Risks of Complications by Attending Physicians After Performing Nighttime Procedures FREE

Increasing public attention has been focused on the role of resident physicians' fatigue in the occurrence of medical errors, including percutaneous needlestick and laceration injuries and postcall motor vehicle crashes.^1- 3 Less is known about the effects of extended-duration work shifts on the performance of attending physicians. Attending physicians who perform emergency operations or perform deliveries during the night often schedule elective procedures or deliveries the following day. It is possible that experienced attending physicians may be better able to cope with the effects of sleep deprivation than residents. Furthermore, the surgical environment differs sufficiently from other medical environments so findings from medical settings may not be directly applicable.⁴ To date, little research either supports or refutes these hypotheses.⁵

In 2003, the Accreditation Council for Graduate Medical Education implemented resident physicians work hour limits of 30 consecutive work hours and 80 weekly hours.⁶ Unlike residents, the work hours of attending physicians in the United States are not restricted. Little is known about the frequency of elective surgical and obstetrical/gynecologic (ob-gyn) procedures by attending physicians who participate in emergency procedures the night before. Even less is known about the risks of complications during these postnighttime procedures. We therefore conducted a study to address these gaps in knowledge.

The study hospital was a 745-bed urban, tertiary care academic trauma center and referral center for high-risk obstetrical procedures. Procedures conducted in either the operating room or labor and delivery suite (L/D) between January 2000 and December 2007 were eligible for matching with control procedures of the same type that were completed between January 1999 and June 2008. Most surgical procedures included a resident physician or other assistant. The human research committee approved the study.

We conducted a matched retrospective cohort study in which we identified all attending surgeons and obstetricians/gynecologists who were involved in overnight and early morning emergency operating room and L/D procedures prior to a daytime procedure. We matched these daytime procedures (postnighttime procedures) that followed overnight emergency procedures performed by the same attending physician at some point between midnight and 6 AM (overnight index procedures) with daytime procedures (control procedures) that did not follow overnight procedures. Each postnighttime procedure was matched by physician with as many as 5 of his or her most contemporaneous control procedures of the same or similar type. Postnighttime procedures without at least 1 matched control were excluded from analysis. Because of the lower complication rate in spontaneous vaginal deliveries,⁷ we enriched the daytime case sampling with cesarean deliveries.

The primary outcomes were the presence of surgical and ob-gyn procedural complications, defined as adverse events occurring as a result of care during an operation or delivery and likely attributable to the performance of the attending surgeon or obstetrician/gynecologist. Preventable complications were complications judged to have likely been due to a surgical error. We excluded the following: adverse outcomes likely due to the patient's condition (eg, septic shock during repair of infarcted bowel); adverse events judged unlikely to be associated with the attending surgeon or obstetrician/gynecologist (eg, anesthetic-induced hypotension); and adverse events of unclear relationship to intraoperative care (eg, pulmonary embolus).

Surgical complications included infection, massive hemorrhage, organ injury, wound failure, and other complications such as failed procedures. Additional obstetrical/peripartum complications included greater than 1000-mL blood loss, stillbirth, umbilical cord prolapse, shoulder dystocia, birth trauma, and fourth-degree perineal tear.^8- 9

Start and end times (procedure duration) were the time of initial skin incision to surgery end, and for vaginal deliveries were the beginning of the second stage of labor to the end of the third stage. Procedure duration did not include estimates of time spent with the patient preoperatively and postoperatively.

We assessed 2 additional predictors of fatigue in postnighttime procedures.¹⁰ Sleep opportunity was defined as the time period between the end of the last overnight index procedure and the beginning of the first daytime postnighttime procedure. For analysis, we dichotomized sleep opportunity to 0 to 6 hours vs greater than 6 hours. The work duration for each postnighttime procedure was the time from the start of the first overnight index procedure to the end of each postnighttime procedure. Consecutive postnighttime procedures accrued unique and increasingly longer work durations. For analysis, work duration was dichotomized into a 0- to 12-hour category vs a greater than 12-hour one.

Physician age, experience as measured by postgraduate years, sex, and specialty were collected, as were patient age, sex, primary diagnosis, comorbidities, Charlson score, and discharge disposition. Procedure data included outpatient or inpatient status, location of procedure in the operating room or L / D, number of procedures performed by the attending physician, and type of procedure by organ system.

