This cross-sectional study uses data from the Multicenter Perioperative Outcomes Group to examine the association between an attending surgeon performing operations overnight and outcomes for operations performed by the same surgeon the next day.
Key Points
Question
Is operating overnight associated with worse outcomes for surgical procedures performed by the attending surgeon the subsequent day?
Findings
In this cross-sectional study of 498 234 daytime operations performed by 1131 surgeons at 20 US institutions, there was no significant association between operating the previous night and the incidence of in-hospital death or major complications (sepsis, pneumonia, myocardial infarction, thromboembolic event, or stroke) for daytime operations performed the subsequent day. After adjusting for confounders, the incidence of death or major complications was 5.89% among daytime operations when the attending surgeon operated the night before compared with 5.87% among daytime operations when the attending surgeon did not.
Meaning
These findings suggest that operating overnight does not appear to be associated with worse outcomes when the attending surgeon continues to operate the next day.
Abstract
Importance
The association between physician fatigue and patient outcomes is important to understand but has been difficult to examine given methodological and data limitations. Surgeons frequently perform urgent procedures overnight and perform additional procedures the following day, which could adversely affect outcomes for those daytime operations.
Objective
To examine the association between an attending surgeon operating overnight and outcomes for operations performed by that surgeon the next day.
Design, Setting, and Participants
In this cross-sectional study, a retrospective analysis of a large multicenter registry of surgical procedures was done using a within-surgeon analysis to address confounding, with data from 20 high-volume US institutions. This study included 498 234 patients who underwent a surgical procedure during the day (between 7 am and 5 pm) between January 1, 2010, and August 30, 2020.
Exposures
Whether the attending surgeon for the current day’s procedures operated between 11 pm and 7 am the previous night. Two exposure measures were examined: whether the surgeon operated at all the previous night and the number of hours spent operating the previous night (including having performed no work at all).
Main Outcomes and Measures
The primary composite outcome was in-hospital death or major complication (sepsis, pneumonia, myocardial infarction, thromboembolic event, or stroke). Secondary outcomes included operation length and individual outcomes of death, major complications, and minor complications (surgical site infection or urinary tract infection).
Results
Among 498 234 daytime operations performed by 1131 surgeons, 13 098 (2.6%) involved an attending surgeon who operated the night before. The mean (SD) age of the patients who underwent an operation was 55.3 (16.4) years, and 264 740 (53.1%) were female. After adjusting for operation type, surgeon fixed effects, and observable patient characteristics (ie, age and comorbidities), the adjusted incidence of in-hospital death or major complications was 5.89% (95% CI, 5.41%-6.36%) among daytime operations when the attending surgeon operated the night before compared with 5.87% (95% CI, 5.85%-5.89%) among daytime operations when the same surgeon did not (absolute adjusted difference, 0.02%; 95% CI, −0.47% to 0.51%; P = .93). No significant associations were found between overnight work and secondary outcomes except for operation length. Operating the previous night was associated with a statistically significant decrease in length of daytime operations (adjusted length, 112.7 vs 117.4 minutes; adjusted difference, −4.7 minutes; 95% CI, −8.7 to −0.8, P = .02), although this difference is unlikely to be meaningful.
Conclusions and Relevance
The findings of this cross-sectional study suggest that operating overnight was not associated with worse outcomes for operations performed by surgeons the subsequent day. These results provide reassurance concerning the practice of having attending surgeons take overnight call and still perform operations the following morning.
Introduction
Duty hour restrictions are common in many industries, such as the airline industry, because of concerns that long work hours may affect job performance.1,2,3,4 In medicine, concerns that long work hours may affect patient safety led the Accreditation Council for Graduate Medical Education to impose strict work hour rules for resident physicians.5 However, no such restrictions exist for attending physicians whose workload may have increased in light of the Accreditation Council for Graduate Medical Education’s limits on residents’ work hours.6,7 In particular, an important question is whether overnight work performed by an attending physician is associated with worse outcomes if the attending physician continues to work on the subsequent day. Understanding the answer to this question has important implications for setting policies at both the hospital and the national level.
The extent to which overnight work among attending physicians is associated with worse outcomes on the subsequent day is not fully understood, particularly when large-scale data and empirical approaches that compare the same physicians in periods when they are exposed to varying levels of overnight work are used. Studies have shown that sleep-deprived medical residents are more likely to commit errors,8,9 but those studies were limited to medical residents and did not examine whether the errors resulted in worse patient outcomes.
Within surgery, a field in which overnight work occurs because of surgical emergencies, a 2018 systematic review10 found little overall association between operating overnight and surgical outcomes for operations performed on the subsequent day; however, many of the underlying studies were too small to detect significant associations concerning mortality or surgical complications. A large retrospective study11 found that attending surgeons who had worked between the hours of midnight and 7 am did not have higher complication rates when they operated on patients the next day, but the sample was limited to surgeons in a single Canadian province and may not apply to care in the US, where surgical volumes and pressures facing surgeons are reportedly higher.12 In addition, the study relied on self-reported overnight work and, because it did not ascertain the actual amount of overnight operative work (ie, whether the surgeon operated for 1 or 7 hours), it is vulnerable to dilution bias (ie, a bias toward finding no effect because working 30 minutes overnight was counted the same as working 7 hours). Moreover, the investigators were not able to evaluate any dose-response relationship (ie, whether working more hours overnight was associated with worse outcomes). An additional methodological limitation of previous studies is potential selection bias. Previous studies typically compare procedures performed by surgeons who worked the previous night with procedures performed by surgeons who did not, but surgeons who opt to take overnight call may differ in experience and skill from those who do not, and their patients may differ, creating a problem of confounding.
