Secondary Logo

Journal Logo

Economics, Education, and Policy: Research Report

Lack of Utility of a Decision Support System to Mitigate Delays in Admission from the Operating Room to the Postanesthesia Care Unit

Ehrenfeld, Jesse M. MD, MPH*†; Dexter, Franklin MD, PhD; Rothman, Brian S. MD*; Minton, Betty Sue BN, MSN§; Johnson, Diane RN, MSN§; Sandberg, Warren S. MD, PhD; Epstein, Richard H. MD, CPHIMS

Author Information
doi: 10.1213/ANE.0b013e3182a8b0bd
  • Free

Abstract

When the phase I postanesthesia care unit (PACU) is at capacity, completed cases need to be held in the operating room (OR) until either a PACU bed becomes available or the patient has recovered sufficiently to be discharged to an alternative location (e.g., a phase II discharge unit). Such “PACU delays” can arise if all physical PACU slots are occupied or there are insufficient nurses to provide safe care, based on the patients’ acuity.

Our group previously published statistical methods that optimize regular workday PACU staffing to achieve the smallest labor cost at a given service level (e.g., 95% of days with no PACU admission delays).1–3 The methodology can also be applied to achieve the highest service level possible without increasing the existing number of staffed hours (i.e., cost).4 For either objective, the approach used is adjustments to the distribution of shifts among start time options and durations that the nurses work.1-4 The computations are performed using historical data for PACU admissions and discharges.1–4 Planning staffing and staff schedules requires computer software, not the back of an envelope, because consideration of hundreds of billions of possible scheduling solutions typically is required.3, 4 A systematic review of other potential strategies to reduce the incidence of PACU delays such as adjusting nurse assignments the day before surgery, changing the sequence of surgical cases, or reducing delays in discharge from the PACU demonstrated that such efforts can be impractical, requiring large changes to organizational processes and lack of availability of sufficient personnel and equipment.4 For example, although, intuitively, case sequencing may seem useful, beneficial reductions in the peak census of the PACU are minimal when ORs have heterogeneous durations of the workday and relatively independent case scheduling, as found epidemiologically to be commonplace.4–6

Another possible approach for reducing PACU delays would be to create a decision support system (DSS) to assist active management of the PACU census on the day of surgery using detection of patient arrival in the PACU from perioperative management systems and automated paging. These processes are similar to those described by Meyer et al.7 in their “INCOMING!” system for expediting the discharge of patients from the PACU to the ward. Such systems could be integrated into electronic PACU bed occupancy displays8 or an alert could be sent to the PACU charge nurse when the PACU is nearly full or at capacity in order to avoid a subsequent PACU delay. Such systems might be especially useful in large PACUs and hospitals with multiple PACUs, where continual manual recalculation of the census and patient flux would be difficult. Such a DSS might also be useful for improving the service level where financial constraints limit staffing to suboptimal levels, or for the infrequent periods when demand is unusually high. However, such a system would provide utility only if the alerts were timely and if resources were available promptly to expedite the exit of PACU patients at the time of the alert. By timely, we mean that the alert needs to be provided sufficiently ahead of the anticipated PACU delay to allow an expedited discharge, but not so far in advance that managers would reasonably hesitate to initiate any special action. Evaluation of whether this can possibly occur was the first of the 2 goals of our article.

Another potential benefit of this DSS would be to help manage contention when 2 ORs are finishing at nearly the same time and are vying for the last available PACU slot. Basing the decision on reducing over-utilized time is an economically rational criterion, because on the day of surgery over-utilized time is expensive compared to the negligible cost of under-utilized time.9 In this scenario, the patient who should be accepted for admission would be the one from the OR that has the larger expected hours of over-utilized time, not necessarily the patient from the room that called first.8 Evaluation of the maximum potential utility of this approach was the second of the 2 goals of our article.

