| JOURNAL HOME | CME HOME | THIS MONTH | PAST ISSUES | ETOC | COLLECTIONS |
| AUTHORS | REVIEWERS | EDITORIAL BOARD | FEEDBACK | RSS | HELP |
| ||||||||||||||
| ||||||||||||||
|
|
|
|
||||||||||||
|
||||||||||||||
From the Departments of *Anesthesia and Health Management and Policy, University of Iowa, Iowa City, Iowa; and Departments of Surgery, and Anesthesiology, SUNY Upstate Medical University, Syracuse, New York.
Address correspondence and reprint requests to Franklin Dexter, MD, PhD, Division of Management Consulting, Department of Anesthesia, University of Iowa, Iowa City, IA 52242. Address e-mail to Franklin-Dexter{at}UIowa.edu or www.FranklinDexter.net.
Abstract
BACKGROUND: Previous studies of operating room (OR) information systems data over the past two decades have shown how to predict case durations using the combination of scheduled procedure(s), individual surgeon and assistant(s), and type of anesthetic(s). We hypothesized that the accuracy of case duration prediction could be improved by the use of other electronic medical record data (e.g., patient weight or surgeon notes using standardized vocabularies).
METHODS: General thoracic surgery was used as a model specialty because much of its workload is elective (scheduled) and many of its cases are long. PubMed was searched for thoracic surgery papers reporting operative time, surgical time, etc. The systematic literature review identified 48 papers reporting statistically significant differences in perioperative times.
RESULTS: There were multiple reports of differences in OR times based on the procedure(s), perioperative team including primary surgeon, and type of anesthetic, in that sequence of importance. All such detail may not be known when the case is originally scheduled and thus may require an updated duration the day before surgery. Although the use of these categorical data from OR systems can result in few historical data for estimating each case’s duration, bias and imprecision of case duration estimates are unlikely to be affected. There was a report of a difference in case duration based on additional information. However, the incidence of the procedure for the diagnosis was so uncommon as to be unlikely to affect OR management.
CONCLUSIONS: Matching findings of prior studies using OR information system data, multiple case series show that it is important to rely on the precise procedure(s), surgical team, and type of anesthetic when estimating case durations. OR information systems need to incorporate the statistical methods designed for small numbers of prior surgical cases. Future research should focus on the most effective methods to update the prediction of each case’s duration as these data become available. The case series did not reveal additional data which could be cost-effectively integrated with OR information systems data to improve the accuracy of predicted durations for general thoracic surgery cases.
Prediction of operating room (OR) case durations using historical data is most accurate when the average is taken of the durations of historical cases with the same combination of scheduled procedure(s), individual surgeon and assistant(s) who will perform the procedure(s), and type of anesthetic(s).1–3 Still, there is residual inaccuracy in case duration estimates. There are two components to inaccuracy of the durations of multiple cases with the same predicted duration: bias and imprecision. An example of bias is all cases taking 8 min less than their predicted durations.4 An example of imprecision is half of cases taking 30 min less than their predicted durations and the other half taking 30 min longer than their predicted durations. By far, the largest source of inaccuracy in predicted case durations is imprecision, not bias.5
The imprecision of the actual durations of cases with the same predicted duration can result in facilities having long patient and surgeon waiting times on the day of surgery.3,5 To mitigate late starts, managers appropriately compensate by budgeting extra OR capacity (i) to permit scheduled buffers of time between surgeons’ successive lists of cases and (ii) to facilitate the movement of cases from one OR to another.1,3 However, the mitigation is costly and imperfect. Ideally, the imprecision itself would be reduced.
Reliance on the triad of procedure(s), surgeon, and anesthetic when estimating case duration is based heavily on studies of OR information systems data.1–3 In this paper, we try to discover additional predictors of case duration by systematically reviewing observational studies that included data from sources such as medical records (e.g., surgeons’ notes written with standardized vocabularies), preanesthesia evaluation forms (e.g., patient weight), and radiology picture archiving and communication systems (e.g., tumor location). Integration of such clinical data with OR information systems might improve the accuracy of case duration prediction for a sufficient number of different procedures as to be cost-effective.
We used general thoracic surgery as a model specialty to discover what additional data could improve the accuracy of predictions of case durations for many cases. Because general thoracic cases are typically long and elective (scheduled), the imprecision in predicted durations influences how many hours of cases are scheduled in each OR each day. Thus, we expected the general thoracic surgical literature to be stocked with case series that used operative duration and other components of case durations as reported end-points.