With the use of data from the Multicenter Perioperative Outcomes Group (MPOG),13 a large multicenter registry of surgical procedures mostly performed in the US, this study examined the association between operating overnight and the risk of mortality and surgical complications for operations performed the next day. Because the registry contained comprehensive information for all operations performed by a given surgeon, we could address confounding by comparing a given surgeon’s outcomes for daytime operations when the surgeon worked the previous night with operations performed when the surgeon did not work the previous night, a within-surgeon analysis.
Methods
Data
In this cross-sectional study, we used the MPOG, a registry of all surgical and diagnostic procedures requiring anesthesia care from more than 50 hospitals across 18 states and 2 countries (US and the Netherlands). The registry has received approval from the institutional review board at the University of Michigan and contributing centers (eAppendix in the Supplement). Because the data are deidentified, the present study was deemed exempt from human participants review by the institutional review board at Stanford University, which also issued a waiver of the requirement for consent. For each surgical or diagnostic procedure, the MPOG includes detailed information, such as surgical start and stop times (corresponding to incision and completion of surgical closure), the Current Procedure Terminology (CPT) code for the given surgery, and an encrypted attending surgeon identifier. The MPOG also reports whether an in-hospital death occurred and provides International Classification of Diseases, Ninth Revision (ICD-9) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnosis codes from the hospital discharge abstract, allowing for risk adjustment and outcome measurement. The present study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies.
In contrast to many other data sets, the MPOG data are comprehensive and include all of a given surgeon’s operations that were performed at the reporting institution. To ensure accuracy, the data must satisfy an array of quality diagnostics before they can be used by researchers,14 and every month, a sample of operations submitted to MPOG from a given institution is reviewed by a clinician or data abstractor located at the submitting institution. The MPOG data have been used in several perioperative outcome studies.15,16,17,18,19,20
Sample
The initial sample consisted of 2 110 956 surgical procedures performed between January 1, 2010, and August 30, 2020. This initial sample was restricted to patients for whom data on comorbidities, mortality, hours the attending surgeon worked the previous night, and surgery CPT codes were present and patients who had a major therapeutic procedure as defined by the Agency for Healthcare Research and Quality.21 We then applied the following exclusion criteria. First, because our methodological approach compared a given surgeon’s outcomes for procedures when the surgeon worked the previous night with procedures when the surgeon did not, we excluded procedures performed by surgeons who never operated overnight (n = 1 332 069). Second, we excluded procedures performed by more than 1 attending surgeon (n = 54 838). Third, we excluded procedures with missing data: patient sex (n = 89) and American Society of Anesthesiologists (ASA) physical status score (n = 7902). Fourth, we excluded procedures with an ASA physical status score of 5, indicating a moribund patient, or 6, indicating an organ donor (n = 1351). Fifth, we excluded the following surgical procedures because they are either commonly performed outside of the operating room or are very low risk: upper endoscopy (CPT codes 43180-43278; n = 47 127),22 lower endoscopy (CPT codes 45300-45393; n = 22 262),23 ophthalmologic surgical procedures (CPT codes 65091-68899; n = 41 581),24 and central venous access procedures (CPT codes 36555-36598; n = 16 435).25 Given our use of fixed effects (ie, indicator variables) for procedures and individual surgeons in the statistical analysis, we excluded CPT codes for which there were fewer than 100 observations (n = 58 753) and procedures performed by surgeons for whom there were fewer than 100 observations (n = 30 295). Our final sample consisted of 498 234 daytime procedures performed by 1131 surgeons at 20 institutions located in the US. A diagram outlining the construction of the sample can be found in the eFigure in the Supplement.
Exposure
For each procedure, the binary exposure variable of interest was whether the attending surgeon had worked the previous night, defined as having operated at any time between 11 pm and 7 am. We chose this time frame because elective procedures in most hospitals start around 7 am; thus, 7 am represents the start of the daytime period and the end of most overnight call shifts. This variable was constructed based on surgical start and stop times recorded for each operation and was constructed using all MPOG data, including procedures that were excluded from the final sample. In addition to this binary exposure, alternative exposure definitions were analyzed and are described in the sensitivity analysis.
Outcomes
The primary outcome was a binary composite measure, taking on a value of 1 if the patient died or experienced a major complication during the inpatient stay. Mortality data were directly obtained from the MPOG. Major complications included thromboembolic events, myocardial infarction, stroke, pneumonia, and sepsis and were identified based on ICD-9 and ICD-10 codes present in the discharge abstract.21 Secondary outcomes included operation length (the difference between start and stop times), as well as each of the following individual outcomes: death, major complications, and minor complications. Minor complications consisted of surgical site infections and urinary tract infections.21 In the case of operation length, we excluded 27 764 procedures with missing data on start and stop times.