We previously have attempted to implement DSS for these approaches at several facilities, all without success. Our impressions are that the sources of the failures were multifactorial, including: lack of timely documentation of PACU arrivals and discharges, inaccurate representations of PACU assignments, reliance on paper-based documentation, behavioral changes resulting from knowledge by the PACU nurses that their activity was being recorded, clerical as opposed to provider data entry (resulting in inaccurate or delayed data capture), and pre-event or post-event documentation (e.g., due to lack of availability of the electronic charting system during transport). Therefore, we elected to study the maximum potential benefit of such a DSS using simulation, as opposed to direct clinical observation. We reasoned that if utility could not be demonstrated under idealized conditions, actual implementation would fail due to lack of cost-utility.

In order to perform our simulations, we capitalized on several unique characteristics of the PACU environment at Vanderbilt University Medical Center. The hospital has an electronic PACU documentation system that allows reconstruction of the nursing and patient census at every minute of the day. In addition, admissions to the PACU are not constrained either by the number of physical recovery slots (Table 1)10 or by inadequate PACU staffing (Fig. 1). These very unusual attributes provided us with real data from a model system for analysis of hypothetical PACU census management DSS alerts under idealized conditions. The lack of capacity constraint allowed us to evaluate utility in our simulations by reducing staffing (performed by reducing the maximum census) and adjusting the census at which an alert would be triggered. Not only did our use of such an unusual PACU bias our results to maximize the potential utility of DSS alerts, we deliberately biased our analysis assumptions to maximize the potential utility of alerts.

Table 1
Table 1:
Projected Average Total Minutes per Regular Workday at Vanderbilt with OR Cases Waiting for PACU Admission Based on Average Daily Census and Physical PACU Bedsa

Since we previously failed at implementation of a PACU census alert system, the hypothesis of our study was that we would not demonstrate sufficient utility for preventing PACU delays to warrant development of a DSS sending alerts when the PACU was at or near capacity.

METHODS

Study Data and IRB Approval

The Vanderbilt University IRB approved the study without requirement for written patient consent. Deidentified PACU data from regular, non-holiday workdays were extracted from January 1, 2010 to December 31, 2012 from the Vanderbilt perioperative information management system databases. Analyzed fields included the times of patient entry into and discharge from the PACU, and the start and end times that individual PACU nurse(s) were involved in the care of each patient. There were 42,199 admissions evaluated (55.7 ± 0.3 per day) during the 782 workdays in the dataset (mean ± SE). The Vanderbilt PACU studied comprises 2 physically separate units in close proximity, each of which has a charge nurse, and to which patients from either of the 2 main OR suites can be sent to recover following anesthesia. In addition, there is an administrative charge nurse who, among other responsibilities, is responsible for the overall flow into and out of the PACUs.

As shown by Table 1 and Figure 1, admissions to the studied PACU were not limited by inadequate physical PACU bed capacity or by insufficient nurse staffing. Again, these very unusual conditions were necessary to study via simulation the maximum possible benefit of an alert system.

Figure 1
Figure 1:
Spaghetti plot of actual staffing versus required staffing in the postanesthesia care unit (PACU) on the 10 busiest PACU days. These days had the highest peak patient census. Actual staffing is displayed as the solid red line for 7:00 AM to 11:00 PMon regular workdays, In panel A, the black lines represent the number of PACU nurses required, assuming a patient to nurse ratio of 2:1 and rounding up to the next integer, on the 10 busiest days of the study interval. In panel B, a patient to nurse ratio of 1.8:1 (i.e., requiring more PACU staff) is assumed as a sensitivity analysis.2 The maximum PACU census observed at any time was 29, with a maximum of 35 PACU beds available for patient care. The gap between the actual and required staffing demonstrates that the PACU was not capacity bound due to a lack of available nurses, and thus suitable for modeling a census management decision support system. The large gap between the PACU staffing and the PACU census is because the PACU nurses function at the start of the day to prepare patients for first case starts in the 35 operating rooms served by the PACU.

Description of the Simulated Census Management Alert System

The DSS was based on the following assumptions: (1) at all times there was an available patient who could be moved from the PACU; (2) at all times an appropriate hospital bed was available for such patient(s); (3) at all times, a PACU bed could be made available 30 minutes after the alert by moving such patients and cleaning the slot, and/or procuring more nurses; and (4) the patient to nurse ratio was not less than 2 to 1. If any criterion were not satisfied, the alert transmitted by the DSS would not result in a benefit. Therefore, these assumptions resulted in deliberate overstatement of the best possible performance of the DSS. In the model, the historical arrival PACU arrival and discharge times for all patients were maintained.

