This paper is only available as a PDF. To read, Please Download here.
Outcomes
1. Contrast examples of symptom expressions and examine the implications of not understanding certain expressions
2. Describe the importance of capturing patient-centered language in developing algorithms to detect symptoms
Importance
Among seriously ill patients, symptom burden is often unrecognized. Computational methods can be used for attending to this burden; these tools can assist with symptom identification for patient monitoring and quality improvement. A natural language processing keyword library can be used as a rudimentary tool to detect “symptom talk” and assist the development of more advanced algorithms.
Objective(s)
To build a PRO-CTCAE symptom keyword library inclusive of the various verbal expressions by which patients and their physicians describe symptoms.
Method(s)
A keyword library was drafted using transcription data from the Communication in Oncologist-Patient Encounters (COPE) trial, which includes audio-recorded outpatient oncology encounters between patient-oncologist dyads. In 93 conversations, three human annotators determined whether symptoms were discussed in each speaker turn, on a scale of 0 (not relevant) to 3 (relevant). For a subset of 48 conversations, words in all turns coded as 3 were extracted and relevant terms added to the developing library. This library was supplemented with language from a prior study extracting symptoms from clinical notes and with a priori additions. The library was tested on 45 additional conversations.
Results
The current library comprises 738 terms, including 137 symptom-related medications and 46 colloquialisms. In the test sample of 45 conversations coded as 3, keywords captured PRO-CTCAE symptom-containing turns with a sensitivity of 78.2%, a specificity of 95.3%, and an accuracy of 83.8%. Preliminary qualitative observations suggest that clinicians more often refer to symptoms via medications (e.g., “Pulmicort”), whereas patients used colloquial verbiage to describe symptoms (e.g., “huffing and puffing”).
Conclusion(s)
This rudimentary keyword library manages to capture most “symptom talk.”
Impact
Patients may verbalize symptoms using different language than clinicians, which clinicians should take care to recognize. In algorithms to recognize symptom talk, supplementing standardized clinical terms with patient-centered terms may better capture symptom burden in cancer and work toward the mitigation of symptom-related suffering in this seriously ill population.
Article info
Identification
Copyright
© 2022 Published by Elsevier Inc.