Additional Variables
Several additional variables were included to adjust for potential confounders: patient age and sex, year of surgery, and ASA physical status score. The ASA physical status score was reported by the attending anesthesiologist and served as a summary measure of the patient’s health status. The score ranges from 1 (healthy patient) to 6 (organ donor), with a letter “E” identifying operations for an emergency indication (eg, “1E” indicates a healthy patient undergoing an emergency operation). In addition, discharge diagnosis codes were used to identify the presence of the comorbidities included in the Elixhauser Index (Table 1).26
Table 1. Characteristics of the Study Samplea.
| Characteristic | Overnight work | P valueb | Hedges gc | |
|---|---|---|---|---|
| Yes (n = 13 098) | No (n = 485 136) | |||
| Patient demographic characteristics | ||||
| Age, mean (SE), y | 53.91 (0.15) | 55.32 (0.02) | <.001 | 0.09 |
| Female, % (SE) | 47.67 (0.44) | 53.28 (0.07) | <.001 | 0.11 |
| Male, % (SE) | 52.32 (0.44) | 46.72 (0.07) | <.001 | 0.11 |
| Comorbidity, % (SE) | ||||
| Congestive heart failure | 8.73 (0.25) | 5.32 (0.03) | <.001 | 0.15 |
| Arrhythmia | 18.46 (0.34) | 13.07 (0.05) | <.001 | 0.16 |
| Valvular disease | 7.72 (0.23) | 5.18 (0.03) | <.001 | 0.11 |
| Pulmonary circulatory disorders | 4.57 (0.18) | 2.94 (0.02) | <.001 | 0.10 |
| Peripheral vascular disease | 9.04 (0.25) | 5.36 (0.03) | <.001 | 0.16 |
| Hypertension, uncomplicated | 32.75 (0.41) | 29.38 (0.07) | <.001 | 0.07 |
| Hypertension, complicated | 11.70 (0.28) | 7.16 (0.04) | <.001 | 0.17 |
| Paralysis | 1.70 (0.11) | 1.08 (0.01) | <.001 | 0.06 |
| Neurologic disorder | 6.66 (0.22) | 4.55 (0.03) | <.001 | 0.10 |
| Chronic pulmonary disease | 14.05 (0.30) | 11.61 (0.05) | <.001 | 0.08 |
| Diabetes, uncomplicated | 11.66 (0.28) | 9.95 (0.04) | <.001 | 0.06 |
| Diabetes, complicated | 5.10 (0.19) | 3.09 (0.02) | <.001 | 0.12 |
| Hypothyroidism | 7.66 (0.23) | 8.29 (0.04) | .01 | 0.02 |
| Chronic renal failure | 12.35 (0.29) | 7.73 (0.04) | <.001 | 0.17 |
| Liver disease | 5.47 (0.20) | 4.00 (0.03) | <.001 | 0.07 |
| Peptic ulcer disease | 0.73 (0.07) | 0.58 (0.01) | .03 | 0.02 |
| AIDS | 0.41 (0.06) | 0.24 (0.01) | <.001 | 0.03 |
| Lymphoma | 1.00 (0.09) | 1.38 (0.02) | <.001 | 0.03 |
| Metastatic disease | 6.06 (0.21) | 13.99 (0.05) | <.001 | 0.23 |
| Solid tumor | 13.83 (0.30) | 24.65 (0.06) | <.001 | 0.25 |
| Rheumatoid arthritis | 2.79 (0.14) | 2.32 (0.02) | <.001 | 0.03 |
| Coagulopathy | 7.59 (0.23) | 4.23 (0.03) | <.001 | 0.17 |
| Obesity | 16.44 (0.32) | 14.96 (0.05) | <.001 | 0.04 |
| Weight loss | 5.79 (0.20) | 3.81 (0.03) | <.001 | 0.10 |
| Fluid or electrolyte disorder | 17.58 (0.33) | 9.94 (0.04) | <.001 | 0.25 |
| Blood loss | 1.21 (0.10) | 0.85 (0.01) | <.001 | 0.04 |
| Iron deficiency anemia | 2.72 (0.14) | 2.07 (0.02) | <.001 | 0.05 |
| Alcohol use disorder | 1.12 (0.09) | 0.66 (0.01) | <.001 | 0.06 |
| Drug use disorder | 4.21 (0.18) | 2.42 (0.02) | <.001 | 0.11 |
| Psychosis | 0.75 (0.08) | 0.51 (0.01) | <.001 | 0.03 |
| Depression | 11.35 (0.28) | 9.95 (0.04) | <.001 | 0.05 |
| ASA physical status scored | ||||
| Physical status score, mean (SE) | 2.62 (0.01) | 2.53 (0.00) | <.001 | 0.13 |
| Emergency procedure, % (SE) | 8.70 (0.25) | 2.60 (0.02) | <.001 | 0.37 |
| High-risk patients, % (SE) | 35.9 (0.4) | 24.7 (0.1) | <.001 | 0.26 |
Abbreviation: ASA, American Society of Anesthesiologists.
Table presents characteristics of daytime operations in the study sample, stratified by procedures performed when the attending surgeon worked the previous night (between 11 pm and 7 am, overnight work) and daytime procedures performed when the attending surgeon did not.
The P values refer to the statistical significance of differences between the 2 groups.
Hedges g refers to the effect size.
The ASA physical status score was provided by the attending anesthesiologist and was a summary measure of the patient’s health status, with a score of 1 indicating a perfectly healthy patient and a score of 6 indicating an organ donor. High-risk patients were patients who were in the upper 25% of estimated risk for mortality or major complication based on the patient’s comorbidities, age, sex, and type of operation.