Terms used in the model are defined in Table 2. If following an alert the PACU census never exceeded the maximum (i.e., the patient flux out of the unit balanced subsequent admissions), the Alert Lead Time was considered to be infinite. A schematic of the model alert system is presented in Figure 2.

Table 2
Table 2:
Definition of Terms Used in the Simulation Model
Figure 2
Figure 2:
Alert system schematic. The function of the hypothetical postanesthesia care unit (PACU) alert system is displayed for a hypothetical 4-bed PACU where the alert was set to trigger when the maximum census of 4 patients is first reached. In this scenario, an alert lead time of 30 minutes is needed to clear a patient to make room for the next patient and thus avoid a PACU delay. At 11:00 AM, the PACU has 3 patients (A, B, and C). The arrival of the next patient (D) triggers the alert. Patient C is identified as being suitable for an expedited discharge and is sent to an available hospital bed at 11:35 AM, reducing the census back to 3 patients. At 12:00 PM, patient E arrives in the PACU and can be accommodated, due to the expedited discharge of patient C. The PACU charge nurse again begins the process of expediting a discharge to make room for the next patient. However, at 12:25 PM, patient F is ready to leave the operating room, but there has not been sufficient time to expedite another discharge, since the arrival occurred at a time interval less than the alert lead time. This results in a PACU delay.

The PACU census tends to rise progressively from near 0 patients at the start of a day to a peak as the first cases of the day end (i.e., there is a positive slope in the census, since more patients are entering the PACU than are leaving the PACU).11 To avoid correlations among alerts on the day of surgery (e.g., due to an exceptionally busy OR schedule), we studied only the first alert of the day. (For details on the large effect of such auto-correlation, readers are referred to Refs. 1, 3, and 6.) This approach further increased the chance of our finding the alert system useful.

Assessment of Utility of Census DSS to Prevent a PACU Delay

The PACU census was determined at each of the 1440 1-minute intervals on every day studied. When the specified PACU census was reached, a hypothetical alert was triggered that was assumed to result in the transfer of a PACU patient at the alert lead time. Again, the fact that in practice transfer of most patients out of the PACU would take longer than the minimum notification time implies that our study deliberately overestimates the potential benefit of alerts.

The census maximum of 24 patients (i.e., corresponding to 12 nurses) was the 95th percentile of PACU occupancy during the 782 workdays studied (Fig. 1). As one sensitivity analysis, all calculations were repeated using an occupancy corresponding to a reduction in PACU staffing by 2 nurses (i.e., to 20 patients). As another sensitivity analysis, calculations were repeated using an alert trigger set point from 1 to 4 patients less than the maximum PACU census. The implication of an alert trigger less than the maximum census is that the saturating number of admissions (i.e., the number of admissions required to produce a PACU delay) would be equal to 1 more than the offset. For example, if the trigger were set at 1 less than the maximum census, it would take a net increase in the PACU census of 2 patients during the alert lead time to cause a PACU delay.

Utility of an alert to prevent a PACU delay was assessed according to whether the PACU census subsequently reached a value exceeding the simulated maximum (Fig. 3). If the alert lead time were less than a minimum notification time of 30 minutes, we considered that the alert would have been ineffective (i.e., too late) in preventing the PACU delay. We based this minimum notification time on the expert opinion of our 2 PACU managers (BSM and DJ), who indicated that identifying a patient suitable for expedited discharge, calling for a bed, arranging transport, calling report, moving the patient out of the PACU, and clearing a slot would take at least 30 minutes under ideal circumstances (i.e., usually >30 minutes). If the alert lead time was more than 60 minutes after the alert, we considered the alert to have been ineffective (i.e., too early). A PACU manager is unlikely to credit the DSS alert as preventing such a delay, given the large flux of patients in the PACU (Table 1, Fig. 1) and the fact that patients identified for expedited discharge would be likely to have been discharged soon regardless of being expedited, since they would already have met discharge criteria. Our PACU experts (BSM and DJ) reported that when trying to anticipate problems, they typically do not look attempt to predict events more than an hour into the future. If the alert trigger set point was reached, but the census never exceeded the PACU, we also considered this alert as ineffective (i.e., a false alarm). Thus, we considered an alert potentially effective in avoiding a PACU delay if the alert lead time was between 30 and 60 minutes, inclusive.

