Determining Optimal Route of Hysterectomy for Benign Indications: Clinical Decision Tree Algorithm.

More than 430,000 hysterectomies were performed in the United States in 2010, most commonly for fibroids (40.7%) and endometriosis (17.7%) ( 1 , 2 ). The American College of Obstetricians and Gynecologists published the committee opinion “Choosing the Route of Hysterectomy for Benign Disease,” ( 2 ) concluding that vaginal hysterectomies have fewer complications and better outcomes than laparoscopic or abdominal, which was reiterated in a Cochrane review ( 3 ). Surgical approach has generally not been standardized because it has been provider-dependent based on physician preferences with emphasis on indications, patient physical characteristics, concomitant procedures, and surgeon experience. Kovac et al ( 4 ) published an expert opinion algorithm on benign hysterectomies, showing success of vaginal hysterectomy in many cases previously performed through laparotomy ( 4 - 6 ). No current validated, evidence-based methods are available to assist gynecologists in choosing the appropriate surgical route for their patients in the robotic surgery era. A national survey found declining overall numbers of hysterectomy, as well as of abdominal and vaginal hysterectomies, with increasing robotic procedures ( 7 ). The reported opinions of residents’ and program directors’ included a concern that some faculty may be uncomfortable with, or unable to perform, vaginal hysterectomy. Current evidence suggests that with the initiation of endoscopic approaches, including robotic surgery, vaginal hysterectomies are underutilized. We retrospectively applied an evidence-based clinical decision tree algorithm to evaluate the surgical approach to simple hysterectomy for benign indications. Our aim was to study the change in practice pattern after the introduction of robotic-assisted surgery and to compare expected and actual rates of vaginal and robotic routes as they relate to outcomes and cost implications.

Five gynecologic surgeons at our institution (J.J.S., J.A.O., J.N.B.-G., S.C.D., and J.B.G.) reviewed published hysterectomy route algorithms and created a modified clinical decision tree algorithm, taking vaginal access (eg, caliber, uterine descent) and uterine size into account ( 4 - 6 ). Addition of robotic assistance as a surgical option was a notable change from previously published algorithms ( Figure 1 ). The Mayo Clinic Institutional Review Board approved this study, and an institutional surgical database was used to identify hysterectomies for benign indications between January 1, 2009, and December 31, 2013 (post-robot cohort). A second cohort (pre-robot) that underwent hysterectomy between January 1, 2004, and December 31, 2005, was identified and used as a baseline to evaluate practice change after the introduction of robotic hysterectomy at our facility in 2007. We excluded 2006 through 2008 to decrease the potential learning curve effects of robotic technology on our results ( 8 ). Strict exclusion criteria were used to create a pure cohort of benign uterine disease, thereby minimizing clinical scenarios that influence the preference toward a particular surgical route, such as cancers, pelvic organ prolapse, and risk-reducing and concomitant procedures ( Box 1 ). We included vaginal, robotic, and abdominal hysterectomies in this study. Total laparoscopic and laparoscopy-assisted vaginal hysterectomies were excluded because of inconsistency in surgical techniques, particularly in the extent of dissection done laparoscopically. During 2004 to 2005, the Division of Gynecologic Surgery at our institution consisted of 9 surgeons, of whom 7 (77.8%) had >2 years of post-fellowship experience as of January 2005. During 2009 to 2013, the Division of Gynecologic Surgery consisted of 13 surgeons (8 overlapping with 2004 to 2005), of whom 10 (76.9%) had >2 years of post-fellowship experience as of January 2011. Patient characteristics, history, physical examination, laboratory and imaging results, surgical information, postoperative