On-site ERCP availability and cholangitis outcomes: Retrospective cohort study

Keywords

Pancreatobiliary (ERCP/PTCD) - ERC topics - Quality and logistical aspects - Quality management - Epidemiology

Background

Endoscopic retrograde cholangiopancreatography (ERCP) is a crucial component in the management of acute cholangitis (AC) [1], which carries mortality rates of up to 10% and 50% with and without treatment, respectively [2] [3] [4]. Meta-analyses and large observational studies support use of early ERCP for biliary decompression to improve patient outcomes [5] [6] [7] [8]. Unlike upper and lower endoscopy, which are widely available at most hospitals, ERCP is offered in certain centers. Patients with AC, however, may present to any hospital. The potential impact of on-site ERCP availability on outcomes in cholangitis is unknown.

Limited availability of ERCP is partially due to a lack of personnel, because endoscopists and nurses require specialized training beyond the skills needed for diagnostic endoscopy. In addition, ERCP is a higher-risk procedure with up to 10% of patients experiencing serious adverse events (AEs) such as pancreatitis, bleeding, infection, or perforation [9]. In Ontario, Canada, ERCP availability is limited to sites where endoscopists perform a larger volume of procedures, a factor associated with fewer AEs and lower procedure failure rates [10] [11]. Currently there is no standardized referral stream for ERCP. When patients present to a hospital without ERCP services, they are either managed conservatively with antibiotics and observation, undergo percutaneous drainage locally, or are transferred to an ERCP site.

Given this evidence gap regarding clinical outcomes for AC, we conducted a retrospective cohort study to compare patients admitted with AC to hospitals with ERCP services (henceforth referred to as ERCP sites) with those admitted to hospitals without ERCP services (henceforth referred to as non-ERCP sites).

Patients and methods

This retrospective cohort study included patients with AC at 27 hospitals in Ontario, Canada that participate in GEMINI [12].

This study received research ethics approval from St. Michael’s Hospital on behalf of all participating hospitals through the Clinical Trials Ontario platform, with a waiver of patient consent due to the use of routinely collected data. We reported this study according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [13].

All diagnosis and procedure codes are available in Supplementary Table 1, Supplementary Table 2, and Supplementary Table 3.

Data source

GEMINI is a hospital research collaborative that collects administrative and clinical data from hospital information systems with 98% to 100% accuracy of selected data elements when compared with manual chart review [14]. GEMINI includes adult admissions to general internal medicine and subspecialty medical wards (cardiology, gastroenterology, hematology, medical oncology, respirology). GEMINI also includes all patients who experienced an intensive care unit (ICU) admission at any point during their hospitalization, including patients who were admitted to surgical wards but required transfer to and/or from an ICU [14].

Patient-level administrative data from individual hospitals are derived from GEMINI as reported to the Canadian Institute for Health Information (CIHI) National Ambulatory Care Reporting System and Discharge Abstract Database [15] [16]. These data include patient demographics, admission and discharge diagnoses, interventions (endoscopic, radiological, surgical), mortality. In addition, laboratory and radiology tests are extracted directly from hospital electronic health record systems into GEMINI. GEMINI data have previously been used to explore provision of endoscopic procedures for management of gastrointestinal bleeding during the COVID-19 pandemic [17]. The participating hospitals serve diverse, multiethnic urban and suburban populations through universal single-payer coverage in Ontario [12].

Cohort

We included adult patients (age ≥ 18) admitted to inpatient general internal medicine and subspecialty medical wards or ICUs at 27 hospitals in Ontario, Canada from April 1, 2015 to December 31, 2021 whose most responsible discharge diagnosis was AC based on International Classification of Diseases version 10 (ICD-10) (Supplementary Table 1). For individuals whose most responsible discharge diagnosis was “other and unspecified cholangitis,” additional Tokyo criteria [1] for suspected AC were required for inclusion in this study.

We did not include patients who were admitted with AC to a surgical ward and had no ICU touchpoint during their hospitalization because these patients are not captured in GEMINI.

