Abstract
ObjectiveTo describe the effect of symptom negation in emergencydepartment (ED) chief complaint data received by the New York City(NYC) Department of Health and Mental Hygiene (DOHMH), and todevise a solution to avoid syndrome and symptom misclassificationfor commonly used negations using SAS Perl Regular Expression(PRX) functions.IntroductionIn July 2016, 77% of ED data was transmitted daily via HealthLevel 7 (HL7) messages, compared to only 27% in July 2015(Figure). During this same period, chief complaint (CC) word counthas increased from an average of 3.8 words to 6.0 words, with atwenty-fold increase in the appearance of the word “denies” in thechief complaint (Figure). While HL7 messages provide robust chiefcomplaint data, this may also introduce errors that could lead tosymptom and syndrome misclassification.MethodsUsing SAS 9.4 and Tableau 9.3, we examined data submissionsfrom 14 EDs responsible for 97% of the occurrences of the word‘denies’ in chief complaints in July 2016.To account for variation in chief complaint format among hospitals,we developed three PRX patterns to identify entire phrases in thechief complaint data field that began with conjugations of the word“deny” followed by various combinations of words, punctuation,spaces, and/or characters.Pattern 1: ''/DEN(Y|I(ES|ED|NG))(\s|\w|(\/)|(\+)|,|(\\)){1,}((\.)|(\|)|($)|(;)|(\))|(-))/’Pattern 2: ''/DEN(Y|I(ES|ED|NG))(\s|\w|(\/)|(\+)|(\\)){1,}((\.)|(\|)|($)|(;)|(\))|(-)|(,))/'';Pattern 3: ''/DENIES:( |\w|\.|,){1,}/'');We separated the ‘denies’ statement from the chief complaint andidentified commonly negated symptoms. We then defined symptomsusing keyword searches of the chief complaint and the ‘denies’statement. We compared symptom classification with and withoutthe consideration of symptom negation.ResultsOf the 14 EDs analyzed, we applied pattern 1 to 8 of the ED’s,pattern 2 to 5 EDs, and patterns 2 and 3 to 1 ED. Approximately98% of denies statements were extracted from chief complaints. Only2% of symptom negation was not captured due to uncommon chiefcomplaint format whose symptom negation didn’t meet one of thepreviously described PRX patterns.The most common words associated with a “denies” statementwere: pain, chest, fever, loc, shortness, breath, vomiting, nausea,travel, headache, recent, trauma, history, abdominal, injury, diarrhea,SOB (shortness of breath), V (vomit), Head, N (nausea), PMH (pastmedical history), suicidal, dizziness, homicidal and D (diarrhea) (seeTable).By not taking negation into consideration in symptom definitions,between 3.5% and 16.5% of symptom visits were misclassified.Symptom misclassification varied greatly by hospital, ranging from0% to 55%.ConclusionsAs hospitals in NYC implement HL7 messaging, symptomnegation is becoming increasingly common in chief complaint data.Current symptom definitions are based on keyword searches that donot take into account symptom negations. This leads to symptommisclassification, and could potentially cause false signals or inflatesyndrome baselines, causing true signals to go undetected. SAS PRXfunctions can be used to flexibly identify symptom negation patternsand exclude them from syndrome definitions. Future studies willquantify the effect symptom negation has had on signal frequency inNYC, and examine symptoms associated with other forms of negationsuch as “Personal Medical History”, “No” and “Negative.”Most Common Symptoms Denied in Emergency Department Chief Complaints