Introducing a methodology for estimating duration of surgery in health services research

  • Donald A. Redelmeier
    Correspondence
    Corresponding author. Sunnybrook Health Sciences Centre, G-151, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada. Tel.: +416-480-6999; fax: +416-480-6048.
    Affiliations
    Department of Medicine, Infectious Disease, and Health Policy Management & Evaluation, University of Toronto, Toronto, Ontario, Canada

    Clinical Epidemiology and Health Care Research Program, Sunnybrook Health Sciences Centre, G-151, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada

    Institute for Clinical Evaluative Sciences in Ontario; Patient Safety Service of Sunnybrook Health Sciences Centre, G-151, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada

    Division of General Internal Medicine, Sunnybrook Health Sciences Centre, G-151, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada
    Search for articles by this author
  • Deva Thiruchelvam
    Affiliations
    Institute for Clinical Evaluative Sciences in Ontario; Patient Safety Service of Sunnybrook Health Sciences Centre, G-151, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada
    Search for articles by this author
  • Nick Daneman
    Affiliations
    Department of Medicine, Infectious Disease, and Health Policy Management & Evaluation, University of Toronto, Toronto, Ontario, Canada

    Institute for Clinical Evaluative Sciences in Ontario; Patient Safety Service of Sunnybrook Health Sciences Centre, G-151, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada
    Search for articles by this author

      Abstract

      Objectives

      The duration of surgery is an indicator for the quality, risks, and efficiency of surgical procedures. We introduce a new methodology for assessing the duration of surgery based on anesthesiology billing records, along with reviewing its fundamental logic and limitations.

      Study Design and Setting

      The validity of the methodology was assessed through a population-based cohort of patients (n=480,986) undergoing elective operations in 246 Ontario hospitals with 1,084 anesthesiologists between April 1, 1992 and March 31, 2002 (10 years).

      Results

      The weaknesses of the methodology relate to missing data, self-serving exaggerations by providers, imprecisions from clinical diversity, upper limits due to accounting regulations, fluctuations from updates over the years, national differences in reimbursement schedules, and the general failings of claims base analyses. The strengths of the methodology are in providing data that match clinical experiences, correspond to chart review, are consistent over time, can detect differences where differences would be anticipated, and might have implications for examining patient outcomes after long surgical times.

      Conclusions

      We suggest that an understanding and application of large studies of surgical duration may help scientists explore selected questions concerning postoperative complications.

      Keywords

      What is new?
      • Long operations are prone to cause major complications, yet duration of surgery is rarely considered when studying patient outcomes postoperatively.
      • This article introduces a new methodology for quantifying the duration of surgery in individual patients undergoing operations for hundreds of different indications.
      • The results show that the methodology is reliable and valid when checked against chart review, clinical plausibility, and cross-center comparisons.
      • The main limitation of the methodology is in obtaining a suitably large and rigorous database of anesthesiology billing claims to underpin the estimates.
      • Future research in predicting and preventing surgical complications may be more fruitful if investigators account for duration of surgery in individual patients.