Figure 3
Figure 3:
Flow diagram of assessment of alert triggered at a specified postanesthesia care unit (PACU) census (threshold) on a given day. The first alert on each day (if present) was evaluated according to the steps in this flow diagram, and was assessed as being potentially useful, or not useful, according to whether the interval from the alert to when the PACU first exceeded the specified maximum capacity. As sensitivity analyses, various intervals (15 to 90 minutes) and census thresholds (20 and 24) were evaluated (see Fig. 4). Alerts arriving outside the interval were not considered as useful as they would have arrived too late to clear the PACU slot, or too early to be causally linked to preventing the subsequent delay.

We conducted sensitivity analyses by extending the range of utility to wider ranges of the minimum to maximum effective notification time, specifically: 15 to 75 minutes, 30 to 90 minutes, and 15 to 90 minutes. Each of these 3 expanded ranges exceeded the limits identified by our PACU experts, and again represents bias in the direction of showing utility of the DSS alert system.

Assessment of Utility of DSS Alerts to Reduce Over-utilized OR Time

This assessment was made with an alert triggered at 1 less than the PACU census. Under such conditions, if 2 ORs call for a PACU slot, the room with the largest estimated amount of over-utilized time should get the slot, as this will reduce the expected minutes of over-utilized time.9 Given that OR’s generally call for a PACU slot following the end of surgery, there is an interval during which an assignment could be rescinded if a call was received from another OR with a higher priority for the institution. We considered the maximum interval during which such a decision could be made to be 15 minutes, corresponding to our previous meta-analyses of prolonged extubation.12,13 That is, if 2 patients arrived in the PACU within 15 minutes following an alert, we considered this as a potential opportunity for changing the OR getting the last available PACU slot. We measured such utility, using alerts triggered at 1 patient less than the maximum census, for patients arriving within 5, 10, and 15 minutes of each other. However, even assuming that all ORs had over-utilized time, the maximum beneficial effect of choosing between 2 ORs selected at random would be 50%, since for half the cases, the correct room would have called first and no change would be made. Since the definition of over-utilized time implies that fewer than half of ORs will have no over-utilized time,14–16 this estimate of 50% is deliberately conservative and biased to show utility of the DSS alert system.

Distribution of Alert Lead Times

The cumulative distribution of alert lead times was calculated for maximum PACU census values between 20 and 24 with the alert trigger set point set at 1 less than the maximum PACU census.

Statistical Methods

The percent utility of each studied alert trigger set point was calculated as the number of days with an alert showing utility divided by the number days with at least one alert. One-sided 95% conservative upper confidence limits were calculated using the method of Clopper-Pearson and are reported using the phrase: “at most.”17 One-sided upper limits were used as we made nonstatistical assumptions biased to show that the alerts would have utility; thus, the confidence intervals should be limited in a similar fashion.

RESULTS

At most, 23% of DSS alerts sent when the PACU census was at the maximum or at 1 less than the maximum, respectively, would have potentially allowed the PACU charge nurse to prevent a subsequent PACU delay (Fig. 4A). The remainder of the alerts would have been false alarms (i.e., the PACU subsequently never exceeded the maximum), either too late (<30 minutes after the alert), or too early (>60 minutes after the alert) to be useful (see Table 3, corresponding to Fig. 4A). When the range of the minimum to maximum effective notification time was extended to 15 to 75 minutes, 30 to 90 minutes, and 15 to 90 minutes, the maximum utility among all simulations would have been 55% (Fig. 4, B–D).