Exposure

The primary exposure was on-site ERCP availability. For each individual hospital, we used three independent methods to characterize participating hospitals as ERCP sites or non-ERCP sites. First, the study authors (JDM, SCG) are practicing endoscopists at an ERCP referral center (St. Michael’s Hospital) and have institutional knowledge of the availability of ERCP across Ontario hospitals. Second, we searched the web pages of individual hospital endoscopy units to identify which endoscopic services are provided. Third, we contacted gastroenterology division heads at each participating hospital to confirm ERCP availability or lack thereof.

With the above information, we grouped patients as having been admitted to an ERCP site or non-ERCP site. If there was any conflict among the three different methods above, we used the information from the gastroenterology division head to classify the respective site.

Two hospitals began providing ERCP services during the study period (Toronto General Hospital, Toronto Western Hospital). We classified patients admitted to these hospitals prior to the ERCP start date as admitted to a non-ERCP and those admitted afterwards as admitted to an ERCP site.

Outcomes

The primary outcome was in-hospital mortality for each episode of care. We defined an episode of care as initial presentation to a hospital with AC and all subsequent continuous transfers between acute care hospitals. Consistent with CIHI definitions, we defined inter-facility transfers as new admissions within 7 hours of discharge or new admissions within 12 hours of discharge if one hospital had coded the transfer. This unit of analysis captured hospital stays that involve inter-hospital transfers for provision of ERCP services. Secondary outcomes were length of stay in hospital, 7- and 30-day readmission rates, ICU admission, and requirement for percutaneous or surgical biliary decompression. Intervention codes for ERCP and percutaneous and biliary decompression are available in Supplementary Table 2 and Supplementary Table 3. Readmissions were captured if they were to a medical or ICU service at a GEMINI hospital. Readmissions coded as elective are not counted as readmissions, and episodes of care ending in death are excluded from readmission analyses because they are not eligible for readmission.

We did not include post-ERCP AE rates of bleeding, perforation, infection, and pancreatitis. We lacked the granular patient-level data required to adjudicate these outcomes, based on a causal attribution system for post-ERCP AEs [18], using the administrative data available in GEMINI, and thus excluded them a priori.

Covariates

We included the following patient-level characteristics as covariates: age, sex, Charlson Comorbidity Index Score at admission [19], use of antibiotics, and presence of underlying biliary stricture or pancreaticobiliary malignancy, which included primary sclerosing cholangitis, chronic pancreatitis, and cancer of the ampulla, bile duct, pancreas, and duodenum (Supplementary Table 2). We also included severity of initial presentation based on Tokyo criteria [1] and provision of antibiotic therapy. We identified antibiotic therapy using an established iterative approach between the GEMINI-RxNorm algorithm and clinical subject matter expert [20]. We considered patients as having severe cholangitis if they had serum creatinine > 176 μmol/L, international normalized ratio (INR) > 1.5, serum platelet count of < 120,000/mm³, or a requirement for intravenous vasopressors, noninvasive ventilation, or invasive ventilation [1].

Analysis

We summarized patient characteristics descriptively using counts with percentages or medians with interquartile ranges (IQRs) where appropriate.

We used logistic regression to estimate the adjusted association between being an ERCP site (compared with a non-ERCP site) and in-hospital mortality, ICU admission, requirement for intervention, and 7- and 30-day readmission. We used negative binomial regression to estimate the adjusted association between ERCP site and length of in-hospital stay. We presented differences in time-to-ERCP and time-to-death by ERCP site using cumulative incidence curves using the Kaplan-Meier method. We used Cox proportional hazards regression to estimate the adjusted association between admission to an ERCP site and time-to-ERCP. Time-to-event analyses censored patients at discharge or death.

Estimates from logistic regression were presented as odds ratios, negative binomial regression as rate ratios (RRs), and Cox regression as hazard ratios (HRs). All models were adjusted for age, sex, presence of disease/malignancy, requirement for antibiotics, and Charlson Comorbidity Index Score. We reported adjusted estimates and 95% confidence intervals (Cis) using hospital-level cluster-robust standard errors to account for the fact that patients are nested within hospitals.

