Extended Work Periods
Among Anesthesiologists and
Postoperative Patient Outcomes
Anjali A. Dixit, MD, MPH
Instructor, Division of Pediatric Anesthesiology
Department of Anesthesiology, Perioperative and Pain Medicine
Stanford University School of Medicine
Agenda
Conclusions
MPOG Study
Background
Disclosures
• Funding from National Institute on Drug Abuse – unrelated to this work
Background: Motivation
Background: Research Question
• Are patient outcomes different when attending anesthesiologists have been working
for an extended time vs not?
• This work built on another MPOG project:
Agenda
Conclusions
MPOG Study
Background
Work
Shift
Length
Patient
Outcomes
Analysis
Methods: Exposure
Work
Shift
Length
Patient
Outcomes
Analysis
Figure created with assistance from OpenAI’s Codex.
Methods: Exposure
Work
Shift
Length
Patient
Outcomes
Analysis
Figure created with assistance from OpenAI’s Codex.
Methods: Exposure
Work
Shift
Length
Patient
Outcomes
Analysis
Methods: Exposure
Figure created with assistance from OpenAI’s Codex.
Methods: Exposure
Work
Shift
Length
Patient
Outcomes
Analysis
Methods: Exposure
Figure created with assistance from OpenAI’s Codex.
Methods: Exposure
Work
Shift
Length
Patient
Outcomes
Analysis
Methods: Outcomes
• Outcome: a composite of in-hospital mortality and other plausible measures of
morbidity (MI, stroke, other thromboembolic event, major respiratory complication, AKI,
anesthesia-related complications, mesenteric ischemia)
Work
Shift
Length
Patient
Outcomes
Analysis
Methods: Analysis
• Fake and Impossible RCT:
Figure created with assistance from OpenAI’s Codex.
Work
Shift
Length
Patient
Outcomes
Analysis
Methods: Fixed Effects
• In our linear probability models, we included fixed effects by: Anesthesiologist; Time-
of-day (by hour); Surgery type
• Also adjusted for patient demographics and comorbidities, ASA status, emergency
status
Work
Shift
Length
Patient
Outcomes
Analysis
Figure created with assistance from OpenAI’s Codex.
Results
• 1,648,720 surgical procedures
• Involving 1,711 unique anesthesiologists across 36 institutions
• 11,556 procedures (0.7%) involved primary anesthesiologists who had worked ≥16
hours at the start of case
• Unadjusted incidence of the composite outcome was:
• Extended work period: 12.3% (95% CI, 11.7%-12.9%)
• Regular work period: 12.0% (95% CI, 12.0%-12.1%)
• Risk difference, 0.3%; 95% CI, −0.3% to 0.9%; P = .38
Results
Results
Results
Results
Results
Agenda
Conclusions
MPOG Study
Background
Conclusions
• These findings leave an important question open
• Possible association between anesthesiologist work shifts and mortality
• Even a 0.3% increase in mortality attributable to physician fatigue is meaningful
• MPOG data can be leveraged to understand this association better as it provides
granular data on our workforce and our patient care patterns
ConclusionsConclusions
Conclusions
Questions, Comments, Future Collaborations
• anjalid@stanford.edu
Thank You!
• Ben Deverett and Eric Sun
• The MPOG community, especially our study collaborators: Robert B. Schonberger, MD, MHCDS (Yale
School of Medicine), Bhiken I. Naik, MBBCh, MSCR (University of Virginia), J David Clark, MD, PhD
(Stanford University School of Medicine), and Robert E. Freundlich, MD, MS (Vanderbilt University
Medical Center).
• The whole MPOG team at University of Michigan, especially Jay Jeong who fielded many of my
questions about the data
Extra Slides
Fixed v Random Effects
• Fixed effects provide a within-anesthesiologist comparison (within the same
anesthesiologist, are outcomes different when they’re working for a long time vs not)
• Accounts for stable anesthesiologist characteristics (like how they practice anesthesiology)
• Random effects combines a within- and between-anesthesiologist estimate that is not
what we want here
• Estimates variation across anesthesiologists so is useful for prediction models
• Assumes clinician-specific intercept is unrelated to other predictors including the exposure
(not necessarily true because some clinicians may work a lot more extended shifts than
others and this may also be related to patient type)