Table 3
Table 3:
Details of Alerts Sent at Various Alert Trigger Set Points as Shown in Figure 4A
Figure 4
Figure 4:
Potential utility of a census management system to avoid potential postanesthesia care unit (PACU) delays. The red circles represent the average utility of an alert system triggered at a maximum census of 20 to 24 patients to prevent a subsequent PACU delay, defined as receipt of an alert within the indicated interval of when the PACU census exceeded the maximum (i.e., alert lead time). The error bars represent one-sided 95% upper confidence limits, reflecting maximum utility of the system. The blue circles correspond to alerts sent when the PACU was at maximum census − 1, green circles at maximum census − 2, cyan circles at maximum census − 3, and orange circles at maximum census − 4. The minimum utility for a decision support system (DSS) providing alerts as automated aids to busy managers, as found in a meta-analysis by Wickens and Dixon,18 is shown on each graph as the dotted red line at 70%.15 Minimum to maximum effective notification times from 30 to 60 minutes (A), 30 to 90 minutes (B), 15 to 75 minutes (C), and 15 to 90 minutes (D) were evaluated. The figures show that the maximum utility was considerably <70% for all alert trigger set points. For the simulation with the widest minimum to maximum effective notification times (D), maximum utility at each census level was an alert trigger set point of maximum census − 1. The display of upper confidence limits only, and multiple nonstatistical assumptions described in the Methods all deliberately biased results to overestimate maximum potential benefit. Thus, the figures demonstrate lack of utility of a census management DSS for preventing PACU delays.

If the alert trigger set point were set at 1 less than the maximum census of 24 patients, at most 45% of alerts would potentially have resulted in reassigning a sole available PACU bed, depending on the range of the minimum to maximum effective notification time to make the change (5 to 15 minutes; Fig. 5). If the alert trigger set point was 1 less than a smaller maximum census of 20 patients (see Methods), potential utility would have occurred for at most 37% of the alerts, similarly depending on the range of effective notification times.

Figure 5
Figure 5:
Potential utility of a postanesthesia care unit (PACU) census management system to reduce hours of over-utilized operating room (OR) time. The green circles represent the maximum utility of the alert system triggered when the census was 1 less than a maximum of 20 patients to reduce over-utilized OR time (i.e., hours of operation beyond those allocated) by rescinding a promised PACU slot between 5 and 15 minutes after the slot was originally assigned. The error bars represents 1-sided 95% upper confidence limits, reflecting maximum utility of the system. The orange circles correspond to alerts sent at an alert trigger set point of 1 less than a maximum census of 24 patients. The figure demonstrates low utility of a census management decision support system (DSS) for managing contention to reduce the amount of over-utilized OR time when the PACU is nearly full and 2 ORs are calling for the last available PACU bed.

The cumulative distributions of the alert lead times are shown in Figure 6 for maximum PACU census between 20 and 24 and an alert trigger set point at 1 less than the maximum census. At an alert lead time of 30 minutes, between 25% and 45% of alerts would have arrived too late for a patient to be discharged from the PACU to avoid a PACU delay. Between 30% and 65% would have arrived 90 or more minutes prior to the admission causing a PACU delay, too early likely to result in an effort to expedite a PACU discharge.

Figure 6
Figure 6:
Distribution of alert lead times. Times from the postanesthesia care unit (PACU) admission triggering an alert to the admission that would have resulted in the PACU exceeding the maximum census (i.e., alert lead time) were calculated for a maximum PACU census between 20 and 24 patients. The alert trigger set point was the maximum census − 1, the optimal value from Figure 4D. Alerts not followed by a census exceeding the maximum were considered to have an infinite alert lead time. For example, the dark blue line, corresponding to a maximum census of 20 patients indicates: (1) 42% of admissions resulting in a PACU delay occurred within 30 minutes of the alert (i.e., alert arrived too late); (2) 35% of such admissions occurred later than 75 minutes after the alert (i.e., alert arrived too early); and (3) approximately 30% of alerts did not result in the PACU ever exceeding maximum census. The utility of the alert for a given range of alert lead times can be calculated by subtracting the cumulative percentage at the lower limit from the upper limit. For the example and a range of 30 to 75 minutes, approximately 23% (42% to 65%) of the alerts would have had utility.

DISCUSSION

Performance was insufficient for the simulated PACU census management DSS. At least three-fourths of the alerts, sent when the census reached 1 patient less than the maximum census, would have failed to prevent a subsequent PACU delay. A DSS would also have low utility for reducing over-utilized time on the day of surgery, despite allowing up to a 15-minute interval to reassign a promised PACU slot to another patient. Actual performance would be expected to be even worse, since many of the assumptions on which the model was predicated would not be satisfied.