visits, and telephone encounters were abstracted from electronic health records by J.J.S. and D.A.C.L. Postoperative complications were categorized and graded according to the expanded Accordion Severity Grading of Postoperative Complications system ( 9 ). The abstracted data were entered directly into a REDCap electronic data capture system, hosted at Mayo Clinic in Rochester, Minnesota, which was designed specifically for this study with the assistance of the study statistician. At the onset of the abstraction, J.J.S. and D.A.C.L. together reviewed the list of parameters to be abstracted and the prespecified response coding in REDCap and determined where to identify the information in the health record. The study statistician performed logical edit checks, and any data discrepancies were reviewed and resolved before the analysis. Dictations are used to document examinations at our institution, which can result in various characteristics noted in the pelvic examination. To ensure consistency, we assumed the inclusion of “normal” or the omission of a description of vaginal caliber, uterine size, or mobility indicated adequate vaginal access and uterine size of <12 weeks of gestation. If no uterine size was dictated, preoperative imaging (if available) was used to estimate the size with previously published uterine weight estimation calculators ( 10 ). If neither an examination nor imaging was available, the pathologic weight of the specimen was used to assign uterine size for retrospective use in the algorithm. For each patient, the expected surgical route was determined by applying the algorithm ( Figure 1 ). Uterine size ≤280 g was chosen as the cutoff for the assignment of vaginal hysterectomy based on the College's Committee Opinion ( 2 ). In any clinical scenario, if the surgeon used a less invasive route than expected by this algorithm, the case was not considered a deviation. For example, for a patient with a large uterus (14 weeks of gestation size) that is accessible transvaginally, the algorithm expected a laparoscopic or robotic approach. If a vaginal hysterectomy was performed instead, the case was not considered a deviation. Statistical analysis was performed with SAS software version 9.3 (SAS Institute Inc). Comparisons between the procedure cohorts were evaluated with χ 2 or Fisher exact test for categorical variables, Wilcoxon rank sum test for ordinal variables, and 2-sample t test for continuous variables. P values were 2-sided, and P <.05 was considered significant. Three separate analyses were performed to identify factors associated with postoperative health: urinary tract infection, surgical site infection, and postoperative complication Accordion grade ≥3, which indicates use of an invasive intervention or treatment. Univariate and multivariable logistic regression models were fit to identify factors associated with each outcome. All of the factors were considered in a multivariable analysis using traditional stepwise and backward variable selection methods to identify a parsimonious model; variables with a P value <.05 were retained in the final model. In addition, variable selection was explored by fitting a cross-validated penalized logistic regression model with a lasso penalty using the GLMSELECT procedure (SAS Institute Inc) and coding the outcome as ±1 ( 11 ). The lasso approach performs variable selection by shrinking the parameter coefficients for associated variables while setting the parameter coefficients of unassociated variables to zero. These 2 approaches identified the same subset of variables when the penalty was loosened, and a final multivariable logistic model was fit using this subset of variables. Associations were summarized using odds ratio (OR) and corresponding 95% confidence interval (CI) estimated from the logistic models. Lastly, a cost estimate was performed comparing expected and actual routes of hysterectomy in the post-robot cohort. Costs were estimated on the basis of historical institutional data, as well as previously published route estimates ( 12 ).