Logistic regression models were fit with the rms package (v6.4–1), negative binomial regressions using the MASS package (v7.3–58.3), cumulative incidence curves using the survminer package (v0.4.9), and Cox regression using the survival package (v3.5–7) [21] [22] [23] [24]. All analyses were completed in R version 4.1.2 [25].

We planned three a priori subgroup analyses based on clinical factors that could plausibly impact the management and/or outcomes of patients with AC:

1. Patients with severe cholangitis, defined using the Tokyo criteria above [1].

2. Patients with underlying biliary stricture or pancreaticobiliary malignancy [26].

3. Patients admitted to hospital on the weekend (from Friday at 5 pm to Monday at 8 am) [27].

We performed post-hoc analyses, using a propensity score-based method to balance baseline patient covariates to account for potential selection bias with: a) the entire cohort; b) only patients who underwent ERCP; and c) only patients admitted to the ICU. The propensity score was calculated using a logistic regression model to calculate the propensity of a patient presenting to a hospital with on-site ERCP, based on the servicer covariates of age, sex, admission Charlson Comorbidity Index Score, presence of severe AC, and underlying biliary stricture or pancreaticobiliary malignancy. Propensity scores were then balanced across patient exposure groups using overlap weighting. Outcomes were then compared in the weighted populations, including in-hospital mortality, length of stay in hospital, 7- and 30-day readmission rates, and ICU admission. We report the effect of admission to an ERCP site as a risk difference (RD), where a positive RD indicates greater risk for those admitted to an ERCP site. Overlap weight modeling was performed using PSweight package in R (version 1.2.0) [28].

Results

Our cohort included 4492 patients, with 3867 (86.1%) at ERCP sites and 625 (13.9%) at non-ERCP sites. Median age was 75 years (IQR 62–84) and 2084 patients (46%) were female. Underlying biliary stricture or pancreaticobiliary malignancy was present in 688 patients (15%) and 822 (18%) had severe AC ([Table 1]).

Table 1 Patient characteristics.

Variable	Admission to ERCP center (N = 3867)	Admission to non-ERCP center (N = 625)
ICU, intensive care unit.
Median age [Q1 - Q3]	74.0 [61.0–84.0]	78.0 [65.0–87.0]
Median Charlson comorbidity score [Q1 - Q3]	0 [0–2.00]	0 [0–2.00]
Female sex, n (%)	1789 (46 %)	295 (47 %)
Severe AC, n (%)	703 (18 %)	119 (19 %)
Weekend admission, n (%)	1090 (28 %)	169 (27 %)
Presence of underlying biliary stricture or pancreaticobiliary malignancy, n (%)	585 (15 %)	103 (16 %)
ERCP performed < 24 h, n (%)	713 (18.4%)	23 (3.7%)
ERCP performed in 24–48 h, n (%)	539 (13.9%)	76 (12.2%)
ERCP performed > 48 h, n (%)	903 (23.4%)	156 (25.0%)
ERCP not performed, n (%)	1712 (44.3%)	370 (59.2%)

ERCP was performed for 713 patients (18.4%) at ERCP sites and 23 patients (3.7%) at non-ERCP sites in less than 24 hours, for 539 patients (13.9%) and 76 patients (12.2%) respectively within 24 to 48 hours, and for 903 patients (23.4%) and 156 patients (25.0%), respectively, in greater than 48 hours. ERCP was not performed during the index hospitalization for 1712 patients (44.3%) and 370 patients (59.2%) at ERCP sites and non-ERCP sites, respectively, ([Table 1], [Fig. 1]). Patients at ERCP sites were more likely to receive ERCP earlier than patients at non-ERCP sites (adjusted (a) HR = 1.76, 95% CI 1.48–2.10). Cumulative incidence estimates are presented in [Fig. 2].