      1. Introduction

      The duration of an operation has clinical, economic, and theoretic implications for modern medical care. It is important for patients when gauging the extent of surgery and predicting their foreseeable speed of recovery. The anticipated duration of surgery is crucial for providers for scheduling operations and for organizing a complicated health-care system [
      • Abouleish A.E.
      • Prough D.S.
      • Whitten C.W.
      • Zornow M.H.
      The effects of surgical case duration and type of surgery on hourly clinical productivity of anesthesiologists.
      ]. The actual duration of surgery also gives direct insights regarding the net combination of procedure difficulty, patient complexity, and surgeon skill [
      • Ikhena S.E.
      • Oni M.
      • Naftalin N.J.
      • Konje J.C.
      The effect of the learning curve on the duration and perioperative complications of laparoscopically assisted vaginal hysterectomy.
      ]. Moreover, duration varies widely from less than 15 minutes for a bladder cystoscopy to more than 15 hours for conjoint twin separation.
      Analyzing surgery duration is easy if intraoperative observation is feasible or if providers are compelled to report times directly [
      • Silber J.H.
      • Rosenbaum P.R.
      • Zhang X.
      • Even-Shoshan O.
      Estimating anesthesia and surgical procedure times from Medicare anesthesia claims.
      ]. More commonly, the duration of surgery is assessed retrospectively from intraoperative records. The classic way in hospitals, for example, is to consult the anesthesiology sheet because patients can rarely report this aspect of their care for themselves. The anesthesiology sheet typically provides a profile of the operation in 15-minute increments with the start time and end time explicitly documented. Interviewing surgeons is another way to estimate surgery duration in clinical practice, although somewhat prone to reporting bias [
      • Redelmeier D.A.
      • Tu J.V.
      • Schull M.J.
      • Ferris L.E.
      • Hux J.E.
      Problems for clinical judgement: 2. Obtaining a reliable past medical history.
      ].
      We are aware of few efforts to develop computerized health services research methods for estimating surgery duration. The usual approach is chart review methodology such as is popular in infectious disease surveillance studies; however, such methods are laborious or focus on volunteer samples [
      • Giger U.F.
      • Michel J.M.
      • Opitz I.
      • Inderbitzin D.
      • Kocher T.
      • Krahenbuhl L.
      Risk factors for perioperative complications in patients undergoing laparoscopic cholecystectomy: analysis of 22,953 consecutive cases from the Swiss Association of Laparoscopic and Thoracoscopic Surgery database.
      ,
      • Leong G.
      • Wilson J.
      • Charlett A.
      Duration of operation as a risk factor for surgical site infection: comparison of English and US data.
      ]. Herein, we develop a new methodology for tackling this challenge. We apply our methodology to studying a cohort of patients undergoing surgery in Ontario. The goal is to offer a quantitative method for studying large numbers of patients. We also provide descriptive results on reliability and validity after applying our methods to real-world data.

      2. Framework

      2.1 Setting

      Ontario is the largest province of Canada accounting for about one-third of the country's total population. Anesthesiologists in Ontario are generally paid fee for service when managing a patient during surgery. Such anesthesiology bills are submitted centrally to a single payer, the Ministry of Health, through the Ontario Hospital Insurance Program (OHIP). Proven cases of fraud are rare, although some over billing and missed billing are inevitable in any payment system [
      • Stelfox H.T.
      • Redelmeier D.A.
      An analysis of one potential form of health care fraud in Canada.
      ]. OHIP bills are collated and available for analysis to researchers at the Institute for Clinical Evaluative Sciences. These databases have undergone extensive prior testing [
      • Juurlink D.N.
      • Mamdani M.M.
      • Lee D.S.
      • Kopp A.
      • Austin P.C.
      • Laupacis A.
      • et al.
      Rates of hyperkalemia after publication of the Randomized Aldactone Evaluation Study.
      ,
      • Mamdani M.
      • Juurlink D.N.
      • Lee D.S.
      • Rochon P.A.
      • Kopp A.
      • Naglie G.
      • et al.
      Cyclo-oxygenase-2 inhibitors versus non-selective non-steroidal anti-inflammatory drugs and congestive heart failure outcomes in elderly patients: a population-based cohort study.
      ,
      ].

      2.2 Total service units

      The elements of anesthesiology care are typically recorded as service units (rather than time or money). For example, anesthesiology for a simple hernia repair might entail 8 service units and for a complicated pancreatic resection might exceed 80 service units. The total number of service units is linked to a specific patient on an identified day undergoing a particular operation. The number of service units is ultimately translated into a dollar amount using a conversion factor that is related to the financial year. In 2007, for example, each service unit was valued at $12.51 CDN, so that a hernia repair totaling eight service units would equate to about a $100 fee (12.51×8).

      2.3 Set-up service units

      Total service units for anesthesiology can be subdivided into four categories, of which the first relates to preparations for the planned surgery and is hereafter termed the “set-up” fee. Surgeries are categorized from a list of more than 2,500 operations, so that each patient is assigned only one set-up fee corresponding to the main surgery. The anesthesiology code matches the surgeon's code, although distinguished by a distinct suffix. This anesthesiology code is then connected to service units using the prevailing OHIP fee schedule. For example, the set-up fee code for a simple hernia repair is E726 (two service units) and for a bipolar arthroplasty is R440 (eight service units).

