Predicting native papilla biliary cannulation success using a multinational Endoscopic Retrograde Cholangiopancreatography (ERCP) Quality Network

Background Success in deep biliary cannulation via native ampullae of Vater is an accepted measure of competence in ERCP training and practice, yet prior studies focused on predicting adverse events alone, rather than success. Our aim is to determine factors associated with deep biliary cannulation success, with/ without precut sphincterotomy. Methods The ERCP Quality Network is a unique prospective database of over 10,000 procedures by over 80 endoscopists over several countries. After data cleaning, and eliminating previously stented or cut papillae, two multilevel fixed effect multivariate models were used to control for clustering within physicians, to predict biliary cannulation success, with and without allowing “precut” to assist an initially failed cannulation. Results 13018 ERCPs were performed by 85 endoscopists (March 2007 - May 2011). Conventional (without precut) and overall cannulation rates were 89.8% and 95.6%, respectively. Precut was performed in 876 (6.7%). Conventional success was more likely in outpatients (OR 1.21), but less likely in complex contexts (OR 0.59), sicker patients (ASA grade (II, III/V: OR 0.81, 0.77)), teaching cases (OR 0.53), and certain indications (strictures, active pancreatitis). Overall cannulation success (some precut-assisted) was more likely with higher volume endoscopists (> 239/year: OR 2.79), more efficient fluoroscopy practices (OR 1.72), and lower with moderate (versus deeper) sedation (OR 0.67). Conclusion Biliary cannulation success appears influenced by both patient and practitioner factors. Patient- and case-specific factors have greater impact on conventional (precut-free) cannulation success, but volume influences ultimate success; both may be used to select appropriate cases and can help guide credentialing policies.


Background
Predicting quality is an important part of determining training, credentialing, and recredentialing thresholds. Endoscopic retrograde cholangiopancreatography (ERCP) is widely performed to diagnose and treat pancreatic and biliary disorders, at 1 per 1000 population, estimated at 3-500,000 annually in the US. Post-ERCP pancreatitis is the most common adverse event (1% to 7%, up to 10-20% in high risk patients) [1][2][3][4], accounts for most of the related mortality (0.1%), an estimated 500 deaths/year in the US [5,6]. Pancreatitis is more likely after repeatedly failed cannulation [7], thus maximizing cannulation success is important, not only for avoiding costly repeat and rescue procedures, but also for decreasing adverse events.
In community practice, ERCP is most often performed for biliary diseases [8], with the rate of successful biliary cannulation remaining the key performance metric. Minimum standards of 80-90% have been proposed internationally [9,10], yet wide practice variations (54%-98%) exist. This rate variation can only be partly explained by variable exclusions of precut sphincterotomy, and cut/ stented papillae [11][12][13][14]. Although "pre-cut" sphincterotomy to facilitate access in difficult cases may increase risk in inexperienced hands, recent meta-analyses of randomized trials concluded that precut appears safer than persistence for experienced endoscopists [15]; and may arguably no longer be regarded as "failure", at least as a secondary outcome. It would also seem preferable to exclude previously stented/cut papillae.
Volumes, training, and practice conditions may play a role [16,17], yet these factors have not consistently been strong predictors [8]. Difficulty of the procedure may also contribute [18,19]. The influence of other factors (trainees, comorbidity, sedation) remains largely unknown. Therefore, data gathered in a unique multinational ERCP Quality Network, were used to investigate the predictors for native papilla biliary cannulation success (with and without "precut") using multilevel logistic regression analyses.

Data sources and study cohort
The data were retrieved from the ERCP Quality Network database, a web-based registry of prospectively entered, consecutive, self-reported, anonymous data from a variety of ERCP practices worldwide ( March 28, 2007-May 18, 2011. Data was cleaned, excluding cut/stented papillae, physicians contributing <30 cases, and cases without biliary cannulation attempts. Informed consent was waived by the Medical University of South Carolina Institutional Review Board, and the study of this database without patient identifiers was granted an exempt status.

