581 Predicting All-Cause 30-Day Mortality Following First ERCP: External Validation of a Model in a Swedish Population-Based First ERCP Cohort

Abstract

among high-risk patients. However, it is unknown if pancreatic duct stent type, diameter, and presence of flanges affects the rate of post-ERCP pancreatitis. Since there have been few randomized controlled trials (RCT) evaluating PD stents, network meta-analysis can be used to evaluate the existing RCTs to determine if any PD stent features affect rates of PEP among high-risk patients. Aim: To perform a network meta-analysis of RCTs evaluating pancreatic stents for the prevention of PEP in high-risk patients. Methods: The MEDLINE, EMBASE and Cochrane Library databases were searched for RCTs evaluating the efficacy of PD stents in preventing post-ERCP pancreatitis from 9/93-9/12. Trials which reported the incidence of PEP in high-risk patients randomized to one versus another type of PD stent or no stent were included in the analysis. RCTs that did not compare PD stents or compared pharmacologic agents to stents were excluded. The predictive probability of ranking best was calculated for all the different stents. The network meta-analysis was performed using Winbugs statistical software. Results: A total of 987 studies were reviewed and 6 RCTs with 561 patients were included in this analysis. Four RCTs compared PD stents with no stents and 2 RCTs compared different types of PD stents head to head. Three RCTs evaluated 5 Fr straight PD stents, 2 RCTs evaluated 5 Fr single pigtail PD stents, and 3 RCTs evaluated 3 Fr single pigtail PD stents (see figure 1). The probability of ranking best was 0.503 for 5Fr single pigtail unflanged PD stent, 0.465 for 5Fr straight flanged PD stent, and 0.031 for 3Fr single pigtail unflanged PD stents (see figure 2). Conclusion: 5 Fr PD stents are superior to 3 Fr PD stents in preventing PEP among high-risk patients. 5 Fr single pigtail, unflanged PD stents and 5 Fr straight, flanged PD stents performed similarly but both performed better than the 3 Fr PD stents in preventing PEP, suggesting that stent diameter is more critical for the prevention of PEP than type of stent or the presence of flanges. Further studies are needed to verify these results. 581 Predicting All-Cause 30-Day Mortality Following First ERCP: External Validation of a Model in a Swedish Population-Based First ERCP Cohort Evangelos Kalaitzakis* Department of Gastroenterology, Skane University Hospital, Lund, Sweden Identifying predictors of 30-day mortality following endoscopic retrograde cholangiopancreatography(ERCP) could facilitate patient selection and the informed consent process. Recently, a 30-day mortality risk estimator following first ERCP was developed based on administrative data from England & Wales (Bodger et al GIE 2011;74:825-33). The estimator incorporates age, underlying cancer, comorbidity, and mode of admission (urgent/elective), and produces the probability of 30-day mortality. The aim of the current study was to externally validate the estimator in a population-based cohort of patients undergoing a first ERCP in Sweden. Methods All patients having had a first in-/out-patient ERCP between 04/2005-06/2008 in Sweden were identified from the Hospital Discharge Register. Data on indication, admission method(urgent/elective), and co-morbid illness were extracted from diagnostic fields(ICD-10 codes) in the register. Patients were linked to the National Death Register to define dead/alive status and the National Cancer Register. The probability of 30-day mortality of each patient according to the published estimator was calculated. Discriminant validity was estimated with the c-statistic, and goodness-to fit with the calibration plot between predicted(estimator) and the observed death probabilities and the Hosmer-Lemeshow test. Results A total of 16478 patients with a first ERCP performed in 66 hospitals were identified (mean age 68(17); 57% female; urgent admission 57%; diagnostic profile: gallstone-related 55.2%, cancer 22.5%, gallstone and cancer 1.4%, other 20.9%). All-cause 30-day mortality was 5%(14.9% in cancer and 1.9% in non-cancer cases). No major differences were observed in case mix or mortality between the original(English) and the current(Swedish) cohorts, apart from a higher proportion of cancer cases in the latter (14.6% vs23.9%). Discriminant validity was very good with a c-statistic of 0.82 (95% CI 0.81-0.83) in the Swedish cohort. However, the Hosmer-Lemeshow test was significant(p 0.001) indicating lack of fit. The calibration plot suggested that the model tended to overestimate mortality(figure 1). Re-calibration was performed with slope adjustment(figure 2) without compromising discriminant validity (c-statistic of new model 0.81, 95% CI 0.80-0.82). However, the HosmerLemeshow test continued to be significant(p 0.001) indicating lack of fit (figure 2). Conclusion A model of 30-day mortality following first ERCP recently developed from administrative data in England and Wales was validated in a Swedish population-based cohort of patients with a first ERCP. Despite adequate discriminant validity, there was lack of fit which remained after re-calibration with slope adjustment. These findings indicate that different factors may be of importance for 30-day mortality prediction following first ERCP in different populations.