Fig. 1 Percentage of patients undergoing ERCP prior to 24 hours, within 24 hours to 48 hours, after 48 hours, and not at all.

Fig. 2 Cumulative incidence curves of time-to-ERCP (solid lines) and time-to-death (dashed lines) using the Kaplan-Meier method, with 0 = non-ERCP sites and 1 = ERCP sites. These time-to event analyses censored patients at discharge or death.

In-hospital mortality

Overall, 197 patients (4%) died in hospital during the episode of care. Patients at ERCP sites did not have significantly different in-hospital mortality rates than patients at non-ERCP sites (unadjusted mortality 4% vs. 4%; aOR 2.19, 95% CI 0.86–5.55) ([Table 2], [Table 3], [Fig. 3]).

Table 2 Unadjusted estimates for primary and secondary outcomes.

Outcome	Admission to ERCP center (N = 3867)	Admission to non-ERCP center (N = 625)
ICU, intensive care unit.
In-hospital mortality, n (%)	173 (4 %)	24 (4 %)
ICU admission, n (%)	441 (11 %)	56 (9 %)
7-day readmission, n (%)	164 (4 %)	31 (5 %)
30-day readmission, n (5)	485 (13 %)	84 (13 %)
Length of stay (days) [Q1 - Q3]	5.22 [3.31–8.99]	6.10 [3.87–9.50]
Percutaneous Intervention, n (%)	290 (7 %)	35 (6 %)
Surgical Intervention, n (%)	30 (1 %)	1 (0.1%)

Table 3 Adjusted primary and secondary outcomes for main and subgroup analyses.

Outcome	Unadjusted estimate	Adjusted estimate
aOR, adjusted odds ratio; aRR, adjusted rate ratio; CI, confidence interval; ICU, intensive care unit. *Significant at P < 0.05. Surgical intervention adjusted estimate not presented because we did not perform adjusted analyses for that outcome.
In-hospital mortality, aOR (95% CI)	1.17 (0.73–1.8)	2.19 (0.8–5.55)
ICU admission, aOR (95% CI)	1.31 (0.86–1.98)	1.96 (1.29–2.98)*
7-day readmission, aOR (95% CI)	0.83 (0.62–1.10)	0.77 (0.56–1.08)
30-day readmission, aOR (95% CI)	0.90 (0.61–1.32)	0.89 (0.59–1.35)
Median length of stay, aRR (95% CI)	0.82 (0.58–1.15)	0.95 (0.81–1.12)
Percutaneous intervention, aOR (95% CI)	1.37 (0.58–3.24)	1.22 (0.53–2.81)
Surgical intervention, aOR (95% CI)	4.71 (0.93–23.84)	–

Fig. 3 Forest plot of adjusted estimates for primary and secondary outcomes.

Compared with non-ERCP sites, patients at ERCP sites had higher mortality rates within the subgroups of severe cholangitis (aOR 2.17, 95% CI 1.17–4.02) and underlying biliary stricture or pancreaticobiliary malignancy (aOR 1.94, 95% CI 1.14–13.58). We observed no difference for in-hospital mortality between ERCP sites and non-ERCP sites within the subgroups of patients admitted on the weekend ([Table 4]).

Table 4 Adjusted estimates for primary and secondary outcomes among subgroups.

Outcome	Severe cholangitis	Weekend admission	Underlying biliary stricture or pancreaticobiliary malignancy
aOR, adjusted odds ratio; aRR, adjusted rate ratio; CI, confidence interval; ICU, intensive care unit. *Significant at P < 0.05. Surgical intervention estimates not presented because we did not perform adjusted analyses for that outcome.
In-hospital mortality, aOR (95% CI)	2.17 (1.17–4.02)*	6.51 (0.60–70.19)	3.94 (1.14–13.58)*
ICU admission, aOR (95% CI)	3.15 (1.90–5.21)*	3.15 (1.49–6.64)*	1.79 (1.01–3.16)*
7-day readmission, aOR (95% CI)	0.74 (0.42–1.30)	0.66 (0.38–1.13)	0.89 (0.57–1.37)
30-day readmission, aOR (95% CI)	0.67 (0.35–1.29)	1.07 (0.67–1.72)	0.88 (0.55–1.41)
Median length of stay, aRR (95% CI)	0.55 (0.40–0.76)*	1.15 (0.98–1.36)	1.32 (1.09–1.59)*
Percutaneous Intervention, aOR (95% CI)	1.23 (0.58–2.61)	3.11 (0.58–16.63)	0.82 (0.37–1.80)