The suggested minimum threshold calculated for a DSS supplying automated alerts as diagnostic aids used by busy managers was found by Wickens and Dixon18 in their meta-analysis to be 70%. They found that below 70% utility, the combined performance of the human and the automation system was worse than had no automation been used at all.18 The model alert system evaluated in this study meets the inclusion criteria for automation in the meta-analysis: “filtering or focusing attention on information deemed to be of interest” (i.e., the current PACU census) and “forming inferences of the state of the world, by integrating information” (i.e., whether or not a PACU delay is imminent).18 The reason for this 70% threshold is that when the false alert rate from a DSS is high, people tend (reasonably) to ignore recommendations. Wickens and Dixon’s findings18 are directly applicable because it is precisely when the PACU is the busiest that it is the most challenging to have extra nursing supervisors and personnel not focus on direct patient care but instead to coordinate PACU and wards and/or PACU and ORs. The implication for our findings (Figs. 4 and 5, Table 3) is there seems little reason to pursue further this type of approach to reducing the incidence of PACU delays. The problems need to be prevented months ahead when staff scheduling is performed.1–4,6,8

Our study has several limitations. First, as stated repeatedly above, the studied PACU and our simulation assumptions greatly overstate the potential benefit of a PACU census management DSS. This approach was sufficient to reach the conclusion of lack of utility of DSS alerts based on PACU census on the day of surgery. However, the specific reported estimates of utility are biased in the positive direction. Second, because our model was evaluated using retrospective data, it ignores latency considerations studied within anesthesia information management systems (e.g., when nurses chart activity in the PACU relative to its actual occurrence19 and the times for transmission of the recommendations to the PACU managers).20,21 Such issues would further reduce the effectiveness of the DSS, as the intervals over which alerts might be useful are reduced. Third, we did not consider the more complicated scenario of >2 patients presenting simultaneously for admission to the PACU when the unit was near full. However, such instances are rare.

In conclusion, we were unable to show potential benefit of a DSS to mitigate PACU delays on the day on the surgery, despite multiple biases that favored effectiveness. Based on our results, the only evidence-based method of reducing PACU delays is to adjust PACU staffing and staff scheduling to the match the historical workload using computational algorithms.1–4,6,8 Such methods are applied months before the day of surgery.1–4

RECUSE NOTE

Franklin Dexter is Statistical Editor and the Section Editor of Economics, Education, and Policy for the Journal. The manuscript was handled by Dr. Steven L. Shafer, Editor-in-Chief, and Dr. Dexter was not involved in any way with the editorial process or decision.

DISCLOSURES

Name: Jesse M. Ehrenfeld, MD, MPH.

Contribution: This author helped design the study and write the manuscript.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: The author declares no conflicts of interest.

Name: Franklin Dexter, MD, PhD.

Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: The University of Iowa performs statistical analyses for hospitals using their operating room information systems, including some in this article. Income from the Division’s consulting work is used to fund Division research. Franklin Dexter has tenure and receives no funds personally, including honoraria, other than his salary and allowable expense reimbursements from the University of Iowa. He and his family have no financial holdings in any company related to his work, other than indirectly through mutual funds for retirement.

Name: Brian S. Rothman, MD.

Contribution: This author helped design the study and write the manuscript.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: The author declares no conflicts of interest.

Name: Betty Sue Minton, BN, MSN.

Contribution: This author helped analyze the data and write the manuscript.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: The author declares no conflicts of interest.

Name: Diane Johnson, RN, MSN.

Contribution: This author helped analyze the data and write the manuscript.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: The author declares no conflicts of interest.

Name: Warren S. Sandberg, MD, PhD.

Contribution: This author helped write the manuscript.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: The author declares no conflicts of interest.

Name: Richard H. Epstein, MD, CPHIMS.

Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript. This author is the archival author.

Attestation: This author has approved the final manuscript.

Conflicts of Interest: Richard H. Epstein is President of Medical Data Applications, Ltd., whose CalculatOR™ software includes the PACU staffing analyses mentioned in this article.

