal clinical notes for a given patient. The new information proportion (NIP) in target notes decreased logarithmically with increasing numbers of previous notes to create the language model. For a given patient, the amount of new information had cyclic patterns. Higher NIP scores correlated with notes having more new information often with clinically significant events, and lower NIP scores indicated notes with less new information. Our analysis also revealed "copying and pasting" to be widely used in generating clinical notes by copying information from the most recent historical clinical notes forward. These methods can potentially aid clinicians in finding notes with more clinically relevant new information and in reviewing notes more purposefully which may increase the efficiency of clinicians in delivering patient care.

Show MeSH
Related in: MedlinePlus