      2.4 Special features service units

      Anesthesiology billing can next be increased with premiums according to special features, hereafter termed “special feature” fees. These features are ascertainable before the start of surgery and relate to complexity. Common premiums, for example, include E007 for age more than 70 years, E008 for moribund patients, or E400 for weekend surgery; hence, one operation can be connected to several special features fees. Each of these premiums is labeled with clear codes and also expressed as service units. For example, the anesthesiology special feature fee code for a moribund patient E022 generates two service units beyond the initial set-up fee.

      2.5 Special events service units

      The anesthesiology fee can also increase as the surgery progresses due to special events, hereafter termed “special events” fees. These events typically relate to additional surgical or patient factors that develop after the surgery has started; hence, one operation can theoretically be connected to several special event fees although most have no special event fees. Some events might be, for example, calling in a second anesthesiologist or providing an intraoperative nerve block. Again, specific codes are available and attached to service units. For example, the anesthesiology special event code for prone positioning is E011 and generates four service units.

      2.6 Duration service units

      The fourth and final category of anesthesiology fee relates to the duration of surgery, hereafter termed the “duration” fee and is the target for our analysis. Duration is measured in 15-minute increments, or part there of, and ascertainable only after the operation is finished. The duration fee starts when the anesthesiologist is first in attendance with the patient. The duration fee ends when the anesthesiologist is no longer in attendance. The fee is expressed in service units and is a complex function of time with some revisions over the years (at present, one unit for each 15 minutes during the first hour, two units for each 15 minutes during the next hour, and other nuances).

      3. Example

      3.1 Patient case

      Consider the following hypothetical case to illustrate anesthesiology billing. A 25-year-old man undergoes an uncomplicated appendectomy on Saturday afternoon in 1992 with a surgical time of 2 hours and 9 minutes. The OHIP schedule in that year identifies a set-up fee of five service units, a premium of 50% due to weekend work, and no other special fees. In this case, the duration service units are computed as four for the first hour, eight for the next hour, and two for the remaining 9 minutes (total=14). This anesthesia care amounts to 28.5 total service units (19×1.50), which converts to $314.64 if reimbursed under the OHIP exchange prevailing in 1992 (one service unit=$11.04).

      3.2 Overall anesthesiology fee

      The challenge for the health services research analysis is to disentangle the duration units from the total service units that constitute the anesthesiologist's reimbursement for patient care. The easiest components to separate are the premiums for special features and special events because each is tagged with a distinctive fee code. This separation makes an assumption (not critical in our work) that cases without such billing involved no special features or events. The remaining code is typically the largest entry; hereafter, referred to as the “basic” fee, and reflects the combination of a set-up fee and a duration fee. Unfortunately, the basic fee does not distinguish the specific set-up component and duration component.

      3.3 Specific anesthesiology fees

      Disentangling the basic fee into the set-up component and the duration component requires extracting the set-up fee based on prevailing standards. For example, if the basic fee for an appendectomy was 19 total service units and an appendectomy set-up fee was 5 service units, the duration fee for this case can be inferred as 14 service units (19−5). Most set-up fees have remained stable over the years, although some refinements are needed for exact calculations. In our work, we used set-up fees from 2006, expressed as service units, because they were the most recent data. In theory, if set-up fees varied dramatically over the years the techniques would still be practical assuming that remote fee schedules were available.

      3.4 Duration of surgery based on fee

      Identifying the duration of surgery for an individual patient then requires translating the duration service units into conventional time units. For example, a duration fee of 10 units indicates longer operative time than a duration fee of 5 units, although the time differential is not necessarily double because some minutes are reimbursed at a higher rate than other minutes. Under the OHIP fee schedule of the 1990s, each 15-minute block in the 1st hour generated one service unit, in hours two to eight generated two service units, and in hours beyond 8 generated three service units. Of note, the formula underwent a large modification in October 1, 2006 so that the thresholds changed from 1st, 2nd, and 8th hour to 1st, 2nd, and 3rd hour (a gain for clinicians).

      3.5 Duration of surgery in minutes

      Consider again the hypothetical patient with an appendectomy in 1992. The basic fee was 19 total service units, of which five related to the set-up fee. The remaining 14 units indicate the duration fee and imply the procedure required several hours. The first full hour amounted to four units in duration fee (1×4), the second full hour amounted to eight units in duration fee (2×4), and the third part hour amounted to the remaining two units (14 - 4 - 8). Hence, this appendectomy had a duration of 2 hours plus one more 15-minute interval. Because anesthesiology duration fees are recorded in intervals of 15 minutes, the precise time is in the range of 2 hours and 1 minute to 2 hours and 15 minutes. In our work, we infer a time of about 2 hours and 15 minutes for this hypothetical patient.