Statistical Analysis
For each outcome, multivariable linear regressions were estimated in which the independent variable of interest was whether the surgeon worked the previous night. To address potential confounding due to differences between surgeons who do and do not take overnight call, the models included surgeon fixed effects (ie, indicator variables for each surgeon), thereby comparing a given surgeon’s outcomes for daytime procedures when the surgeon worked the previous night with daytime procedures when the same surgeon did not. In addition to adjusting for potential confounding due to differences between surgeons who did and did not take overnight call, this within-surgeon approach also implicitly adjusted for differences across institutions (ie, unobservable differences in patient acuity, differences in operating room staffing) that could confound our results. Although most of the outcomes were discrete variables, linear probability models were estimated because the inclusion of a large number of fixed effects can lead to biased estimates in nonlinear models such as logistic regression27 and can also create issues with complete or quasi-complete separation.28,29
In addition to surgeon fixed effects and our binary exposure variable, covariates included fixed effects for operation type (surgical CPT code) and the aforementioned patient characteristics (ie, indicator variables for each of the comorbidities listed in Table 1). Moreover, because surgeons may have become more fatigued as the day progressed, the model also included fixed effects for the hour at which the operation started. This model therefore compared outcomes of daytime procedures when the attending surgeon had operated overnight with the estimated outcomes for similar daytime procedures (ie, similar procedures type and start time and patient comorbidities) when the same surgeon did not operate overnight.
A prespecified analysis plan outlining the details of the statistical analysis was approved and registered by the MPOG Perioperative Clinical Research Committee prior to data acquisition. All analyses were performed with Stata, version 14.0 (StataCorp LLC). Standard errors were clustered at the surgeon level, and 2-sided P values were used to assess statistical significance, with a threshold of .05.
Subgroup Analyses
A prespecified exploratory subgroup analysis on high-risk patients was performed because these patients may be at highest risk from a surgeon’s fatigue due to the complex condition of the patient or the complexity of the surgical procedure. To do so, a multivariable logistic regression was used to estimate the predicted probability of in-hospital death or in-hospital postoperative complications on the basis of age, sex, ASA physical status score, surgical procedure, and the comorbidities listed in Table 1. High-risk patients were defined as those in the top quartile of predicted probability for either in-hospital death or major postoperative complications (n = 124 559; predicted probability ≥6.5%).
We also performed several post hoc subgroup analyses for the following surgical subspecialties: cardiovascular surgery (CPT codes 33016-37799), orthopedic surgery (CPT codes 20100-29999), urology (CPT codes 50010-55899), and neurosurgery (CPT codes 61000-64999). We chose these subspecialties because they are common subspecialties representing a wide range of underlying procedure risk.
Sensitivity Analyses
The primary exposure was a binary variable equal to 1 if a surgeon operated for any amount of time between 11 pm and 7 am the previous night. A sensitivity analysis modeled the amount of time worked during this period using 2 alternate approaches. The first allowed for a nonlinear association between hours worked the night before and surgical outcomes (eg, an additional hour of work may matter more if it is the sixth vs the first hour of work). Five mutually exclusive and exhaustive categories were created based on the total amount of operative time between 11 pm and 7 am: no work, less than 2 hours, 2 hours to less than 4 hours, 4 hours to less than 6 hours, and 6 to 8 hours. The second approach estimated a linear and continuous association between the number of hours worked between 11 pm and 7 am and the primary outcome.
In addition to these analyses, we performed several sensitivity analyses related to changes in our statistical approach. First, we performed an analysis in which we excluded surgeon fixed effects. Second, we performed an analysis in which we used a multivariable logistic regression; this regression used the same set of explanatory variables as for our linear regression analysis. Third, we performed an analysis in which we retained surgeons who performed at least 25 procedures as opposed to our baseline analysis, which imposed a threshold of at least 100 procedures. Fourth, we performed an analysis in which we clustered errors at the institution level.
Results
Overall, 498 234 daytime procedures involving 1131 unique surgeons were analyzed (mean [SD] patient age, 55.3 years [16.4 years]; 233 494 [46.8%] of the patients were male and 264 740 [53.1%] were female). In 13 098 of these procedures (2.6% of total), the attending surgeon operated between 11 pm and 7 am the previous night. Among those operations, the mean (SD) time worked during the previous night was 107 (129) minutes (median, 57 minutes; IQR, 14-148 minutes). The 5 most common daytime surgical CPT codes were 55866 (laparoscopic prostatectomy; n = 9118), 52 332 (cystourethroscopy; n = 8192), 20 680 (implant removal; n = 8098), 27 447 (total knee arthroplasty; n = 7850), and 38 525 (lymph node biopsy; n = 7684). A list of the 50 most common daytime surgical CPT codes is provided in eTable 1 in the Supplement.
