Considering more than 70% of healthcare information is stored in unstructured clinical notes, healthcare providers increasingly demand effective text-search systems for clinical care, QI reporting and research projects. However, clinical notes are known for containing spelling variations, typos, local-practice-generated acronyms, synonyms, and informal words. We developed a novel search system (QREK) that can guide users to expand their input query by suggesting spelling variations, acronyms and other semantically relevant words, all identified through an artificial intelligence (AI)-driven unsupervised shallow learning algorithm, designed by our our Natural Language Processing team.
How does searching in QREK differ from Epic’s search functionality?
1. Go beyond searching a single patient: Epic’s search functionality allows a search to be conducted on one patient at a time. By contrast, QREK searches across all patients within Epic, making it versatile for finding cohorts.
2. Broader suggestions: Epic’s search functionality currently includes expanding a user’s input keyword(s) using known ontologies. However, it doesn’t match notes with misspellings, typos, and locally practiced shorthands. For example, a search for “tonsillectomy” will not find a common misspelling “tonsilectomy” (one “L” only instead of double). By contrast, QREK does consider this and in so doing, broadens user queries to maximize the chance of finding all relevant notes.
Total matches: QREK provides the user with an approximate total number of notes, visits/encounters, and patients that match the search criteria.
Suggestions: QREK provides the user with a list of suggested synonyms, spelling variations, abbreviations, semantically relevant words, and words otherwise relevant to the query keyword.
Auto-complete: QREK will auto complete common words found in clinical notes as the user types out a search term.
Advanced search: Users can create very advanced and nested queries by applying multiple AND, OR, or NOT conditions.
Contextual Estimates: QREK provides the user with an approximation of how many matched results might be incidental and/or unwanted matches such as negative mentions, or patient’s family history. Users will be able to see samples by clicking on each category.
Filters: Users can filter results by visit date and note type.
Patient lookup: Users can lookup a patient using first name, last name, patient id, date of birth, or a combination of those.
Export the results: Users can export up to 4,000 records with a selection of fields for further analysis.
Filters: Filters are available only for note types and visit date. It is not possible to filter by author of a note, ICD codes, etc.
Updates: The data for QREK updates every other month. This translates to searches being conducted on slightly old data.
Formatting: Formatting of data (e.g. fonts, tables) are removed.
Data source: QREK incorporates most but not “all” note types within Epic.
Approximate numbers: Some results such as number of encounters, patient, and contextual estimates are just approximated numbers, and not exact.
Export: Only up to 4,000 records can be exported. Users are encouraged to narrow their search result using filters and advanced searches prior to downloading results. Please note that QREK's results are extracted from clinical notes, not code-based data, allowing keyword matches. This feature will generate more relevant results, but results can still be wrong or inaccurate due to keyword matching challenges such as ambiguity, matching negative mentions or template words, etc. Go back to the basics of data science, “all models can be wrong, but some are useful”.
Example use caseStomachache is a symptom often reported in clinical notes but with many different variations. Clinical notes at NCH show “stomachache” as: stomachache, stomach ache, stomachaches, bellyache, belly ache, bellyaches, tummy ache, stomach pains, and other variations. QREK provides users with a list of suggested spelling and documentation variations, along with possible relevant keywords of interest such as “vomiting” and “nausea” to enable users to expand their query and potentially find more relevant clinical notes and encounters and patients associated with them.
DeepSuggest can guide users to expand their input query by suggesting spelling variations, acronyms and other semantically relevant words, all identified through an artificial intelligence (AI)-driven unsupervised shallow learning algorithm, designed by our our Natural Language Processing team.
Watch the video
How to get access?
Non-NCH entities interested in adopting and deploying the platform over their own data repositories can contact Tyler Lieser at Tyler.Lieser@nationwidechildrens.org for more information and contact details.
This research is funded by a Patient-Centered Outcomes Research Institute (PCORI) Award (ME-2017C1-6413). If you are interested in QREK, for access and collaboration please contact us at R&DIntake@nationwidechildrens.org