We matched postnighttime procedures with same or similar controls by the same attending physician within 12 months for surgical procedures and within 6 months for ob-gyn procedures. We matched as many as 5 controls for each postnighttime procedure preferentially using the same International Classification of Diseases, Ninth Revision (ICD-9) procedure code, and if necessary, allowed matches of 1 higher coding level (ie, of slightly lesser specificity). Potential control procedures were excluded from matching if they occurred on the same day or 1 day after any postnighttime case. Control cases could be used only once for matching.

We electronically screened for complications using ICD-9-Clinical Modification (CM) periprocedural surgical and ob-gyn complications codes in the Agency for Healthcare Research and Quality Guide to Patient Safety Indicators¹¹ and the Complications Screening Program.¹² We also used a previously validated set of administrative adverse event screens including 30-day readmissions.¹³ We created data warehouse queries using the Obstetrics Adverse Outcome Index¹⁴ to capture additional obstetrical complications including transfusion during delivery.

Following administrative screening, we used a 2-tiered approach for identifying procedural complications.¹⁵ Administrative screen-positive procedures underwent medical record review by 2 trained data abstractors. Interrater reliability testing was conducted prior to data collection. Abstracted cases with suspected complications were independently evaluated by clinician pairs with experience in rating complications. Complications were judged on the likelihood of preventability and severity of injury using a previously developed rating system.¹⁶ Disagreements were resolved by consensus. Chart abstractors and clinician reviewers were blinded to whether incidents occurred during postnighttime or matched control procedures, overnight sleep opportunities, and shift durations.

In analyses relating procedure type (postnighttime or control) with other variables, generalized linear mixed models were used to account for the clustering of patients within attending physicians and within matched sets.^17- 18 We reported means and proportions of the baseline characteristics and used generalized linear mixed models and Wald P values for differences between postnighttime and control procedures. We initially compared differences in outcomes using generalized linear mixed models odds ratios (ORs), P values, and 95% confidence intervals (CIs) without adjusting for any confounders (but accounting for clustering), and then used generalized linear mixed models to compute multivariable-adjusted ORs between outcomes and procedure type and 95% CIs. We also conducted analyses comparing complication rates of postnighttime procedures in which attending physicians had 0 to 6 hours of sleep vs greater than 6-hour sleep opportunities and work duration was 0 to 12 hours vs greater than 12 hours.

A priori confounders included age, comorbidities, and sex for operating room procedures. Clustering is incorporated in generalized linear mixed models such that we did not need to adjust for matching factors. Any covariate was kept in a multivariable model if it led to a 10% change in the OR estimate between outcome and procedure type.

For complication ratings among reviewers, the k coefficient was used to assess interrater reliability. All analyses were conducted using Proc Glimmix or Proc Freq in SAS version 9.2, (SAS Institute Inc, Cary, North Carolina). All tests were 2-tailed, and a P value of less then .05 was considered significant. Since testing for the association of complications with procedure type, sleep opportunity, and work duration were planned a priori, we have not adjusted the type 1 error to account for multiple comparisons; thus, the P values should be interpreted cautiously.

This study had 80% power (α = 5%) to detect a decrease in the complication rate from 4% in postnighttime procedures to 2.65% in control procedures using the generalized linear mixed models Wald test. With the observed complication rate of 5.4% in postnighttime procedures, we had 80% power to detect a decrease to 3.8%.

A total of 217 953 operating room and 69 509 L/D procedures were performed during the study period. We identified 4059 operating room index procedures (1.9%) and 17 886 L/D ones (25.7%) that ended or started between 12 AM and 6 AM (eFigure). There were 1135 operating room and 2945 L/D procedures performed by the same physician on the same day as the index procedure. We were able to match 919 postnighttime operating room procedures (81%) and 957 postnighttime L/D procedures (32.5%) to at least 1 control. We matched an average of 3.9 operating room control procedures with each postnighttime one (n = 3552) and 4.1 L/D control procedures with each postnighttime one (n = 3945). There were no differences between patients who underwent control and postnighttime procedures except that control ob-gyn patients were slightly older than postnighttime ones (Table 1).