Daytime procedures performed when the attending surgeon operated the night before were more likely to involve patients with comorbidities (Table 1); however, the differences were generally small in magnitude (Hedges g value <0.2), with the exception of procedures denoted as urgent by the attending anesthesiologist (8.70% for procedures that followed overnight operative work vs 2.60% for those that did not; P < .001; Hedges g = 0.37), high-risk patients, defined as patients in the upper quartile of estimated risk for mortality or a major complication (35.9% for procedures that followed overnight operative work vs 24.7% for those that did not; P < .001; Hedges g = 0.26), solid tumors (13.83% for procedures that followed overnight operative work vs 24.65% for those that did not; P < .001; Hedges g = 0.25), and metastatic disease (6.06% for procedures that followed overnight operative work vs 13.99% for those that did not; P < .001; Hedges g = 0.23). These differences tended to be smaller in magnitude and were less likely to be statistically significant when adjusted for surgeon fixed effects, an analysis that mirrors the study’s statistical design by comparing patient characteristics for daytime procedures with and without previous overnight work done by the same surgeon (eTable 2 in the Supplement).
The unadjusted incidence of in-hospital death or major complication was 9.83% (1287 of 13 098; 95% CI, 8.44%-11.21%) among daytime procedures when the attending surgeon worked the previous night compared with 5.76% (27 945 of 485 136; 95% CI, 5.28%-6.24%) among daytime procedures when the attending surgeon did not (absolute difference, 4.06%; 95% CI, 2.86%-5.27%; P < .001). However, after adjustment for potential confounders, the incidence of in-hospital death or major complication was 5.89% (95% CI, 5.41%-6.36%; Table 2) among daytime procedures performed by surgeons who operated the previous night compared with 5.87% (95% CI, 5.85%-5.89%) among daytime procedures performed by surgeons who did not (absolute adjusted difference, 0.021%; 95% CI, −0.47% to 0.51%; P = .93).
Table 2. Overnight Work by Surgeons and Incidence of Secondary Outcomes for Daytime Proceduresa.
| Outcomes for daytime procedures | Overnight work | Absolute difference, % (95% CI) | P value | |
|---|---|---|---|---|
| Yes (n = 13 098) | No (n = 485 136) | |||
| Primary outcome (in-hospital death or major complication), % (95% CI) | ||||
| Unadjusted analysis | 9.83 (8.44 to 11.21) | 5.76 (5.28 to 6.24) | 4.06 (2.86-5.87) | <.001 |
| Adjusted analysisb | 5.89 (5.41 to 6.36) | 5.87 (5.85 to 5.89) | 0.021 (−0.47 to 0.51) | .93 |
| Operation length, mean (95% CI), minc | ||||
| Unadjusted analysis | 120.2 (111.0 to 129.4) | 117.2 (112.6 to 121.8) | 3.0 (−5.3 to 11.3) | .48 |
| Adjusted analysisb | 112.7 (108.8 to 116.5) | 117.4 (117.3 to 117.5) | −4.7 (−8.7 to −0.8) | .02 |
| In-hospital death, % (95% CI) | ||||
| Unadjusted analysis | 1.47 (1.15 to 1.79) | 0.88 (0.77 to 0.98) | 0.59 (0.29 to 0.88) | <.001 |
| Adjusted analysisb | 0.87 (0.68 to 1.06) | 0.89 (0.88 to 0.90) | −0.02 (−0.22 to 0.17) | .82 |
| Major complications, % (95% CI)d | ||||
| Unadjusted analysis | 9.10 (7.78 to 10.41) | 5.24 (4.80 to 5.69) | 3.86 (2.71 to 5.00) | <.001 |
| Adjusted analysisb | 5.40 (4.92 to 5.87) | 5.34 (5.33 to 5.35) | 0.05 (−0.43 to 0.54) | .83 |
| Minor complications, % (95% CI)e | ||||
| Unadjusted analysis | 2.99 (2.54 to 3.45) | 2.25 (2.06 to 2.44) | 0.74 (0.33 to 1.15) | <.001 |
| Adjusted analysisb | 2.34 (2.03 to 2.64) | 2.27 (2.26 to 2.28) | 0.07 (−0.24 to 0.39) | .67 |
Table presents unadjusted and adjusted means of the primary outcome (composite of in-hospital death or major complication), operation length, incidence of in-hospital death, and incidence of in-hospital complications (major and minor). Overnight work refers to daytime procedures performed when the attending surgeon worked the previous night between 11 pm and 7 am; no overnight work refers to daytime procedures performed when the attending surgeon did not.
Adjusted for patient characteristics, operation type, and surgeon fixed effects; 95% CIs were adjusted for clustering at the surgeon level.
Defined as the difference between start and stop times.
Included thromboembolic events, myocardial infarction, stroke, pneumonia, sepsis, and stroke.
Included surgical site infections and urinary tract infections.
After adjusting for potential confounders, no significant association was observed between overnight work and most of the secondary outcomes we examined, such as the incidence of death, minor complications, and major complications (Table 2). The sole exception was operation length: operating the previous night was associated with a statistically significant adjusted decrease in length of daytime procedures (adjusted length, 112.7 vs 117.4 minutes; adjusted difference, −4.7 minutes; 95% CI, −8.7 to −0.8, P = .02), although this difference is unlikely to be meaningful. In a subgroup analysis confined to patients at high estimated risk for the primary outcome, the adjusted incidence of death or major complication was 20.68% (95% CI, 19.49%-21.86%) for daytime procedures when the attending surgeon worked the previous night compared with 20.41% (95% CI, 20.36%-20.45%) for daytime procedures when the attending surgeon did not (adjusted difference, −0.27%; 95% CI, −0.96% to 1.50%; P = .67).