Among the 1,318 patients in the post-robot cohort (2009-2013), 120 (9.1%) were excluded because of inability to assign an expected hysterectomy route due to insufficient information on accessibility of the uterus, leaving 1,198 patients for analysis. Among the 497 patients in the pre-robot cohort (2004-2005), 24 (4.8%) were excluded (2 lacked a uterine size and 22 lacked information on accessibility of the uterus), leaving 473 patients for analysis. Among the patients in the analysis cohorts, uterine size was obtained from the pelvic examination of 81.1% of patients, from the preoperative imaging of 5.7%, and from final pathologic analysis of 13.2%. The cohorts were similar in proportion of uteri >280 g based on final pathologic analysis (post-robot vs pre-robot, 19.9% vs 20.9%), no vaginal parity (30.7% vs 28.4%), and prior laparotomy (38.2% vs 36.4%). Table 1 summarizes the route performed and the expected route per the algorithm for each cohort. In the pre-robot cohort, 320 hysterectomies (67.7%) were performed vaginally and 153 (32.3%) with laparotomy; 46 of 305 hysterectomies (15.1%) were performed with laparotomy when vaginal hysterectomy was the algorithm's expected route. In the post-robot cohort, 672 hysterectomies (56.1%) were performed vaginally, 390 (32.6%) robotically, and 136 (11.4%) through laparotomy. Among the 743 hysterectomies in the post-robot cohort for which the vaginal route was expected, 38 (5.1%) involved laparotomy and 154 (20.7%) involved robotic technique. Thus, when the algorithm indicated vaginal hysterectomy was feasible, more frequent deviation occurred after robotic surgery was initiated (post-robot vs pre-robot, 25.8% vs 15.1%; P <.001). Among the post-robot cohort patients for whom the vaginal route was indicated by the algorithm, the woman's age, American Society of Anesthesiologists score, postoperative blood transfusions, severe postoperative complications (Accordion complications grade ≥3), and readmission rates were not significantly different between vaginal and robotic hysterectomy ( Table 2 ). Body mass index, vaginal and cesarean parities, preoperative diagnosis of endometriosis or dysmenorrhea, uterine weight >280 g, operative time, urinary tract infection, and surgical site infection were significantly more common in those patients where robotic hysterectomy (rather than expected vaginal hysterectomy as per the algorithm) was performed ( Table 2 ). Vaginal parity and cesarean parity were significantly different between the vaginal and robotic approaches ( P <.001). A vaginal parity of zero was noted in 50 of the vaginal hysterectomies (9.1%); 18 (47.4%), abdominal hysterectomies; and 88 (57.5%), robotic approaches performed when the vaginal route was expected. Robotic hysterectomy was performed for women with vaginal parity and when vaginal hysterectomy was expected: 65 (42.5%) had ≥1 vaginal deliveries and 37 (24.2%) had ≥2 vaginal deliveries. Vaginal hysterectomy was expected and performed for 49 patients (8.9%) who had 1 cesarean delivery, 17 (3.1%) with 2, and 6 (1.1%) with ≥3. Compared with the robotic group, the difference was significant: 24 patients (15.7%) with 1, 28 (18.3%) with 2, and 8 (5.2%) with ≥3 cesarean deliveries ( P <.001). When the hysterectomies were performed for a preoperative diagnosis of endometriosis, a statistically significant difference was found between the actual routes of hysterectomy when the algorithm expected a vaginal hysterectomy: vaginal, 15 (2.7%); robotic, 37 (24.0%); and abdominal 13 (34.2%) ( P <.001). Of the vaginal hysterectomies performed for endometriosis, none involved intraoperative route conversions. In cases where vaginal hysterectomy was expected, 4 patients (0.7%) required intraoperative route conversion among the vaginal hysterectomies and 2 (1.3%) among the robotic hysterectomies. Of the 4 vaginal cases, 2 were converted to laparotomy for a urinary tract injury and adhesions; 1 to laparoscopy to obtain adequate utero-ovarian pedicle hemostasis; and 1 to laparoscopy first, then laparotomy because of dense adhesions. Robotic cases were converted to laparotomy because of adhesions and inability to gain access to the peritoneal cavity. No significant differences were found in overall rates of postoperative