Length of stay and readmission

We observed no difference between patients at ERCP sites versus non-ERCP sites for hospital length of stay (median 5.22 days [IQR 3.31–8.99] vs 6.10 days [IQR 3.87–9.50]; adjusted RR 0.95, 95% CI 0.81–1.12), 7-day readmission rates, (4% vs 5%; aOR 0.77, 95% CI 0.56–1.08), or 30-day readmission rates (13% vs 13%; aOR 0.89, 95% CI 0.59–1.35) ([Table 2], [Table 3], [Fig. 3]).

Patients at ERCP sites with severe cholangitis had shorter LOS (aRR 0.55, 95% CI 0.40–0.76), and those with underlying biliary stricture or pancreaticobiliary malignancy had longer LOS (aRR 1.32, 95% CI 1.09–1.59). We observed no significant difference between sites for other subgroups for the outcomes of length of stay or 7- or 30-day readmission ([Table 4]).

Hospital resource utilization

Patients at ERCP sites were more likely to experience an ICU admission (11% vs 9%; aOR 1.96, 95% CI 1.29–2.98). We observed no difference between patients at ERCP sites versus non-ERCP sites for percutaneous intervention rates (7% vs 6%; aOR 1.22, 95% CI 0.53–2.81). For the outcome of surgical intervention, we could not perform an adjusted analysis due to a small number of events. Thirty patients (1%) at ERCP sites and one patient at a non-ERCP site (0.1%) underwent surgical intervention ([Table 2], [Table 3], [Fig. 3]).

Patients at ERCP sites with severe cholangitis (aOR 3.15, 95% CI 1.90–5.21), underlying biliary stricture or pancreaticobiliary malignancy, (aOR 1.79, 95% CI 1.01–3.16) and those admitted on the weekend (aOR 3.15, 95% CI 1.49–6.64) had higher odds of ICU admission. We observed no difference between ERCP sites and non-ERCP sites for receipt of percutaneous intervention among any predefined subgroup ([Table 4]).

Propensity score with overlapping weights analysis

The cohorts were perfectly balanced after overlap weighting, by definition (Supplementary Table 4). There was no significant difference between patients at ERCP sites compared with those at non-ERCP sites for in-hospital mortality (RD 0.09, 95% CI -0.12 to 0.03), LOS (RD 0.4, 95% CI -1.27 to 2.19), 7-day readmission (RD -0.02, 95% CI -0.07 to -0.13), 30-day readmission (RD -4.76, 95% CI -0.25 to 0.05), or ICU admission (RD 3.15, 95% CI -0.06 to 0.08) (Supplementary Table 5). When including only patients who underwent ERCP, there were no differences between groups for in-hospital mortality, LOS, 7- and 30-day readmission, or ICU admission. When including only patients who were admitted to the ICU, there were no differences between groups for in-hospital mortality, LOS, or 7-day readmission. Patients at ERCP sites had a lower risk of 30-day readmission (RD 16.09, 95% CI -0.25 to 0.01), although the event rate for patients at non-ERCP sites was < 6 (Supplementary Table 5).

Discussion

In this retrospective analysis of patients admitted with AC at 27 large urban hospitals, we observed no difference in mortality between patients admitted to ERCP sites compared with non-ERCP sites. Patients at ERCP sites were more likely to experience an ICU admission. Subgroups of patients with underlying biliary stricture or pancreaticobiliary malignancy and severe cholangitis had higher mortality at ERCP sites compared with non-ERCP sites. To our knowledge, this is the first study to examine the effect of on-site ERCP availability on cholangitis outcomes.