Sensitivity analyses also suggested no difference between overnight work and the primary outcome. In the first sensitivity analysis, in which procedures were grouped into 1 of 5 categories based on the amount of time the attending surgeon worked the previous night (no work, <2 hours, 2 to <4 hours, 4 to <6 hours, and 6 to 8 hours), there was no statistically significant difference in the incidence of death or major complication for daytime procedures in any of the overnight procedure–length groups compared with those in the no-work group (Table 3). In the second sensitivity analysis, each additional hour worked the previous night was associated with a statistically nonsignificant decrease in the probability of death or a major complication for daytime procedures (absolute adjusted difference, −0.05% per hour; 95% CI, −0.22% to 0.12% per hour; P = .57) (Table 3). Several sensitivity analyses related to alternative model specifications, such as the use of a logistic regression, produced results that were qualitatively similar to our main findings (eTable 3 in the Supplement).
Table 3. Overnight Work by Surgeons and Incidence of Death or Major Complication for Daytime Procedures, Alternative Definitions of Overnight Work Exposurea.
| Length of time worked between 11 pm and 7 am the previous night | Adjusted incidence of death or major complication, % (95% CI) | Absolute adjusted difference, % (95% CI) | P value |
|---|---|---|---|
| Hours operated night before, modeled as a categorical variable | |||
| None (n = 485 136) | 5.87 (5.85 to 5.88) | [Reference] | NA |
| >0 to <2 h (n = 9117) | 6.05 (5.50 to 6.60) | 0.18 (−0.38 to 0.75) | .52 |
| 2 to <4 h (n = 2116) | 5.47 (4.35 to 6.59) | −0.40 (−1.52 to 0.72) | .49 |
| 4 to <6 h (n = 883) | 6.01 (4.17 to 7.85) | 0.15 (−1.70 to 1.99) | .88 |
| 6-8 h (n = 982) | 5.21 (3.60 to 6.83) | −0.65 (−2.27 to 0.97) | .43 |
Abbreviation: NA, not applicable.
Table shows the results of an analysis in which daytime procedures were classified into 1 of 5 categories based on the amount of time the surgeon worked the previous night. A second analysis, modeled linearly, calculated the adjusted increase in risk per additional hour of overnight work: −0.05% per hour (95% CI, −0.22% to 0.12% per hour; P = .57). All analyses were adjusted for patient characteristics, procedure type, and surgeon fixed effects. The 95% CIs were adjusted for clustering at the surgeon level.
Discussion
The association between physician fatigue and patient outcomes is important to understand but has been difficult to examine given methodological and data limitations. Surgeons frequently work overnight performing urgent procedures and perform additional procedures the following day, providing an opportunity to examine whether fatigue due to overnight work is associated with adverse patient outcomes. In this cross-sectional study, we examined whether performing operations overnight was associated with worse outcomes for operations the same surgeon performed the next day. Our analysis revealed no such associations, either for patients overall or for those estimated to be at the highest risk of death or surgical complications. This finding was robust to alternative definitions of overnight work and held across a variety of outcomes. The lack of association is unlikely to be due to statistical imprecision because both the estimated effects and the upper bounds of the CIs were small in magnitude. This lack of association is striking because according to several measures—such as estimated risk of death or complications—daytime procedures performed when the surgeon had operated the previous night were associated with higher risk compared with procedures when the same surgeon did not. Our data therefore suggest that, in general, surgeons who continued to operate even after having operated the previous night tended to perform only higher-risk, more urgent procedures and that despite this, continuing to operate the next day was not associated with worse outcomes.
Previous studies have generally found no association between surgeon fatigue and perioperative outcomes,10,12 but most have had significant limitations. First, because the incidences of death and surgical complications are rare, many studies were not sufficiently powered to identify statistically significant results effects. To our knowledge, the present study’s sample was the largest analyzed to date. Many previous studies also compared outcomes across surgeons rather than outcomes among a given surgeon’s procedures, which could lead to bias if surgeons who take call and perform procedures overnight differ in experience and skill or if their surgical case mix differs from the case mix of those who do not. In addition, several previous studies examined Canadian surgeons only, relied on self-reports of overnight work, and did not measure the amount of time a surgeon operated overnight. Our study addressed these limitations by using a large multicenter registry of US institutions with information on all procedures performed by surgeons at those institutions, allowing us to explicitly compare a given surgeon’s outcomes for procedures when the surgeon did and did not work the previous night.
