Trauma Bay Team Lead Gender Bias During Trauma Resuscitation:
An Objective Assessment
We are grateful for support of this project by Janet Cortez, Angel Picos, Becky Utz, and the Center for High Performance Computing at the University of Utah.
This project was supported by a 1U4U award from the U of Utah and by funds provided by Raminder Nirula, MD.
Conclusions
- Men and women ran equally quiet trauma bays.
- Residents and attendings ran equally quiet trauma bays.
- There were not enough Female Attendings to enable testing for differences by gender and rank combined.
- For all leaders, “quiet” requests were effective.
- Only evidence of bias was a marginally significant pattern that Supervising Attendings’ “quiet” request was more effective than others’. Future research is needed to test the robustness of this pattern.
- Audio-recordings of trauma bays are inherently messy. More audio cleaning techniques are needed.
Recommendations
- Technology can efficiently analyze trauma bay noise and communication clarity, providing valuable feedback.
- Quiet requests are effective.
- Highly structured work environments may reduce the likelihood of gender bias.
- Outside of the trauma bay, there is gender bias among team members towards leaders, despite no difference in men’s and women’s leadership quality.2
- Gender bias in the trauma bay would pose a threat to team efficiency and patient outcomes.3
- Whisper, an advanced neural network-based ASR by OpenAI, transcribes audio reliably across languages and challenging conditions, including noise and varied accents.5
- A-Weighting in audio processing adjusts sound measurements to reflect human ear sensitivity6, commonly used to assess perceived loudness and analyze sound levels.7,8
Research Questions
- Does the noise level in the trauma bay differ based on leaders’ gender and/or rank?
- Does a “quiet” request impact communication clarity, and does this differ by leaders’ gender or rank?
- Does a “quiet” request impact noise level, and does this differ by leaders’ gender or rank?
- Human coders identified the leader's gender (man/woman) and rank (resident/attending physician) in videos.
- Audio was transcribed using Whisper 5 to quantify “quiet” utterances. Human coders identified the gender and rank (resident/attending/other team member) of who said “quiet.”
- Two transcripts per video were compared before and after “quiet” utterances using word overlap and embedding similarity to assess communication clarity.
- Sound pressure levels were analyzed with an A-weighted transformation to approximate loudness 6. Average noisiness was measured for the entire resuscitation and also in ten 30-second segments before and after “quiet.”
- Models initially adjusted for patient characteristics (SBP, GCS, Pulse, ISS, RBC, Plasma, and Platelet count), and then were re-estimated including only the significant predictors, as noted in Analysis Strategy.
Significant Controls: trauma bay, patient systolic blood pressure
Number of Videos: 677
Jacqueline M. Chen, PhD1, Becky Neufeld, M.S., Ph.D.1, Mattia M. Grespan, MSc, Ph.D.2, Hailee Davis, B.S.1, Jefferson Dillon, B.S.1, Brian R.W. Baucom, PhD1, Vivek Srikumar, PhD2, & Raminder Nirula, MD3
1Department of Psychology, University of Utah, Salt Lake City, UT; 2 Kahlert School of Computing, University of Utah, Salt Lake City, UT; 3Department of Surgery, University of Utah, Salt Lake City, UT
RQ 1: Noisiness of the trauma bay did not differ by leader gender or rank
RQ 2: “Quiet” request increased communication clarity under all leaders
RQ 3: “Quiet” request decreased noise levels under all leaders
Significant Controls: Plasma vol (ML) given in first 4 hrs
Number of Videos: 63
Significant Controls: trauma bay
Number of Videos: 147
818 total videos
- 155 videos where “quiet” was said as a command
Subset to Videos with Complete Data for Patient Controls
RQ 1: N = 196
RQ 2: N = 54
RQ 3: N = 57
Model With All Control Variables
RQ 1: Higher systolic blood pressure - quieter rooms; Bay 2 louder than Bay 1
RQ 2: Higher plasma - louder rooms
RQ 3: Bay 2 louder than Bay 1
Final Model Interpretation
All plotted effects above are controlling for the significant control variables
Model 1: OLS Regression; Models 2 and 3: Multilevel Modeling
Increase Subset to Videos with Complete Data for Significant Controls
RQ 1: N = 677
RQ 2: N = 63
RQ 3: N = 147
- We analyzed 818 videos recorded in the trauma bay for indications of negative team functioning including noise level, communication clarity, and instances of “quiet” requests (in 155 videos).
University of Utah Trauma Service Training Practice 9
Table 1
Trauma Leader Characteristics
University of Utah Trauma Bay
Scan the QR code for our references
- Human coders identified the leader's gender (man/woman) and rank (resident/attending physician) in videos.
- Audio was transcribed using Whisper 5 to quantify “quiet” utterances. Human coders identified the gender and rank (resident/attending/other team member) of who said “quiet.”
- Two transcripts per video were compared before and after “quiet” utterances using word overlap and embedding similarity to assess communication clarity.
- Sound pressure levels were analyzed with an A-weighted transformation to approximate loudness 6. Average noisiness was measured for the entire resuscitation and also in ten 30-second segments before and after “quiet.”
- Models initially adjusted for Trauma Bay (1, 2) and patient characteristics (SBP, GCS, Pulse, ISS, RBC, Plasma Vol, Platelet Vol), and then were re-estimated including only the significant controls, as noted in Analysis Strategy.