Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
Text mining health records (english)
Transcript of Text mining health records (english)
Manual rewievs of health records are resource intensive Improvement of patient safety Challenges! A frequently used validated method for structured reviews of health records
Two reviewers and a supervisor read e.g. 20 records from a hospital each month
It includes a list of 56 triggers (markers) for possible adverse events/patient harms
Possible harms are registered in parallel
It typically takes 40-50 minuts per patient record
The quality of the method is discussed, e.g. because the results can be inconsistent Manual method: Global Trigger Tool Algoritmerne er (især) gode til at frasortere journaler uden problemer…! Patients who have fallen during a hospital stay, whether the incident caused injuries or not Results :: Example :: Falls A need for refining the definitions of triggers and adverse events?
Other ways to measure harms? What is »the truth«…? A central problem Collaboration There is a large potential for using text analytics in health care
The work is demanding, particularly when working with a small language as Danish
We are trying to collaborate nationally Thank you for your attention! Purpose Methods Results Develop-ment The art of getting the right data in the hands of the right people with the right tools at the right time… And project management… The language in health records True sentences versus SMS-style, e.g.
»Patienten har selv seponeret sit kateter«
»Pt selv sep kat«
Spelling errors, e.g.
Tryksår = Trygsår, tryk sår, tyksår, decibitus, dukibitus… etc.
Abbreviations (unauthorised; slang), e.g.
aff, at, blp, cic, epi, epid, fl, forb, ga, gens, gop, hyg, inf, ka, kath, knag, kvalmest, kvo, mob, msu, omk, ooa, opv, seq, sh, sik, stg, suff, uls, vt
Complete nonsense, e.g.
»Bor på landbrug med dyr hånd« (speech recognition) The contents of health records Complicated and variant structures
Different kinds of notes
Things that are changing over time
Quantity & quality of texts
Considerable differences in how much and how well people write in health records
Redundant information (repeats)
Implicit information (what is not written) Departments and hospitals are different Important to realise…!
Different patient groups
Different ways of recording informations
Different languages (»dialects« and slang) Module-based algorithms are advantagous Consultant, MD, DMSci, assistant professor
Center for Quality ∙ Region of Southern Denmark ∙ Middelfart
Institute of Regional Health Resarch ∙ University of Southern Denmark The technique has large potentials
There are a number of challenges
National cooperation is desirable Conclusions Practically... Manuel
læsning Computer Which triggers and events? Questions…? Sharing information...! An integrated IT-solution for practical use SAS Enterprise Text Miner
SAS Content Categorization Software How do we do it...? A prototype is about to be tested We are testing the algorithms with several series of new (unknown) health records