      4. Limitations

      4.1 Missing data

      The first failing in our methodology for estimating surgery duration relates to missing data. For example, when we analyzed 480,986 elective operations from April 1, 1992 to March 31, 2002 we found no anesthesiology bills in 98,194 cases (20%). This attrition occurs because some procedures do not require an anesthesiologist (e.g., colonoscopy), anesthesia is sometimes provided by clinicians who do not bill OHIP (e.g., those on alternative payment plans or nurse practitioners), and some bills are never paid (e.g., missing or faulty insurance number) [
      • Task Force on sedation and analgesia by non-anesthesiologists
      Practice guidelines for sedation and analgesia by non-anesthesiologists.
      ]. In addition, our technique restricts database searches to clinicians who have an anesthesiology specialty and thereby misses cases where general practitioners deliver anesthesia. The net result is cohort attrition, which may be relevant in estimating population-based counts and rates.

      4.2 Self-serving exaggerations

      A natural limitation of the data that are actually available relates to faulty billing by clinicians (in our analysis, the data reflect work by 1,084 anesthesiologists). Given financial realities, the general direction of such bias is probably toward exaggerating the time to generate greater reimbursement. The size of this bias is unknown, although likely to be less than an hour based on preliminary reports [
      • Kesselheim A.S.
      • Brennan T.A.
      Overbilling vs. downcoding—the battle between physicians and insurers.
      ]. The relative size of this bias, therefore, may be smaller for the more complicated operations. In many cases, this imprecision will not be important compared to imprecisions in other variables related to the analysis (such as measuring patient age to the nearest year rather than nearest month or day).

      4.3 Imprecision related to clinical diversity

      Additional sources of imprecision relate to ambiguities beyond financial self-interest. Depending on the observer, the start of each operation can be defined as the patient's arrival into the room, placement onto the table, contact with a clinician, initiation of anesthesia, first incision, or other event. The end of each operation is also susceptible to the same inconsistencies. Time recording by clinicians is inherently imperfect, as implicitly acknowledged by anesthesia fee schedules, which count time in 15-minute increments. Finally, all written records are prone to errors related to legibility, transcription, digit preference, and other processing failures.

      4.4 Upper limits

      One more special problem arises because the billing records available for researchers at the Institute for Clinical Evaluative Sciences in Ontario may exclude special cases that undergo manual funding review. For the surgical context, such cases are typically extremely complicated operations that generate over 100 service units in a basic anesthesiology fee. This limitation is unlikely to be germane in most studies because such extreme cases reflect less than 0.1% of total surgeries conducted. Moreover, such unique cases are generally suppressed from scientific presentations due to safeguards for patient confidentiality. This limitation, however, does place an upper bound of around 8 hours on the maximum duration of surgery that can be analyzed.

      4.5 Updates over years

      Ongoing negotiations between payers and providers lead to some awkwardness given that the fee schedule undergoes review with potential revisions on an annual basis. For research purposes, the simplest updates are changes to the conversion ratio that translates service units into dollar figures. For clinical purposes, however, additional updates occur to the set-up fee that reflect advances in care and other changes to anesthesia standards. In addition, the thresholds at which time blocks become valued at two or three service units will likely undergo further revisions in the years ahead. Hence, some effort needs to be devoted to maintaining archives of fee schedules when conducting retrospective analyses of surgical durations.

      4.6 National differences

      Canada's single payer medicare system implies that the methodology pertains throughout the province but is not universal because billing patterns vary in other locations. Anesthesia fee schedules in other Canadian provinces typically reflect 15-minute increments, thereby allowing the methodology to be directly adapted to other parts of the country. The exceptions are Alberta where billing reflects 5-minute increments and Prince Edward Island where billing reflects 30-minute increments; thus, the precision of estimates can vary by location. In all Canadian provinces, anesthesia billing is generally handled by clinicians and their management groups, so that recent changes in hospital systems have no direct effect on the methodology.