Outcomes and study variables
The primary outcome was "conventional" deep biliary cannulation success, with use of precut considered a "failure". The secondary outcome was overall/ultimate biliary cannulation success (allowing precut, if success occurred during that same procedure). Deep cannulation was defined as the tip of the catheter passing freely beyond the sphincter segment. Other cannulation maneuvers (e.g. wire-guided cannulation or temporary pancreatic stenting), without precut, were considered "conventional" techniques for our purposes.
Endoscopist-specific data were gathered at a baseline survey before Network participation. It had 6 categories for the endoscopist's prior hands-on training volume: 0 (no formal training), 1-100, 101-150, 151-200, 201-250, >250 procedures. Years of prior ERCP practice was recorded. Lifetime volume (estimated cumulative number of prior ERCPs) and annual volume (estimated by number of ERCPs performed the preceding year) were surveyed.
In addition to the baseline survey data, 2 endoscopistspecific variables were created as surrogates of "efficiency" in standard cases: grade-1-case procedure time (median time from inserting to removing scope in grade-1-difficulty cases); grade-1-case fluoroscopy time (median fluoroscopy time in grade-1-difficulty cases).
Case-specific variables included: trainee involvement, difficulty, ASA, sedation, inpatient/outpatient status at time of ERCP, and indications (stone, imaging abnormality, chronic pain, abnormal liver enzymes, pancreatitis, tumor ablation, and post-surgical problems). Endoscopist-level variables included: country (United States, United Kingdom (UK), others), academic/community setting, experience in years, lifetime volume, annual volume, training volume, grade-1 procedure time, and grade-1 fluoroscopy time.

Statistical analysis and power considerations
For most numerical variables (experience measures, grade-1-difficulty times), distributions were positively skewed, so they were split by median or quartile.
To account for inherent clustering (i.e., same endoscopist performing multiple procedures), a multilevel model with random intercepts was constructed. First, univariate multilevel logistic regression analyses were performed yielding adjusted cannulation success rates. Correlations between variables were evaluated using Spearman' rank correlation coefficients. Second, variables with an adjusted univariate p-value < 0.2 were entered into multivariate multilevel logistic regression. A backward stepwise approach was used, and adjusted odds ratios (OR) and corresponding 95% confidence intervals were reported. All tests were 2-sided, and p values <0.05 were considered statistically significant (SAS v9.2 (SAS Institute Inc, Cary, NC)).
With 13,018 subjects, we had high (96%) power to detect very small (2%) cannulation rate differences, for case-specific variables. For endoscopist-level factors, power was lower but still reasonable: for 84 doctors (eg. stratified by median annual volume), differences in cannulation success rates of >11% (e.g. 85% vs. 96%) could be detected with 80% power, assuming an intraclass correlation coefficient of 0.4 within endoscopists.

Results
In the Network, 13,018 ERCPs in native papillae were performed by 85 endoscopists. Qualifying endoscopists (contributing >30 cases), consecutively entered a median of 8.9 ERCP/week (range 5-27) over up to almost 3 years (median entry 80 days). Although consecutive entry was not able to be audited directly, each endoscopist's actual entered case volume was higher than their respective expected case volume for this time frame (using annual volume at baseline survey), consistent with consecutive case entry. Conventional deep biliary cannulation success rate was 89.8% (ranging 63.9%-100% for different endoscopists). Precut sphincterotomy was performed in 876 (876/13018, 6.7%) ERCPs, and deep biliary cannulation was achieved in 745 (85.1%). Overall deep biliary cannulation success rate (including precut-assisted cases) was 95.6% (ranging 80.2%-100% for different endoscopists).

Case-specific characteristics
Briefly, 6235 (47.9%) were grade 1 difficulty and 3746 (28.8%) were grade 3, 30.5% were classified as ASA III-V, and just over half (55.3%) had MAC or general anesthesia ( Table 1). The pre-ERCP status of patients was evenly split between outpatient and inpatient. Trainees were involved in 4113 (31.6%) procedures. The most common indication (36.8%) was suspected stone.

Endoscopist-specific characteristics
Most endoscopists (71%) were from the US with 19% from the UK; other countries included Canada, Australia, Brazil, Norway, and Venezuela (Table 2). About half the endoscopists did not receive formal ERCP training, and there was a broad range of lifetime experience and volumes. Surrogates of ERCPist "efficiency" included median procedure and fluoroscopy times in grade-1-difficulty cases of 25 min and 3 min, respectively.