complications between the robotic and vaginal approaches for the patient group in whom vaginal approach was expected. However, rates of UTI (4.1% vs 8.1%, P =.05) and surgical site infection (0.2% vs 4.7%, P <.001) were higher in the robotic group ( Table 2 ). No statistically significant difference was detected in age at surgery, American Society of Anesthesiologists score, or postoperative blood transfusion for patients who had vaginal or robotic hysterectomy, among the post-robot cohort patients from whom the robotic route was indicated by the algorithm ( Table 3 ). Operative times and history of cesarean delivery differed among the groups. Severe postoperative complications (Accordion complications grade ≥3) were greater in robotic than vaginal hysterectomies (4.0% vs 0.0%, P =.05) but were no different in robotic vs abdominal hysterectomies (4.0% vs 6.7%, P =.48). Similarly, readmission rates were higher among robotic than vaginal hysterectomies (6.1% vs 1.0%, P =.04) but were not different between robotic and abdominal procedures (6.1% vs 11.9%, P =.16). Of the 1,318 patients in the post-robot cohort, 1,269 had sufficient postoperative follow-up within the 30 days after surgery or a complication within the first 30 days. Among the 1,269 patients, 188 (14.8%) had at least 1 postoperative complication, of which the worst complication was an Accordion grade 1 for 55 patients and a grade 2 for 103 patients; 30 patients (2.4%) had a grade 3 or higher complication. Of interest, 29 (2.3%) patients had a surgical site infection and 60 (4.7%) had a urinary tract infection. The results from the univariate analyses evaluating factors for an association with presence of an Accordion grade ≥3 postoperative complication are summarized in Appendix 1 , available online at http://links.lww.com/xxx . Based on univariate analyses, past history of a stroke or transient ischemic attack (TIA), longer operating time, and route of hysterectomy were each significantly associated with having grade ≥3 postoperative complication ( P <.05). The multivariable analysis was limited with only 30 events; a grade ≥3 postoperative complication was 5.5 (95% CI for adjusted OR, 1.2-26.2) times more likely among patients with past history of a stroke or TIA compared with those who have no past history, and 4.6 (95% CI, 1.6-13.3) and 3.4 (95% CI, 1.4-8.3) times more likely among patients with abdominal or robotic hysterectomy, respectively, than those with vaginal hysterectomy. When the same list of factors were evaluated for an association with surgical site infection, the following factors were significant, based on univariate analyses ( P 2, diabetes, coronary artery disease, asthma, longer operating time, and route of hysterectomy ( Appendix 2 , available online at http://links.lww.com/xxx ). On multivariable analysis, patients with an American Society of Anesthesiologists score >2 (adjusted OR, 3.8; 95% CI, 1.7-8.4), longer operating time (adjusted OR, 1.5 per 1-hour increase; 95% CI, 1.0-2.2), and who underwent an abdominal or robotic hysterectomy (adjusted OR, 13.6; 95% CI, 2.8-66.7 and adjusted OR, 7.5; 95% CI, 1.5-37.1) were more likely to have a surgical site infection. None of the evaluated factors were significantly ( P <.05) associated with development of a postoperative urinary tract infection based on univariate analysis (data not shown). When stratified by individual surgeon (lettered A-M in Table 4 ), the percentage of surgeons who followed the algorithm when the expected route was vaginal varied from 50% to 100%. Of note, the surgeons also perform a large volume of hysterectomies for clinical scenarios excluded in this study, and therefore the surgeons have a larger total number of hysterectomies performed annually in each respective practice than is represented in Table 4 . Woelk et al ( 12 ) reported the unadjusted mean cost of vaginal, robotic, and abdominal hysterectomies as $10,318, $14,402, and $15,079, respectively, at our institution. Given that 38 hysterectomies were performed abdominally and 154 robotically when the algorithm predicted a vaginal approach, the costs saved when following this algorithm would have been approximately $800,000 for the post-robotic surgery cohort (2009-2013). The 30-day cost was not determined.