      4.7 International differences

      Surgery takes time, but different nations have different ways of reimbursing anesthesiologists for their time. American medicare anesthesia fee schedules are based on 15-minute increments, and are potentially straightforward to analyze because billing data explicitly list case duration. Israeli medicare anesthesia fee schedules have no accounting for time because hospital-based anesthesiologists are salaried, submit no claims, and generate no large data source for our method. Most other countries fall between the American and Israeli extremes, with complex billing systems in some places. Determining how nations outside of North America support the methodology is further challenged by the delicate aspects of privacy around billing in some locations.

      4.8 Claims base limitations

      The final major limitation of our methodology reflects the general weaknesses of billing data. Clinicians have other priorities beyond financial reimbursement so that sometimes billing data are submitted incorrectly or not at all. Billing data are transferred to accountants and other intermediates thereby raising the possibility of fallible transcription or other errors related to data entry. The internal processing by governmental financial accountants is also a bit uncertain and not open to scientific scrutiny. The entire process takes time so that substantial lags develop between the day of surgery and the day that data are available for analysis. Finally, all this work needs to respect prevailing guidelines for ensuring patient confidentiality and privacy.

      5. Results

      5.1 Sensibility

      We identified the 50 most frequent elective surgeries (n>1,000 for each procedure) conducted between April 1, 1992 and March 31, 2002 (10 years) to obtain pilot results using the methodology. For each operation, we calculated the distribution of surgical durations (Table 1). Overall, we found the method yielded plausible values for the mean and median times of each procedure, along with impressive levels of variation in some cases. The general pattern of results from the method seemed to match our clinical experience and surveillance reports from the United States and United Kingdom on surgical times [
      National Nosocomial Infections Surveillance (NNIS) System Report. Data summary from January 1992 through June 2004.
      ]. For example, the mean duration for a left-sided hemicolectomy with anterior resection was about 45 minutes longer than a right-sided hemicolectomy without anterior resection (2.8 vs. 2.1 hours, P<0.001).
      Table 1Duration for top 50 surgeries (h)
      CodeMeanMedianFirst quartileThird quartile
      Cardiac surgery with pump bypassE6504.44.33.85.0
      Total prostatovesiculectomy and lymph node dissectionS6413.93.82.84.8
      Proctectomy with abdomino-perineal resectS2143.83.52.84.5
      Abdominal aorta aneurysm repairR8023.83.53.04.5
      Retropubic radical prostatectomy/vasectomyS6513.43.32.84.0
      Pulmonary lobectomy without radical node dissectionM1433.43.32.54.0
      Revision total hip arthroplastyR2413.43.32.54.3
      Femoral–popliteal bypass without endarterectomyR7913.23.02.33.8
      Thoraco-abdominal radical nephrectomyS4163.13.02.33.8
      Spinal cord posterior laminect of one or two levelsN1852.82.52.03.3
      Proctosigmoidectomy with anterior resectionS2132.82.52.03.3
      Revision total knee arthroplastyR2442.82.52.03.5
      Left hemicolectomy with anterior resectionS1712.82.52.03.4
      Carotid endarterectomy without bypass or patchN2202.82.82.33.3
      Staging pelvic lymphadenectomy for prostatic cancerS6522.72.51.53.5
      Total hip replacement with takedown of fusionR5532.62.52.03.0
      Carotid endarterectomyR7922.42.31.82.8
      Large intestine resection with anastamosisS1672.32.01.82.8
      Spinal hemilaminectomy with osteophyte removalR4572.32.01.52.8
      Bipolar hip arthroplastyR4402.32.31.82.8
      Total knee hemiarthroplasty of both compartmentsR4412.22.01.82.5
      Right hemicolectomy including ileumS1662.12.01.52.5
      Hysterectomy through abdominal approachS7581.81.51.42.0
      Hysterectomy through vaginal approachS7571.71.51.42.0
      Laparotomy without biopsyS3121.71.41.32.0
      Vitrectomy of eye by infusing suction cutterE1481.71.51.32.0
      Rotator cuff repair of shoulderR5941.61.51.41.8
      Scleral resection or buckling procedureE1521.61.51.31.8
      Open repair of trochanteric hip fractureF1001.61.41.31.8
      Repair of massive incisional herniaS3441.61.41.31.8
      Oophorectomy or oophorocystectomyS7451.61.41.31.8
      Modified radical mastectomy without biopsyR1091.61.51.31.8
      CholecystectomyS2871.51.41.31.8
      Major forefoot reconstructionR3601.51.41.11.8
      Retropubic urethropexy procedureS5491.51.41.11.5
      Cataract extraction (all types)E1401.41.41.31.5
      Postoperative ventral hernia repairS3401.41.31.11.5
      Anterior and posterior vaginal repairS7171.41.31.11.5
      Simple mastectomy without biopsyR1081.31.31.11.4
      MediastinoscopyZ3301.31.31.11.4
      Partial mastectomy with wedge resectionR1111.21.11.01.4
      Transurethral resection of prostateS6551.21.31.01.4
      Anterior or posterior vagina repairS7161.21.11.01.4
      Permanent cardiac electrodes and generator insertionZ4441.21.11.01.3
      Inguinal or femoral herniotomyS3231.11.00.81.3
      Breast tissue biopsyR1071.01.00.81.1
      Excision of multiple miscellaneous tumorsZ6340.90.80.81.1
      Unilateral orchidectomyS5890.90.80.81.0
      Excision of single tumor 1–2 cm diameterZ6320.80.80.51.0
      Hemorrhoidectomy without sigmoidoscopyS2470.80.80.81.0