REFERENCES

1. Dexter F, Epstein RH, Penning DH. Statistical analysis of postanesthesia care unit staffing at a surgical suite with frequent delays in admission from the operating room–a case study. Anesth Analg. 2001;92:947–9
2. Dexter F, Wachtel RE, Epstein RH. Impact of average patient acuity on staffing of the phase I PACU. J Perianesth Nurs. 2006;21:303–10
3. Epstein RH, Dexter F, Traub RD. Statistical power analysis to estimate how many months of data are required to identify PACU staffing to minimize delays in admission from ORs. J Perianesth Nurs. 2002;17:84–8
4. Dexter F, Epstein RH, Marcon E, de Matta R. Strategies to reduce delays in admission into a postanesthesia care unit from operating rooms. J Perianesth Nurs. 2005;20:92–102
5. Marcon E, Dexter F. Impact of surgical sequencing on post anesthesia care unit staffing. Health Care Manag Sci. 2006;9:87–98
6. Marcon E, Dexter F. An observational study of surgeons’ sequencing of cases and its impact on postanesthesia care unit and holding area staffing requirements at hospitals. Anesth Analg. 2007;105:119–26
7. Meyer MA, Sokal SM, Sandberg W, Chang Y, Daily B, Berger DL. INCOMING!–A web tracking application for PACU and post-surgical patients. J Surg Res. 2006;132:153–8
8. Dexter F. Bed management displays to optimize patient flow from the OR to the PACU. J Perianesth Nurs. 2007;22:218–9
9. Dexter F, Willemsen-Dunlap A, Lee JD. Operating room managerial decision-making on the day of surgery with and without computer recommendations and status displays. Anesth Analg. 2007;105:419–29
10. Schoenmeyr T, Dunn PF, Gamarnik D, Levi R, Berger DL, Daily BJ, Levine WC, Sandberg WS. A model for understanding the impacts of demand and capacity on waiting time to enter a congested recovery room. Anesthesiology. 2009;110:1293–304
11. Dexter F, Tinker JH. Analysis of strategies to decrease postanesthesia care unit costs. Anesthesiology. 1995;82:94–101
12. Dexter F, Bayman EO, Epstein RH. Statistical modeling of average and variability of time to extubation for meta-analysis comparing desflurane to sevoflurane. Anesth Analg. 2010;110:570–80
13. Wachtel RE, Dexter F, Epstein RH, Ledolter J. Meta-analysis of desflurane and propofol average times and variability in times to extubation and following commands. Can J Anaesth. 2011;58:714–24
14. Strum DP, Vargas LG, May JH, Bashein G. Surgical suite utilization and capacity planning: a minimal cost analysis model. J Med Syst. 1997;21:309–22
15. McIntosh C, Dexter F, Epstein RH. The impact of service-specific staffing, case scheduling, turnovers, and first-case starts on anesthesia group and operating room productivity: a tutorial using data from an Australian hospital. Anesth Analg. 2006;103:1499–516
16. Wang J, Dexter F, Yang K. A behavioral study of daily mean turnover times and first case of the day start tardiness. Anesth Analg. 2013;116:1333–41
17. Dexter F, Marcon E, Aker J, Epstein RH. Numbers of simultaneous turnovers calculated from anesthesia or operating room information management system data. Anesth Analg. 2009;109:900–5
18. Wickens WD, Dixon S. The benefits of imperfect diagnostic automation: a synthesis of the literature. Theoretical Issues Erg Sci. 2007;8:201–12
19. Epstein RH, Dexter F, Ehrenfeld JM, Sandberg WS. Implications of event entry latency on anesthesia information management decision support systems. Anesth Analg. 2009;108:941–7
20. Epstein RH, Dexter F, Rothman B. Communication latencies of wireless devices suitable for time-critical messaging to anesthesia providers. Anesth Analg. 2013;116:911–8
21. Rothman BS, Dexter F, Epstein RH. Communication latencies of apple push notification messages relevant for delivery to time-critical information to anesthesia providers. Anesth Analg. 2013;117:398–404
© 2013 International Anesthesia Research Society