Limitations
This study has limitations. First, many of the centers were academic institutions, and the results may not apply to nonacademic settings with different staffing models or policies regarding overnight work. Second, although our empirical approach addressed many potential confounders, additional factors, such as unobserved patient characteristics or unobserved time-varying surgeon characteristics, may have introduced bias to our results. In particular, daytime procedures performed after a surgeon had operated overnight could differ from a surgeon’s usual daytime surgical caseload in many ways, some of which cannot be captured by the variables available in the data set. For example, one possibility is that surgeons could preferentially schedule less risky procedures on postcall days. However, the data suggest this was not the case in this sample. Based on observable factors, such as patient comorbidities and type of surgical procedure, operations when the surgeon operated the night before tended to have higher, not lower, estimated mortality or complications than other procedures. Third, the MPOG data were limited to in-hospital outcomes and did not permit examination of whether a surgeon’s overnight work was associated with worse postdischarge outcomes, such as readmission rates. Fourth, the study focused on differences in patient outcomes rather than problems that might emerge only during a surgical procedure, such as increased transfusion rates. However, there was no clinically significant difference in surgery length for procedures performed when the surgeon worked the previous night compared with procedures performed when the same surgeon did not, suggesting that intraoperative problems were unlikely to occur more often in instances when the surgeon worked the night before. Fifth, the empirical approach compared outcomes of procedures performed by the same surgeon in instances when the surgeon worked vs did not work the night before, an association that was, by definition, estimated among only those surgeons who chose to perform both overnight and daytime operations. Overnight work among surgeons who do not typically operate overnight may have different repercussions for patient outcomes. Sixth, the study did not examine whether any potential associations with overnight work varied across subgroups of surgeons (eg, based on age, experience, or surgical specialty). Seventh, the study did not measure time spent on nonoperative types of overnight work, such as performing consultations. In particular, most procedures the surgeon performed overnight may have required additional work besides the operation itself, such as obtaining the history and performing a physical examination or providing a consultation note; therefore, our study may have underestimated the extent to which an attending surgeon worked overnight. Along these lines, our study was unable to ascertain the extent to which a surgeon may have worked overnight at other institutions not included in our sample. Eighth, the study did not consider other outcome measures, such as the need for reoperation or readmission. Ninth, the study focused on surgeons; the association between physician fatigue and patient outcomes may differ among various specialties and practice settings. The study also did not include consideration of the potential associations of fatigue with other measures of physician well-being, such as physician burnout.
Conclusions
In this cross-sectional study, a large multicenter registry of surgical procedures was used to compare outcomes of daytime procedures in which a surgeon worked the previous night with daytime procedures when the same surgeon did not, and no association was found between perioperative outcomes and overnight work. Combined with previous studies, these results provide reassurance concerning the practice of having attending surgeons take overnight call and still perform procedures the following morning. Although our results do not establish that this practice is always safe or that fatigue does not affect outcomes, they suggest that the potential risk was managed well enough to avoid patient harm in this sample of surgeons. Therefore, these results suggest that policies limiting attending surgeon work hours may not be necessary, so long as sufficient policies and incentives exist to ensure that institutions and surgeons manage any potential risk from fatigue. Whether these findings apply to settings other than those studied here, as well as how fatigue affects clinicians, is an area for future research.
eAppendix. MPOG (Multicenter Perioperative Outcomes Group) Member Hospitals
eFigure. Sample Construction Flow Diagram
eTable 1. 50 Most Common Surgical Procedures
eTable 2. Characteristics of the Study Sample After Adjusting for Surgeon Characteristics
eTable 3. Additional Sensitivity and Subgroup Analyses
References
- 1.Neu D, Kajosch H, Peigneux P, Verbanck P, Linkowski P, Le Bon O. Cognitive impairment in fatigue and sleepiness associated conditions. Psychiatry Res. 2011;189(1):128-134. doi: 10.1016/j.psychres.2010.12.005 [DOI] [PubMed] [Google Scholar]
- 2.Goode JH. Are pilots at risk of accidents due to fatigue? J Safety Res. 2003;34(3):309-313. doi: 10.1016/S0022-4375(03)00033-1 [DOI] [PubMed] [Google Scholar]
- 3.Fullagar HH, Skorski S, Duffield R, Hammes D, Coutts AJ, Meyer T. Sleep and athletic performance: the effects of sleep loss on exercise performance, and physiological and cognitive responses to exercise. Sports Med. 2015;45(2):161-186. doi: 10.1007/s40279-014-0260-0 [DOI] [PubMed] [Google Scholar]
- 4.Federal Aviation Administration; Department of Transportation. Flightcrew member duty and rest requirements. Fed Regist. 2014;79(236):72970-72975. [Google Scholar]
- 5.Accreditation Council for Graduate Medical Education.Common Program Requirements. Accessed April 15, 2022. https://www.acgme.org/what-we-do/accreditation/common-program-requirements/ [Google Scholar]