      5.2 Correspondence

      Our next evaluation of the method was based on a chart review of 50 patients undergoing operations in one year (2002). For each patient, we obtained the actual reported surgical time from the in-hospital anesthesia chart records. We then compared these data to the estimated surgery duration times that were based on our method and computed in a manner blind to the actual duration of surgery. For example, we found that the duration of surgery for the first patient was 3.0 hours when based on our method compared to 3.0 hours when based on chart review. The overall pattern of results showed high correlation with a small trend toward upcoding of anesthesiology times (Pearson=0.94, P<0.001). In only one case did the method underestimate surgical duration by more than 30 minutes (Fig. 1).
      Figure thumbnail gr1
      Fig. 1Scatterplot of surgical times for individual patients comparing values in medical record to values estimated from anesthesiology billing records. X-axis and Y-axis both show time expressed in hours, along with line of identity shown as a diagonal. Correlation coefficient (r) based on Pearson's statistic. Main finding shows good correlation with some upward displacement of estimated times relative to chart review times.

      5.3 Consistency

      We checked the reliability of our method by examining the same surgical procedure conducted on different years and patients. To do so, we split the overall 10-year data set into patients treated during the first 5 years (first half) and the second 5 years (second half). The mean surgical duration times were then computed on each of the top 50 procedures for the two separate samples. The overall results showed remarkable consistency (Pearson=0.98, P<0.001). For example, the most common operation (transurethral resection of the prostate) had mean durations of surgery in the first and second 5 years that were within 2 minutes of each other (1.24 vs. 1.26 hours, respectively). Moreover, visual inspection revealed no major anomalies and in only one case were the two mean estimates off by more than 30 minutes (Fig. 2).
      Figure thumbnail gr2
      Fig. 2Scatterplot of mean surgical times for individual procedures comparing values in first 5 years to values in second 5 years of cohort accrual. X-axis and Y-axis both show time expressed in hours, along with line of identity shown as a diagonal. Correlation coefficient (r) based on Pearson's statistic. Main finding shows good correlation with some increase in variance for relatively longer surgeries.

      5.4 Differentiation

      We also explored whether our method detected differences where differences would be anticipated. To do so, we classified each operation using available hospital codes as being conducted in a teaching hospital or a community hospital. We reasoned that the same procedure conducted in a teaching hospital would tend to take a bit longer due to either added case complexity or academic teaching [
      • Dimick J.B.
      • Cowan J.A.
      • Colletti L.M.
      • Upchurch G.R.
      Hospital teaching status and outcomes of complex surgical procedures in the United States.
      ]. The results were in accord with this idea, with about a 11-minute mean increase (95% confidence interval: 6, 16; P<0.001) in teaching hospitals relative to community hospitals. For example, the mean duration of surgery for a cholecystectomy was about 15 minutes longer in a teaching hospital than in a community hospital (1.72 vs. 1.50 hours, P<0.001). Increases in time were evident for both longer and shorter operations (Fig. 3).
      Figure thumbnail gr3
      Fig. 3Scatterplot of mean surgical times for individual procedures comparing values in teaching hospitals to values in community hospitals. X-axis and Y-axis both show time expressed in hours, along with line of identity shown as a diagonal. Correlation coefficient (r) based on Pearson's statistic. Main finding shows good correlation with upward displacement in teaching compared to community hospitals.