Despite similarities in uterine size, absence of vaginal parity, and history of laparotomy between the 2 cohorts, abdominal and vaginal hysterectomies decreased after the introduction of robotic technology. Patients benefit from a decrease in laparotomies that robotic technology provides; however, the drawback is the associated underutilization of vaginal surgery. This is illustrated by a greater deviation from the algorithm in post-robot cohort, suggesting that the hysterectomies that would have been done vaginally in the pre-robot cohort were done robotically in the post-robot cohort. Among patients in the post-robot cohort who were expected to have vaginal hysterectomy but had robotic surgery performed, patients had longer operations and higher infection rates than those who underwent a vaginal hysterectomy. Conversely, when a robotic approach was expected and a vaginal was performed, patients had fewer complications (10.7% among vaginal and 20% among robotic). These findings indicate deviating toward the less invasive approach does not increase complications, consistent with other publications ( 13 - 18 ). In our study, when the expected route was vaginal, a difference was found in the percentage of patients with a specimen uterine weight >280 g among the routes performed (5.3% among vaginal and 10.4% among robotic). Either the surgeons were not accurately determining uterine size or their documentation was not sufficient. A national downward trend has been seen for hysterectomies since the introduction of less invasive alternatives to treat menstrual disorders ( 1 , 19 ). Although hysterectomies are decreasing, the number of surgical approaches is increasing, posing a challenge to trainees. Each modality takes repetition to master, which has been well documented in robotic surgery ( 8 ). The Accreditation Council for Graduate Medical Education requires a minimum number of 15 vaginal hysterectomies during residency, which is less than laparoscopic ( 20 ) and abdominal (35) ( 20 ). Median hysterectomy numbers were 17 vaginal, 37 laparoscopic, and 50 abdominal for residents in 2013 ( 21 ). The results of this study indicate that most hysterectomies can be performed vaginally or endoscopically, suggesting that these modalities should be emphasized over abdominal hysterectomy. Recent literature contains suggestions for improving education through simulation models ( 22 ), objectively evaluating vaginal surgical skills ( 23 ), and techniques to address a difficult hysterectomy ( 24 ). Another way to improve trainee experience is to increase vaginal hysterectomies. Collaboration or referral to those who can safely perform vaginal hysterectomies has been suggested to increase vaginal surgery and provide effective care ( 25 ). Support of this includes data that high-volume surgeons have shorter operative times, lower morbidity and mortality rates, including operative injury, transfusion rates, and intensive care unit admissions ( 26 , 27 ). Certainly, cost implications are a factor in the variation in surgical approach to hysterectomy. Several publications have compared costs among the various surgical approaches, consistently favoring vaginal surgery ( 12 , 28 , 29 ). In January 2015, United HealthCare Services, Incorporated, released a network bulletin outlining changes in the company's requirements for preauthorization of hysterectomy procedures ( 30 ). Outpatient vaginal hysterectomies would not need prior authorization, and a denial would be issued when the authorization process was not completed for all other approaches. If claims are denied, they cannot be billed because they are not believed to be medically necessary ( 30 ). The algorithm presented can serve as a guide for choosing the most appropriate route of hysterectomy, reducing health care delivery costs, as shown by the estimated $800,000 savings to our division, as well as complying with the third-party payer requirements. Furthermore, larger implications for surgical cost are present with postoperative complications. Bakkum-Gamez et al ( 31 ) found that for patients having a hysterectomy for endometrial cancer, each superficial surgical site infection added nearly $10,000 to the 30-day costs, as well as nonmeasurable costs including time off of work and lost wages. Our study has multiple strengths, including the cohort size of nearly 1,700 patients. Every surgeon in our division was included, improving the variability in individual practices. In addition, we included a pre-robotic cohort to compare surgical practice pre- and post-robotic introduction at our institution, with a gap in time allowing for the learning curve of robotic surgery. Study limitations include our retrospective application of the algorithm, because we were likely unable to account for all variables used in decision-making. Our algorithm used vaginal access and uterine size to determine preferred approach. Multivariate regression shows that surgical complications are independently associated with surgical route. Given that many of the postoperative outcomes that we evaluated were rare, the study had limited power to evaluate these associations, and the multivariable analysis to identify factors associated with postoperative outcomes should be cautiously interpreted as exploratory. Finally, we were required to exclude 7.9% of patients because the expected route could not be assigned. Previously published algorithms to guide physician selection of hysterectomy route successfully improved the rates of vaginal hysterectomy and, subsequently, the clinical outcomes of patients ( 4 ). Our algorithm is similar, with minor adjustments—most notably, the addition of the robotic hysterectomy as a surgical option. Use of our algorithm likely results in a higher rate of vaginal hysterectomy, improvement in patient outcomes, and lower costs of health care delivery. Prospective use of a clinical decision tree algorithm is underway at our institution as the next step to improve health care delivery and to validate these findings.