Management of cholangitis includes antibiotics, supportive treatment such as fluids or vasopressors, and biliary decompression [2]. Although medications and supportive treatment are provided at all hospitals, biliary decompression often requires endoscopic or percutaneous drainage and is not available universally. In Ontario, patients can be transferred between hospitals when they require advanced endoscopic services that are unavailable at their presenting hospital. In our study, patients had higher absolute rates of undergoing ERCP at ERCP sites and were more likely to undergo it sooner than patients at non-ERCP sites. Despite this, we found no difference for in-hospital mortality overall.

Several secondary results warrant further discussion. First, we found increased mortality among patients with severe cholangitis and underlying biliary stricture or pancreaticobiliary malignancy at ERCP sites. Although we anticipated that on-site ERCP availability and provider experience with cholangitis at ERCP sites would be associated with improved outcomes, it is possible that there are unmeasured clinical factors that explain higher in-hospital mortality. Indeed, patients at ERCP sites experienced higher rates of ICU admission. There are several possible underlying reasons. There may be more patient complexity and associated morbidity at ERCP sites. In addition, provision of an invasive procedure such as ERCP may lead to ICU admission. Patients at ERCP sites may also undergo ICU admission to facilitate emergent ERCP after hours. These potential clinical pathways warrant further study.

In addition, approximately 44% of patients at ERCP sites and 59% at non-ERCP sites with a diagnosis of AC did not undergo ERCP during the index hospitalization. This is in keeping with other population-based studies of ERCP in cholangitis [27] [29]. There are several potential reasons for this finding. First, some patients may respond to antibiotics and supportive care before procedural decompression is performed. If there is adequate clinical response, providers may decide that risk of an ERCP outweighs the benefit. In addition, some patients may go on to be discharged and have ERCP at a later time as outpatients. Third, invasive procedures may not be within certain patients’ goals of care. Finally, patients may have been misclassified as having AC, and thus, would not have warranted biliary drainage. Further work to validate diagnostic codes for cholangitis in administrative data sets is warranted.

There are several limitations in our present study. First, we did not capture patients who are admitted to surgical services with cholangitis or those from rural hospitals outside the GEMINI network. Second, we did not capture delayed AEs, such as post-ERCP pancreatitis, due to the limitations of administrative data. These AEs may have contributed to worse outcomes for patients at ERCP sites for certain subgroups and warrant investigation in a prospective study. Third, accuracy of diagnosis and interventions in an administrative data set are dependent upon the accuracy of initial coding. Fourth, we were unable to capture patients who underwent ERCP as outpatients after their index hospitalization. Fifth, our analyses of time-to-ERCP do not consider the competing risk of in-hospital death; mortality rates in our cohort were very low and this does not meaningfully impact results. Finally, we were unable to account for factors that affect ERCP performance, such as hospital and provider-level volume [10] [11]. Our study also has several strengths. We used a comprehensive inpatient medical database that captures variables known to affect cholangitis outcomes. In addition, we used objective and relevant outcomes such as mortality, intensive care utilization, and readmission.

Conclusions

In this cohort study of 4492 patients at 27 large urban hospitals, on-site ERCP availability did not impact in-hospital mortality rates. Compared with non-ERCP sites, patients at ERCP sites with severe cholangitis and underlying biliary stricture or pancreaticobiliary malignancy had higher in-hospital mortality rates, which warrants further study for potential unmeasured variables that are associated with adverse outcomes.