METHODS
PubMed was searched on July 6, 2007, for general thoracic surgery articles that included data on one or more components of case duration. The following search protocol was used:
The search yielded 347 abstracts, which were screened according to the following criteria:
No restrictions were placed on the language, year, or quality of the article. Studies were excluded that presented before/after groups as assessing trends over time (e.g., last decade to this decade). Studies were included that presented before/after groups as assessing historical control versus new intervention (e.g., older data for an older procedure to newer data for a new procedure).
The search yielded 71 abstracts, whose full papers were read and further screened according to the following criterion:
For articles that included two means and standard deviations, and stated that the difference was statistically significant, but did not report a P value, we performed Student’s t-test ourselves. However, articles were excluded in which group comparisons included differences in the numbers of procedures performed (e.g., mediastinoscopy followed by lung resection as two sequential surgical cases versus one surgical case). We already know that performing multiple procedures will take at least as long as the longest of the individual procedures, and numbers of procedures do not represent additional data obtainable from electronic medical records.6
RESULTS
There are differences in case duration based on the type of procedure(s), surgeon and assistant(s) performing the procedure, and the type of anesthetic used. For type of procedure(s), multiple studies report differences in components of case duration based on different anatomic procedures used for the same medical condition (Table 1)7–15 and based on different methods or approaches used to achieve the same anatomic result (Table 2)16–38. For surgeon, day-to-day differences in case duration result from varying composition of the surgical team (Table 3),39–42 but not from surgical technique (Table 4)43–48 since each surgeon tends to select one technique over another. For type of anesthetic, differences are also reported (Table 5).49–53
|
|
|
|
|
Tables 1–5 are ordered in declining sequence of the median differences in case duration. The sequence matches the relative impact expected from prior analyses of OR information system data.2 Different anatomic results have the largest impact (Table 1), surgical approach the second largest (Table 2), and so forth.
The information shown in Tables 1–5 are not all used when cases are originally booked at some facilities (e.g., those of the authors). For example, Table 2 includes the surgical approach, which may be decided only once all imaging is available. Table 3 includes the complete surgical team, most of who may not be known until the working day before surgery. Table 5 includes the specific type of anesthetic, which may not be known until after completion of the preanesthesia evaluation and assignment of the anesthesiologist. Thus, routine usage of the information listed in Tables 1–5 would generally involve changes to the prediction of case duration as data are updated.
The use of all of the data from Tables 1 to 5 to select historical cases from the OR information system to use to predict a future case’s duration can result in there being few historical case durations for each future case. Depending on each facility’s relationship between how a case is scheduled and the selection of the surgeon preference card(s), the consequence can be a reduction in how often each preference card is used. Having many preference cards can contribute to inaccuracies and infrequent updates, even for components such as medications for which inaccuracy can harm patients.54 Table 6 shows that small sample sizes themselves are unlikely to contribute paradoxically to increased bias and imprecision in predictions of case durations (e.g., from intraoperative delays).
|
By analyzing OR information system data, it is not possible to judge whether a predictor of case duration (e.g., anesthetic) is causing the difference or is just a marker for an underlying patient characteristic, disease characteristic, etc., that is an actual cause of the difference in case duration and would itself be a more accurate predictor. The preceding 45 studies had many such characteristics that were known preoperatively and that were compared statistically between the groups for which there were large differences in case duration (Tables 1–5). Table 7 shows that almost all of the characteristics did not differ significantly between groups. Thus, the previously identified predictors of case duration (Tables 1–5) were, in fact, directly causing the differences, not serving simply as markers for individual data in Table 7. If one or more of the characteristics in Table 7 were used for case duration prediction, their use would need to be in addition to the data in Tables 1–5, not in lieu of one or more of those data.
|
|
|
DISCUSSION
Four Implications of Our Findings
First, although we used an entirely different approach to evaluating differences in case duration than the statistical analysis of OR information system data for all cases at a facility,1–6 the results were strikingly similar.
Second, the data stored in OR information systems are the principal ones needed for the prediction of durations for most cases (Tables 1–5). The cost of adding data from other information systems such as electronic medical records would be fixed regardless of the few cases for which benefits may accrue. Thus, to improve the accuracy of estimates, initial efforts should focus on the most effective methods and computer–human interfaces to determine and use the data already1–6 known to be important (Tables 1–5), not searching for more data. We speculate that many facilities are using the information in Tables 1, 2, and 4, but need to add use of their knowledge of staff assignment (Table 3) and type of anesthetic (Table 5).
Third, because case durations have marked uncertainty, there is no singular "the duration." Because the decision whether to perform a case in a specific OR at a specific facility on a specific day tends to be affected little by inaccuracy in predicted case duration,2–5 obtaining all of the information at the time the case is scheduled is unnecessary. In contrast, close to the day of surgery, decisions (e.g., case sequencing)2,3 depend on the longest and shortest times that cases may take and are sensitive to inaccuracy in case duration prediction. We recommend that for decisions made the working day before surgery, the original prediction of case duration not be used, but a more accurate prediction including the additional data then available, such as the expected perioperative team (Table 3) and anesthetic (Table 5).