- 6.Vanderveen K, Chen M, Scherer L. Effects of resident duty-hours restrictions on surgical and nonsurgical teaching faculty. Arch Surg. 2007;142(8):759-764. doi: 10.1001/archsurg.142.8.759 [DOI] [PubMed] [Google Scholar]
- 7.Hutter MM, Kellogg KC, Ferguson CM, Abbott WM, Warshaw AL. The impact of the 80-hour resident workweek on surgical residents and attending surgeons. Ann Surg. 2006;243(6):864-871. doi: 10.1097/01.sla.0000220042.48310.66 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Barger LK, Ayas NT, Cade BE, et al. Impact of extended-duration shifts on medical errors, adverse events, and attentional failures. PLoS Med. 2006;3(12):e487. doi: 10.1371/journal.pmed.0030487 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Taffinder NJ, McManus IC, Gul Y, Russell RC, Darzi A. Effect of sleep deprivation on surgeons’ dexterity on laparoscopy simulator. Lancet. 1998;352(9135):1191. doi: 10.1016/S0140-6736(98)00034-8 [DOI] [PubMed] [Google Scholar]
- 10.Gates M, Wingert A, Featherstone R, Samuels C, Simon C, Dyson MP. Impact of fatigue and insufficient sleep on physician and patient outcomes: a systematic review. BMJ Open. 2018;8(9):e021967. doi: 10.1136/bmjopen-2018-021967 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Govindarajan A, Urbach DR, Kumar M, et al. Outcomes of daytime procedures performed by attending surgeons after night work. N Engl J Med. 2015;373(9):845-853. doi: 10.1056/NEJMsa1415994 [DOI] [PubMed] [Google Scholar]
- 12.Pozen A, Cutler DM. Medical spending differences in the United States and Canada: the role of prices, procedures, and administrative expenses. Inquiry. 2010;47(2):124-134. doi: 10.5034/inquiryjrnl_47.02.124 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Multicenter Perioperative Outcomes Group. Home page. Accessed April 15, 2022. https://mpog.org
- 14.Colquhoun DA, Shanks AM, Kapeles SR, et al. Considerations for integration of perioperative electronic health records across institutions for research and quality improvement: the approach taken by the Multicenter Perioperative Outcomes Group. Anesth Analg. 2020;130(5):1133-1146. doi: 10.1213/ANE.0000000000004489 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lee LO, Bateman BT, Kheterpal S, et al. ; Multicenter Perioperative Outcomes Group Investigators . Risk of epidural hematoma after neuraxial techniques in thrombocytopenic parturients: a report from the Multicenter Perioperative Outcomes Group. Anesthesiology. 2017;126(6):1053-1063. doi: 10.1097/ALN.0000000000001630 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.de Graaff JC, Pasma W, van Buuren S, et al. Reference values for noninvasive blood pressure in children during anesthesia: a multicentered retrospective observational cohort study. Anesthesiology. 2016;125(5):904-913. doi: 10.1097/ALN.0000000000001310 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Berman MF, Iyer N, Freudzon L, et al. ; Multicenter Perioperative Outcomes Group (MPOG) Perioperative Clinical Research Committee . Alarm limits for intraoperative drug infusions: a report from the Multicenter Perioperative Outcomes Group. Anesth Analg. 2017;125(4):1203-1211. doi: 10.1213/ANE.0000000000002305 [DOI] [PubMed] [Google Scholar]
- 18.Aziz MF, Brambrink AM, Healy DW, et al. Success of intubation rescue techniques after failed direct laryngoscopy in adults: a retrospective comparative analysis from the Multicenter Perioperative Outcomes Group. Anesthesiology. 2016;125(4):656-666. doi: 10.1097/ALN.0000000000001267 [DOI] [PubMed] [Google Scholar]
- 19.Kheterpal S, Healy D, Aziz MF, et al. ; Multicenter Perioperative Outcomes Group (MPOG) Perioperative Clinical Research Committee . Incidence, predictors, and outcome of difficult mask ventilation combined with difficult laryngoscopy: a report from the Multicenter Perioperative Outcomes Group. Anesthesiology. 2013;119(6):1360-1369. doi: 10.1097/ALN.0000435832.39353.20 [DOI] [PubMed] [Google Scholar]
- 20.Sun E, Mello MM, Rishel CA, et al. ; Multicenter Perioperative Outcomes Group (MPOG) . Association of overlapping surgery with perioperative outcomes. JAMA. 2019;321(8):762-772. doi: 10.1001/jama.2019.0711 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Agency for Healthcare Research and Quality. Surgery flags software for services and procedures . 2020. Accessed December 10, 2020. https://www.hcup-us.ahrq.gov/toolssoftware/surgeryflags_svcproc/surgeryflagssvc_proc.jsp
- 22.Codify by AAPC. Surgical procedures on the esophagus CPT® code range 43020-43499. 2021. Accessed June 11, 2021. https://www.aapc.com/codes/cpt-codes-range/43020-43499/
- 23.Codify by AAPC. Surgical procedures on the colon and rectum CPT® code range 45000-45999. 2021. Accessed June 11, 2021. https://www.aapc.com/codes/cpt-codes-range/45000-45999/
- 24.Codify by AAPC. Surgery CPT® code range 10004-69990. 2021. Accessed June 11, 2021. https://www.aapc.com/codes/cpt-codes-range/10004-69990/10
- 25.Codify by AAPC. Vascular introduction and injection procedures . 2021. Accessed June 11, 2021. https://www.aapc.com/codes/cpt-codes-range/36000-36598/
- 26.Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11):1130-1139. doi: 10.1097/01.mlr.0000182534.19832.83 [DOI] [PubMed] [Google Scholar]
- 27.Greene W. The behaviour of the maximum likelihood estimator of limited dependent variable models in the presence of fixed effects. Econom J. 2004;7(1):98-119. doi: 10.1111/j.1368-423X.2004.00123.x [DOI] [Google Scholar]
- 28.Hellevik O. Linear versus logistic regression when the dependent variable is a dichotomy. Qual Quant. 2009;43(1):59-74. doi: 10.1007/s11135-007-9077-3 [DOI] [Google Scholar]
- 29.Wooldridge JM. Econometric Analysis of Cross Section and Panel Data. MIT Press; 2010. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eAppendix. MPOG (Multicenter Perioperative Outcomes Group) Member Hospitals
eFigure. Sample Construction Flow Diagram
eTable 1. 50 Most Common Surgical Procedures
eTable 2. Characteristics of the Study Sample After Adjusting for Surgeon Characteristics
eTable 3. Additional Sensitivity and Subgroup Analyses