Physician characteristics are provided in Table 2 for the 86 surgeons (17.5%) and 134 obstetricians/gynecologists (40.5%) who performed postnighttime and matched control procedures. The mean and median number of postnighttime and control procedures per physician are provided in Table 3. The most common operating room procedures were gynecologic, digestive, and cardiovascular cases.

Results of the electronic administrative screens and medical record reviews are provided in the eFigure. Clinician reviews determined that there were complications in 68 postnighttime operating room procedures (7.4%) and 253 operating room control procedures (7.1%) (OR of a postnighttime vs control procedure having at least 1 complication, 1.02; 95% CI, 0.74-1.40), and 55 and 251 preventable complications, respectively (OR, 0.88; 95% CI, 0.64-1.22; Table 4). We also found complications in 33 postnighttime L/D procedures (3.4%) and 112 L/D control ones (2.8%) (OR of a procedure having at least 1 complication, 1.21;95% CI, 0.79-1.84), and 32 and 100 preventable complications, respectively (OR, 1.30; 95% CI, 0.85-1.99; Table 4). The adjusted ORs from the multivariable model were similar to the unadjusted ORs. The interrater reliability for clinician ratings for the presence, severity, and preventability of complications were good: k = 0.83, 0.78, and 0.60, respectively.

The severity of complications associated with the postnighttime and control operating room procedures was judged as follows: unable to determine or insignificant, 0 and 3 (1.1%); significant, 25 (36.2%) and 112 (40.2%); serious, 42 (60.9%) and 144 (51.7%); life-threatening, 2 (2.9%) and 13 (4.7%); and fatal, 0 and 7 (2.5%), respectively (P = .84).

The severity of complications associated with the postnighttime and control L/D procedures was judged as follows: unable to determine or insignificant, 1 (2.9%) and 2 (1.8%); significant, 25 (71.4%) and 74 (66.1%); serious, 9 (25.7%) and 32 (28.6%); and life-threatening, 0 and 4 (3.6%), respectively (P = .98).

The most common surgical complications were organ injuries and bleeding; surgical site infections and organ injury were the most common obstetrical complications. We did not detect a difference in types of complications between postnighttime and control procedures (Table 5). Complicated postnighttime and control operating room procedures (135 and 129 minutes, respectively) were longer than uncomplicated procedures (64.5 and 66.2 minutes; eTable). Similarly complicated postnighttime and control L/D procedures (71.5 and 60.4 minutes, respectively) were longer than uncomplicated procedures (46.5 and 44.8 minutes). The overall duration of postnighttime L /D procedures (47.2 minutes) was longer than for control procedures (45.2 minutes; P = .002).

Among postnighttime cases, a higher rate of procedural complications occurred when there were 6 or fewer hours of sleep opportunity (6.2%) compared with postnighttime procedures in which there were more than 6 hours of sleep opportunity (3.4%) (OR, 1.72; 95% CI, 1.02-2.89); this was predominantly due to operating room (surgical) complications (8.5% vs 3.1%, respectively). Postnighttime procedures performed when work duration exceeded 12 hours showed nonsignificantly higher complication rates compared with shifts of 12 hours or less (6.5% vs 4.3%; OR, 1.47; 95% CI, 0.96-2.27) (Table 6).

Procedures performed the day after attending surgeons and obstetricians/gynecologists worked at some point during the night were not associated with significantly increased complication rates compared with control cases that were not preceded by nighttime work. However, the duration of sleep opportunity available to surgeons and obstetricians/gynecologists following overnight cases varied widely. Among surgical (operating room) but not obstetrical/gynecologic (L/D) postnighttime cases with 6 hours or less of sleep opportunity, there was a substantially elevated rate of complication compared with cases in which sleep opportunity exceeded 6 hours. Furthermore, we found a nonsignificant increase in the risk of complications among combined operating room and L/D cases for work durations greater than 12 hours as compared with shifts of 12 hours or less. These data suggest that attending physicians, like residents and nurses,^1,19 may be at increased risk of making errors when sleep deprived or working extended shifts.

Our findings add to the limited literature on the effects of attending physician sleep deprivation and extended work shifts on clinical outcomes. In a study of surgical errors, fatigue was self-reported as a contributing factor in 16% of preventable adverse events.²⁰ Ellman et al⁴ conducted a retrospective study of adverse outcomes in cardiac surgical procedures performed by sleep-deprived surgeons performed within the 24-hour interval following an overnight procedure. They found no difference in complication rates in procedures conducted by sleep-deprived vs nonsleep-deprived surgeons. Similar to Ellman et al,⁴ we found no overall difference in complication rates among those performing surgery after working during the night, although both limited sleep opportunity and increased work duration predicted increased complication rates.