Fourth, OR information systems need to incorporate the statistical methods designed for small numbers of prior surgical cases. Tables 1–5 show the importance of obtaining information on the procedure(s) to be performed (e.g., "thoracoscopic wedge resection of lung" versus "video assisted thoracoscopic surgery" and/or "wedge resection"), surgical team (e.g., "Dr. Smith and 1st year cardiothoracic resident who has not previously performed the procedure" versus "Dr. Smith"), and type of anesthetic (e.g., "general anesthesia" versus "anesthesia"). This use of so much detailed information results in small sample sizes. More than 75% of procedure(s) scheduled may be scheduled fewer than three times per year.59,60 Additionally, more than 25% of cases may involve procedure(s) likely scheduled by the case’s surgeon less than twice per year.60,61 Pooling data among facilities generally does not suffice to compensate for the small sample sizes when estimating case durations, because procedure(s) that are rare at one facility tend to be rare elsewhere.59,62 Table 6 provides new information showing that if all of the information of Tables 1–5 were used for case duration prediction, there is a low chance of an indirect reduction in the accuracy of case duration estimates caused by the reduction in sample sizes. Previous studies3 showed the benefit of predicting the longest and shortest times that a case may take using the statistical methods designed for small numbers of prior surgical cases. The surgeon’s estimate and the information of Tables 1–5 are used to estimate the center (median) of the statistical distribution of case durations, just as done currently by most OR information systems. Then, the proportional uncertainty is estimated using data from many cases,3 not just those classified as in Tables 1–5. These so-called Bayesian statistical methods are both nonproprietary and accurate,3 making their inclusion into commercial systems appropriate.
Limitations
Wright et al. showed that once historical case duration data have been used to provide a prediction of a new case’s duration, permitting the surgeon to adjust the estimate up or down by a reasonable percentage (e.g., 10%) can decrease the imprecision.63 At the time of case scheduling, often there is limited knowledge of who will assist the surgeon.63,64 On the basis of our findings, we speculate that the information in Tables 1–5 were the data about each case being used by the surgeon63 to adjust his estimate. However, from our study we cannot discern more.
Our limited focus on case duration prediction resulted in our consideration of the detail with which procedures are scheduled (Table 1) but not the process. Other ways that anesthesia providers use the scheduled procedure(s) depend on those procedures being specified using a standardized vocabulary (e.g., Current Procedural Terminology), for six reasons.1–6,63,65–68 (i) When procedures are classified using physician billing codes, the physiological complexity of the case can be quantified automatically from the corresponding anesthesia basic (startup) units. This information permits automatic and prompt evaluation of the appropriateness of a case for a facility.65–67 (ii) When a patient is scheduled for physiologically complex surgery, the schedulers can be prompted to arrange a preanesthesia clinic appointment. (iii) Appropriate patient instructions can be selected automatically. (iv) Preapproval of insurance can be automated. (v) Automatic assignment of the case to the anesthesia providers can be based on the computer knowing the physiological complexity and rareness of each procedure. (vi) Expected postoperative bed requirements can be checked automatically.68
Our study was limited to procedures performed by general thoracic surgeons. However, our results matched those of a previous study of total hip replacement and knee replacement.69
Finally, our objective was not to average differences in operative times as part of a meta-analysis, but rather to organize and review the pertinent scientific literature. We limit our conclusions to saying that many general thoracic surgeons were sufficiently interested in the time that they take to complete their procedures that they included such end points in their clinical trials and case series. They found many factors to be important, almost all of which (Tables 1–5) matched those found to be relevant using the entirely different approach of statistically analyzing OR information system records for all specialties simultaneously.1–6
Footnotes
Accepted for publication December 5, 2007.
Dr. Franklin Dexter, Section Editor for Economics, Education, and Policy, was recused from all editorial decisions related to this manuscript.
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  R. E. Wachtel and F. Dexter Influence of the Operating Room Schedule on Tardiness from Scheduled Start Times Anesth. Analg., June 1, 2009; 108(6): 1889 - 1901. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. E. Wachtel and F. Dexter Reducing Tardiness from Scheduled Start Times by Making Adjustments to the Operating Room Schedule Anesth. Analg., June 1, 2009; 108(6): 1902 - 1909. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. Dexter, R. H. Epstein, J. D. Lee, and J. Ledolter Automatic Updating of Times Remaining in Surgical Cases Using Bayesian Analysis of Historical Case Duration Data and "Instant Messaging" Updates from Anesthesia Providers Anesth. Analg., March 1, 2009; 108(3): 929 - 940. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