      5.5 Implications

      We also conducted one analysis to examine the clinical implications of long surgical times; namely, by correlating duration of surgery with total length of stay in hospital. As expected, longer surgeries generally required longer inpatient recoveries (Pearson=0.73, P<0.001), so that each hour of surgery was followed by about 3 days of hospital stay (95% confidence interval: 2, 4). For example, the mean duration of surgery for an abdominal aorta aneurysm repair was about 3.8 hours and the mean duration of hospital stay was 12.4 days. The overall results showed moderate consistency (Fig. 4). The largest anomaly was for an open repair of a trochanteric hip fracture that involved relative short surgery (averaging about 1.8 hours) but relatively long hospital stays (averaging about 16 days).
      Figure thumbnail gr4
      Fig. 4Scatterplot of mean surgical times and length of hospital stay for individual procedures. X-axis shows time expressed in hours, Y-axis shows time expressed in days, diagonal line represents ratio of 1 hour to 3 days. Correlation coefficient (r) based on Pearson's statistic. Main finding shows good correlation with largest exception being trochanteric hip fracture surgery (surgery=1.8 hours, stay=16 days).

      6. Summary

      Estimates of surgical duration are rare in health services research and not previously available in Canada to our knowledge. Most literature on surgical duration, instead, reflects clinical surveillance in the United States and the United Kingdom with limited power, crude statistical models, or selection bias [
      National Nosocomial Infections Surveillance (NNIS) System Report. Data summary from January 1992 through June 2004.
      ,
      • Traverso L.W.
      • Koo K.P.
      • Hargrave K.
      • Unger S.W.
      • Roush T.S.
      • Swanstrom L.L.
      • et al.
      Standardizing laparoscopic procedure time and determining the effect of patient age/gender and presence or absence of surgical residents during operation. A prospective multicenter trial.
      ,
      • Ammori B.J.
      • Larvin M.
      • McMahon M.J.
      Elective laparoscopic cholecystectomy: preoperative prediction of duration of surgery.
      ,
      • Pandit J.J.
      • Carey A.
      Estimating the duration of common elective operations: implications for operating list management.
      ]. Herein, we introduce a method that is feasible, rigorous, novel, ethical, and reasonable. We also provide technical details about the method, list the main limitations, and offer validating findings. The goal is to foster health services research toward making surgical care in the future better than it is today. Surgical site infections, postoperative delirium, and many other complications are widespread problems that partially reflect surgical duration.
      Our methodological approach has several strengths. First, the underlying anesthesiology database is available, objective, important, directly linked to other health-care databases, and undergoes ongoing scrutiny as a core-funding program. Second, the method yields data that are immediately meaningful to patients, clinicians, and policy makers. Third, the method avoids biases related to the propagation of human error given that the anesthesiologist is typically not the responsible surgeon nor the main postoperative consultant. Fourth, the method avoids the costs, delays, and errors of manual chart review. Finally, the necessary computer programing is straightforward and requires no proprietary software.
      Surgical time is an appealing factor for investigators interested in anticipating and preventing postoperative complications. One application might be in developing new prediction rules, which can be applied before surgery for distinguishing high-risk from low-risk surgical cases. Other applications might be in explaining postoperative complications or adjusting for different baseline predispositions toward surgical complications. The main limitation of our method is in obtaining the detailed anesthesiology billing records. The main strength of our method is the capacity to study thousands of patients with relative ease. The most relevant studies are likely to be on the unintended adverse events from surgical procedures.

      Acknowledgments

      This work was supported by the Canada Research Chair in Medical Decision Sciences, the Canadian Institutes of Health Research, the National Institutes of Health Resuscitation Outcomes Consortium, the Clinician Scientist Training Program of the University of Toronto, and the PSI Foundation of Ontario. All authors have participated in the study design, interpretation of results, and approval of the final draft. Donald Redelmeier had full access to all the data, final authority for the decision to submit for publication, and ongoing responsibility for accuracy. We are indebted to the following people for helpful comments on specific points: Marianne Alexiadis, Daniel Hackam, Damon Scales, Homer Tien, Edward Etchells, Michael Schull, Jack Williams, and Duminda Wijeysundera.