Univariate multilevel logistic regression analysis
Five case-specific factors were significantly associated with conventional cannulation success, adjusted for doctor clustering (Table 3): trainees, difficulty, ASA, outpatient status, and indications. Of the endoscopistspecific factors, only country was significant.
Similarly, 5 case-specific factors were significantly associated with overall/ultimate success (Table 4): trainees, difficulty, ASA, sedation type, and indications. Four endoscopist-specific factors were significant: country, annual volume, and practice "efficiency" surrogates (median procedure and fluoroscopy times in grade-1difficulty cases).

Multivariate multilevel logistic regression analysis
Because of the above inter-correlated variables, lifetime volume and procedure time were dropped for multivariate  (Table 5). Relationships with trainee involvement were complex: a high degree of trainee involvement (hands-on >50% of the case) was associated with higher success (OR 1.58) than not having a trainee, whereas low levels of trainee involvement were associated with failure. Of note, none of the endoscopist-specific factors was significant. Similar factors were found to be independently associated with overall/ultimate success, including trainee involvement, comorbidity index, and certain indications (

Discussion
Deep biliary cannulation success in native papillae is a widely accepted measure of competence in ERCP during training, and quality of an endoscopist in ERCP practice. Identifying predictors for successful biliary cannulation in native papilla, both at a case-and at an endoscopist-/ team-level, have important implications in improving the quality of ERCP and patient care. Further multivariate analyses suggest that only case-specific factors are significantly associated with conventional native papilla biliary cannulation success, and that endoscopist-and institution-level factors may not be as important.
Pre-procedure evaluations considering complexity and indications are important, to weigh anticipated success rates into decision-making and consent. Prior studies correlating higher difficulty score and lower success were heterogeneous without sufficient adjustment for confounders; Verma et al. found no correlation between conventional cannulation success and procedure difficulty for trainees [21]. Our results supported a relationship, although the absolute differences seen were small. Overall "case" complexity and difficulty is determined by many factors (of which cannulation is just one), and so, does not necessarily correlate with "cannulation" difficulty. The negative randomized trials of ERCP in mild to moderate acute gallstone pancreatitis should already limit its use in active pancreatitis due to limited efficacy [22]; but pancreatitis also predicted lower success rates in our study, perhaps related to duodenal edema, providing more reason to avoid this context. Obstructive jaundice (mostly cancers) predicts lower success than in suspected stone cases; this is in keeping with a recent randomized trial advising against ERCP in obstructive jaundice from surgically resectable tumors mostly because of morbidity related to cannulation/stenting failures and rescue procedures (69%/83% success in drainage at ERCP in community/academic centers, respectively) [23]. Post-surgical biliary issues (e.g. leaks, strictures) also predict lower success; this has not been previously reported; anatomic distortion, edema, or need for atypical positioning (e.g. supine) because of surgical wounds may contribute.
Trainee involvement (modeled as yes or no) has been shown to increase post-ERCP pancreatitis [24]. The British Society of Gastroenterology (BSG) study reported trainees reduced cannulation success to 54% [12]. However, we explored the proportion of involvement, and found conventional and precut-assisted cannulation success rates decreased only with more casual trainee involvement. Trainees with brief involvement in a case may cause papilla edema, and lower the chance of the supervisor's success. Lastly, confounding could exist if more skilled endoscopists allowed more trainee handson time.
In the present study, high ASA score was surprisingly another factor predicting lower cannulation success. Tenuous sedation (requiring a more hurried procedure) or patient positioning (e.g. due to obesity) may hinder cannulation in some way. Our results suggest that outpatient ERCP, adjusted for other factors, may have a higher success rate, even though adverse event rates may be comparable [25,26]. Previous reports have found higher technical success rates achieved under deep sedation and general anesthesia than moderate ("conscious") sedation because of better patient tolerance and compliance [27,28]. However, our results showed that deeper sedation only predicted success if precut was allowed; this may be explained by an improved ability to use advanced rescue  techniques with deeper sedation. Country predicted success in univariate analysis, but when corrected for differences in sedation use and other factors, it did not; the univariate association is likely confounded by international sedation