Presentation Title
Identifying Connectional Silence in Palliative Care Consultations
Abstract
Context
Improvement of communication among clinicians, seriously ill patients, and family members is a current national priority. To better understand, publicize, and incentivize high-quality serious illness communication, conversational features in the natural clinical setting must first be made systematically measureable. Automating or semi-automating the measurement of conversation features using machine learning (ML) alongside human coding (HC) for qualitative measurements is a promising tandem method that shows increased productivity when compared to HC alone. This tandem ML-HC method was tested for reliability, efficiency, and sensitivity in identifying Connectional Silence as an important feature of the serious illness conversation.
Connectional Silence
In the context of suffering, current literature identifies the existence of pauses that occur between patients and clinicians which can represent pivotal moments of shared understanding and presence. As they are moments which can be recognized and set apart as involving a quantifiable moment of human connection, we refer to these moments as Connectional Silences. Connectional Silences were classified in the HC codebook as Invitational (signaling a moment of trust-building and assured listening between patient/family and clinician), Emotional (involving a moment of respectful silence following an emotional/gravitational moment or expression), and Compassionate (signaling verbal confirmation of a moment of difficulty, emotion, and/or gravity).
Methods
This was a cross-sectional analysis of 354 audio-recorded inpatient palliative care consultations from the Palliative Care Communication Research Initiative multisite cohort study. A ML screening algorithm was used to predict pauses within the dataset of 2 or more seconds, and 1000 of these were randomly selected and produced with surrounding context. HC’s qualitatively searched these clips for contextually important silences and developed a codebook of meaningful silence types. HC’s also evaluated 100 minutes of non-clipped audio data to both validate the ML algorithm and compare efficiency to tandem ML-HC.
Results
Connectional Silences were found to be rare (5.5%) among all two-second or longer pauses in palliative care conversations. Tandem ML-HC demonstrated strong reliability (kappa 0.62; 95% confidence interval: 0.47– 0.76). HC alone required 61% more time than the Tandem ML-HC method. No Connectional Silences were missed by the ML screening algorithm.Thus, the ML-HC method was found to be reliable, sensitive, and efficient for identifying Connectional Silence in serious illness conversations.
Primary Faculty Mentor Name
Robert Gramling
Faculty/Staff Collaborators
Viktoria Manukyan (Graduate Student Mentor)
Status
Undergraduate
Student College
College of Arts and Sciences
Program/Major
Biochemistry
Second Program/Major
French
Primary Research Category
Health Sciences
Identifying Connectional Silence in Palliative Care Consultations
Context
Improvement of communication among clinicians, seriously ill patients, and family members is a current national priority. To better understand, publicize, and incentivize high-quality serious illness communication, conversational features in the natural clinical setting must first be made systematically measureable. Automating or semi-automating the measurement of conversation features using machine learning (ML) alongside human coding (HC) for qualitative measurements is a promising tandem method that shows increased productivity when compared to HC alone. This tandem ML-HC method was tested for reliability, efficiency, and sensitivity in identifying Connectional Silence as an important feature of the serious illness conversation.
Connectional Silence
In the context of suffering, current literature identifies the existence of pauses that occur between patients and clinicians which can represent pivotal moments of shared understanding and presence. As they are moments which can be recognized and set apart as involving a quantifiable moment of human connection, we refer to these moments as Connectional Silences. Connectional Silences were classified in the HC codebook as Invitational (signaling a moment of trust-building and assured listening between patient/family and clinician), Emotional (involving a moment of respectful silence following an emotional/gravitational moment or expression), and Compassionate (signaling verbal confirmation of a moment of difficulty, emotion, and/or gravity).
Methods
This was a cross-sectional analysis of 354 audio-recorded inpatient palliative care consultations from the Palliative Care Communication Research Initiative multisite cohort study. A ML screening algorithm was used to predict pauses within the dataset of 2 or more seconds, and 1000 of these were randomly selected and produced with surrounding context. HC’s qualitatively searched these clips for contextually important silences and developed a codebook of meaningful silence types. HC’s also evaluated 100 minutes of non-clipped audio data to both validate the ML algorithm and compare efficiency to tandem ML-HC.
Results
Connectional Silences were found to be rare (5.5%) among all two-second or longer pauses in palliative care conversations. Tandem ML-HC demonstrated strong reliability (kappa 0.62; 95% confidence interval: 0.47– 0.76). HC alone required 61% more time than the Tandem ML-HC method. No Connectional Silences were missed by the ML screening algorithm.Thus, the ML-HC method was found to be reliable, sensitive, and efficient for identifying Connectional Silence in serious illness conversations.