      References

        • Abouleish A.E.
        • Prough D.S.
        • Whitten C.W.
        • Zornow M.H.
        The effects of surgical case duration and type of surgery on hourly clinical productivity of anesthesiologists.
        Anesth Analg. 2003; 97: 833-838
        • Ikhena S.E.
        • Oni M.
        • Naftalin N.J.
        • Konje J.C.
        The effect of the learning curve on the duration and perioperative complications of laparoscopically assisted vaginal hysterectomy.
        Acta Obstet Gynecol Scand. 1999; 78: 632-635
        • Silber J.H.
        • Rosenbaum P.R.
        • Zhang X.
        • Even-Shoshan O.
        Estimating anesthesia and surgical procedure times from Medicare anesthesia claims.
        Anesthesiology. 2007; 106: 346-355
        • Redelmeier D.A.
        • Tu J.V.
        • Schull M.J.
        • Ferris L.E.
        • Hux J.E.
        Problems for clinical judgement: 2. Obtaining a reliable past medical history.
        CMAJ. 2001; 164: 809-813
        • Giger U.F.
        • Michel J.M.
        • Opitz I.
        • Inderbitzin D.
        • Kocher T.
        • Krahenbuhl L.
        Risk factors for perioperative complications in patients undergoing laparoscopic cholecystectomy: analysis of 22,953 consecutive cases from the Swiss Association of Laparoscopic and Thoracoscopic Surgery database.
        J Am Coll Surg. 2006; 203: 723-728
        • Leong G.
        • Wilson J.
        • Charlett A.
        Duration of operation as a risk factor for surgical site infection: comparison of English and US data.
        J Hosp Infect. 2006; 63: 255-262
        • Stelfox H.T.
        • Redelmeier D.A.
        An analysis of one potential form of health care fraud in Canada.
        CMAJ. 2003; 169: 118-119
        • Juurlink D.N.
        • Mamdani M.M.
        • Lee D.S.
        • Kopp A.
        • Austin P.C.
        • Laupacis A.
        • et al.
        Rates of hyperkalemia after publication of the Randomized Aldactone Evaluation Study.
        N Engl J Med. 2004; 351: 543-551
        • Mamdani M.
        • Juurlink D.N.
        • Lee D.S.
        • Rochon P.A.
        • Kopp A.
        • Naglie G.
        • et al.
        Cyclo-oxygenase-2 inhibitors versus non-selective non-steroidal anti-inflammatory drugs and congestive heart failure outcomes in elderly patients: a population-based cohort study.
        Lancet. 2004; 363: 1751-1756
      1. Tu J.V. Penfold S.P. McColgan P. Laupacis A. Access to health services in Ontario: ICES Atlas. Institute for Clinical Evaluative Sciences, Toronto2005
        • Task Force on sedation and analgesia by non-anesthesiologists
        Practice guidelines for sedation and analgesia by non-anesthesiologists.
        Anesthesiology. 2002; 96: 1004-1017
        • Kesselheim A.S.
        • Brennan T.A.
        Overbilling vs. downcoding—the battle between physicians and insurers.
        N Engl J Med. 2005; 352: 855-857
      2. National Nosocomial Infections Surveillance (NNIS) System Report. Data summary from January 1992 through June 2004.
        Am J Infect Control. 2004; 32: 470-485
        • Dimick J.B.
        • Cowan J.A.
        • Colletti L.M.
        • Upchurch G.R.
        Hospital teaching status and outcomes of complex surgical procedures in the United States.
        Arch Surg. 2004; 139: 137-141
        • Traverso L.W.
        • Koo K.P.
        • Hargrave K.
        • Unger S.W.
        • Roush T.S.
        • Swanstrom L.L.
        • et al.
        Standardizing laparoscopic procedure time and determining the effect of patient age/gender and presence or absence of surgical residents during operation. A prospective multicenter trial.
        Surg Endosc. 1997; 11: 226-229
        • Ammori B.J.
        • Larvin M.
        • McMahon M.J.
        Elective laparoscopic cholecystectomy: preoperative prediction of duration of surgery.
        Surg Endosc. 2001; 15: 297-300
        • Pandit J.J.
        • Carey A.
        Estimating the duration of common elective operations: implications for operating list management.
        Anaesthesia. 2006; 61: 768-776