On call workflow, attendings (sometimes) giving feedback hours later
dashboard components
Windows server running XAMPP, using Apache, MySQL, PHP, Python, and ASP
Components:
Discrepancy Logger
Report Comparator
Report Difference Chart
discrepancy logger
Javascript plugin using PACS API (iSite 3.6, Philips)
button in PACS interface loads case metadata (time, date, modality, subspecialty, and accession) and allows comments, choosing RadPeer classification.
emails trainee directly and logs info in MySQL DB
report comparator
CRON job scripted in Perl
queries mirrored RIS for prelim reports, saves in MySQL DB
separate queries for finalized reports, matches and saves both versions in DB
Allow logging in to web service to view diffs
Authors' DB contains 1 million reports
report difference chart
Python script calculates Levenshtein distance between prelim and final reports
Normalizes distance for longer reports by calculating percentage
Allows visualization of data in various formats, e.g.
Trainees can then view the average # of changes between trainees globally and for each attending-trainee pair
Trainees ranked in chart so each trainee can track their performance relative to their peers
Color coded to highlight outliers
figures
Average Lev dist % for trainee-attndg pair on top, avg % for particular study type on bottom
Trainees ordered by performance across top of chart, better performance to the right
Individual reports for each attending/trainee pair
Advantages
Streamlines feedback workflow, esp when not in close proximity
Encourages feedback by making process less onerous
Allows trainees to gauge relative performance
can drill down to specific poorly performing modalities
Possible use as objective measure of trainee competency
Showed decrease in Levenshtein percentage in cohort of residents over time
disadvantages
Many sources of differences between reports
Dictation software mistakes
Attending wording preferences
Mishearing attending during readout
Disagreement in interpretation
New clinical information following preliminary report
Makes assumption that more changed words equates with poorer report quality
cannot distinguish the clinical significance of a change, and a one word change may have high significance (e.g. inserting "no", or changing L3 --> L4)