practice differences. ASGE, based on learning curves, recommends at least 180-200 cases (at least 50% therapeutic) to achieve competency in cannulation [9]. A recent study, however, found that at least 350 supervised procedures were needed for an 80% native papilla biliary cannulation success rate [21]. In the ERCP network, 69.5% endoscopists received <200 procedures during their training (54.2% less than 100), comparable to a recent survey: 60.4% responders completed <180 training cases [29]. Training volumes did not predict cannulation success in our cohort. However, highvolume-trained endoscopists (>500) were underrepresented; higher-volume training might have impact. Another consideration is that endoscopists with <200 cases in training tended to have higher years performing  ERCP (data not shown); as such, their learning curve may have already risen and plateaued in practice, minimizing the apparent impact of their lower volume training. Many believe that after proper training, experience (volume, years) and annual volume contribute to outcomes, but consensus on relative importance of annual volume vs cumulative experience, and on recredentialing volume thresholds, is lacking. The British Joint Advisory Group (JAG) recommends that endoscopists should perform >75 ERCPs/year [30]. Both an American [11] and Austrian study showed that >50 ERCPs/year had higher cannulation  ERCPs/year) demonstrated no associations between success and volume or experience [8]. Weaknesses of the above mentioned studies include the lack of separating out cannulation success with and without precut, or native vs cut/stented papillae. Our data showed that higher annual volume (using quartiles) had a small but nonsignificant trend toward higher conventional success and a significant trend toward overall (precut-allowed) success (>239 ERCP/ yr: OR 2.79). Perhaps some of the higher overall success of the more active endoscopist might be due to their ability to comfortably use a more advanced rescue technique like precut sphincterotomy. We explored other cutoffs for annual volume (data not shown), including 50 as others have suggested, and 100; neither was significant. Fluoroscopy time can be influenced by several procedural factors, as well as endoscopist and X-ray technician experience, trainee involvement, and equipment quality [32][33][34][35]. Being a radiation-efficient endoscopist (averaging ≤ 3 min use in grade-1 cases) predicted overall success, suggesting quality in one aspect of practice might be associated with quality in another, which is a novel concept.
There are limitations with our study. First, as stated above, we have lower power to detect the effects of some doctor-level factors on biliary cannulation success due to the modest number of endoscopists. Second, the selfreported data could not be audited for accuracy; however the anonymous design should have reduced bias. Despite this, we acknowledge there could have been some selective reporting of more successful procedures; the number of procedures entered appeared to be similar to the number expected for each provider (based on a priori reported volumes), so we hope this bias was minimal. In addition, the generalizability of our conclusions may be limited by the fact that the volunteering endoscopists may not reflect average ERCP practice worldwide; however, this more pertains to the overall success rates, and less likely to affect the generalizability of the predictors themselves. Fortunately, the spectrum of training, volume, years in practice, practice settings, and success rates are comparable to that of other studies, and does not suggest a homogenous, highly skilled cohort of tertiary clinicians.

Conclusions
In conclusion, our results based on this unique international dataset indicate that case-specific factors may have greater impact on the most important ERCP quality metric (biliary cannulation success), than endoscopistspecific ones. Annual volume and sedation practices may influence ultimate success when precut-assisted success is not considered a failure. With regard to endoscopist experience factors, annual volume appears to perhaps be more important than prior experience, and the ideal volume (>200/year) may be considerably higher than the 50/year previously published. Further study, with an even larger number of endoscopists, could further explore the minimum annual volume for maintenance of ERCP competence. Competing interests PBC and JR have done consulting for Olympus, who sponsored the registry, but otherwise, there are no potential conflicts of interest.
Authors' contributions CYP participated in study design, acquisition of data, statistical analysis, interpretation of data, manuscript draft, and manuscript revision. PJN obtained funding and was involved in the statistical analysis and interpretation of data. PBC participated in study design, acquisition of data, and critical revision of the manuscript. DTL was involved in study design and revision of the manuscript. JR was responsible for study design, acquisition of data, statistical analysis, and critical revision of the manuscript. All authors read and approved the final manuscript.