The 2008 Institute of Medicine report on resident duty hours did not comment on attending physician work hour limitations.²¹ It remains unclear whether aging attending physicians are more or less able to cope with the physiologic effects of extended work shifts and fatigue than are younger resident physicians.²² A recent survey found that attending physicians generally have greater concerns than do residents about the potentially harmful effects of work hour restrictions for trainees on their professional development.²³ Attending physicians may be less likely to acknowledge the potentially harmful effects of extended work shifts than trainees. Some attending physicians may also be less inclined than residents to postpone electively scheduled surgical procedures even when they are aware of the possibility of decreased alertness from insufficient overnight sleep.

Measures to reduce attending physician extended-duration work shifts have the potential to influence patient care. Although our results suggest that improving sleep opportunities may improve care, work hour restrictions could also lead to disruptions in care continuity or delays.²⁴ Continuity of care is not a major issue, however, when considering appropriate rest periods between the performances of procedures on different patients.

There have been no US legislative or professional initiatives concerning sleep deprivation or work hours for attending physicians.²⁵ While noncompliance with the Accreditation Council for Graduate Medical Education resident work hour standards has been problematic,²⁶ attending physician work schedule restrictions would likely be even more challenging to monitor, enforce, and achieve. European restrictions on attending and resident physician work schedules have been complicated by increased surgical staff requirements and staff shortages.²⁷

Our data suggest that the business as usual of a “limitless work week” for attending physicians is not optimal for patient care.²⁵ Several initiatives could mitigate the risks of unsafe levels of fatigue during procedures. Large physician groups can avoid scheduling elective procedures following overnight on-call responsibilities or use hospital-based clinicians, such as obstetrical “laborists” and surgical hospitalists, to cover nighttime emergencies.^28- 29 A culture of teamwork,³⁰ along with critical redesign of schedules, can mitigate the chance of unduly fatigued attending physicians performing procedures.^10,31 When possible, adequate backup personnel should be available to relieve physicians who detect impaired performance due to severe fatigue in themselves and others.³² Better sleep hygiene or the appropriate use of caffeine as a stimulant should be considered.¹⁰

Individuals should be educated about the effects of sleep deprivation on performance and how to recognize its effects. While cancellations following sleep-deprived nights are uncommon and may cause emotional and logistical challenges for patients and staff,³³ attending physicians should consider canceling or postponing elective procedures if they are not alert enough to safely operate. While technologies exist for monitoring alertness during certain tasks such as driving, they have not been tested for use as a “fitness for duty test” in health care or other settings.

Our study has several limitations including possible unidentified confounders or biases not accounted for in a retrospective cohort study. Our results might not be generalizable, especially to nonteaching hospitals; the presence of resident physicians may either have compensated for the effects of attending physician fatigue or have added to it if the residents themselves were sleep deprived. While most of the physicians in our study primarily operated at the study hospital, we were unable to determine if the rested attending physicians conducted overnight emergency procedures elsewhere. We were unable to determine if additional surgeons and obstetricians/gynecologists were requested to assist postnighttime procedures. We were also unable to identify if the physicians who performed control procedures were actually awake overnight. Judging the preventability of complications may have been affected by reviewers' hindsight bias.

We found that the rate of complications for surgeons and obstetricians/gynecologists who performed procedures the preceding night in our study hospital was not higher overall than for those who did not. Surgeons who had limited opportunity for sleep had significantly higher rates of complications than those who had a longer sleep opportunity. We did not observe an increased rate of complication among obstetricians/gynecologists who had limited opportunity for sleep; but given the lower rate of complications in labor and delivery as compared with surgery, larger studies with increased statistical power will be needed to further explore the effects of sleep deprivation in this setting.

Prospective studies are needed to more definitively quantify the effects of sleep deprivation on attending physicians across specialties, and to determine the safety of performing surgery after working at night in nonteaching settings. For situations in which it is necessary for attending physicians to conduct life-saving procedures following overnight work, effective strategies to minimize the effects of fatigue should be adopted into practice.