• References

• 1 Kiriyama S, Kozaka K, Takada T. et al. Tokyo Guidelines 2018: diagnostic criteria and severity grading of acute cholangitis (with videos). J Hepatobiliary Pancreat Sci 2018; 25: 17-30
• 2 Lee JG. Diagnosis and management of acute cholangitis. Nat Rev Gastroenterol Hepatol 2009; 6: 533-541
• 3 Mosler P. Diagnosis and management of acute cholangitis. Curr Gastroenterol Rep 2011; 13: 166-172
• 4 Lee F, Ohanian E, Rheem J. et al. Delayed endoscopic retrograde cholangiopancreatography is associated with persistent organ failure in hospitalised patients with acute cholangitis. Aliment Pharmacol Ther 2015; 42: 212-220
• 5 Tan M, Schaffalitzky de Muckadell OB, Laursen SB. Association between early ERCP and mortality in patients with acute cholangitis. Gastrointest Endosc 2018; 87: 185-192
• 6 Iqbal U, Khara HS, Hu Y. et al. Emergent versus urgent ERCP in acute cholangitis: a systematic review and meta-analysis. Gastrointest Endosc 2020; 91: 753-760 e4
• 7 Hou LA, Laine L, Motamedi N. et al. Optimal timing of endoscopic retrograde cholangiopancreatography in acute cholangitis. J Clin Gastroenterol 2017; 51: 534-538
• 8 Parikh MP, Wadhwa V, Thota PN. et al. Outcomes associated with timing of ERCP in acute cholangitis secondary to choledocholithiasis. J Clin Gastroenterol 2018; 52: e97-e102
• 9 Andriulli A, Loperfido S, Napolitano G. et al. Incidence rates of post-ERCP complications: a systematic survey of prospective studies. Am J Gastroenterol 2007; 102: 1781-1788
• 10 Varadarajulu S, Kilgore ML, Wilcox CM. et al. Relationship among hospital ERCP volume, length of stay, and technical outcomes. Gastrointest Endosc 2006; 64: 338-347
• 11 Coté GA, Imler TD, Xu H. et al. Lower provider volume is associated with higher failure rates for endoscopic retrograde cholangiopancreatography. Med Care 2013; 51
• 12 Verma AA, Guo Y, Kwan JL. et al. Patient characteristics, resource use and outcomes associated with general internal medicine hospital care: the General Medicine Inpatient Initiative (GEMINI) retrospective cohort study. CMAJ open 2017; 5: E842
• 13 Von Elm E, Altman DG, Egger M. et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007; 370: 1453-1457
• 14 Verma AA, Pasricha SV, Jung HY. et al. Assessing the quality of clinical and administrative data extracted from hospitals: the General Medicine Inpatient Initiative (GEMINI) experience. J Am Med Inform Assoc 2021; 28: 578-587
• 15 Canadian Institute for Health Information. Discharge abstract data metadata. https://www.cihi.ca/en/discharge-abstract-database-metadata-dad
• 16 Canadian Institute for Health Information. National Ambulatory Care Reporting System metadata (NACRS). https://www.cihi.ca/en/national-ambulatory-care-reporting-system-metadata-nacrs
• 17 Khan R, Saha S, Gimpaya N. et al. Outcomes for upper gastrointestinal bleeding during the first wave of the COVID-19 pandemic in the Toronto area. J Gastroenterol Hepatol 2022; 37: 878-882
• 18 Forbes N, Elmunzer BJ, Keswani RN. et al. Consensus-based development of a causal attribution system for post-ERCPadverse events. Gut 2022; 71: 1963-1966
• 19 Charlson M, Szatrowski TP, Peterson J. et al. Validation of a combined comorbidity index. J Clin Epidemiol 1994; 47: 1245-1251
• 20 Waters R, Malecki S, Lail S. et al. Automated identification of unstandardized medication data: a scalable and flexible data standardization pipeline using RxNorm on GEMINI multicenter hospital data. JAMIA open 2023; 6: ooad062
• 21 Therneau T. A package for survival analysis in R. R package version 3.5–7. 2023.
• 22 Harrell F. Regression modeling strategies. R package version 6.4–1. 2023.
• 23 Venables W, Ripley B. Modern applied statistics with S. Fourth Edition. New York, NY, United States: Springer; 2002
• 24 Kassambara A, Kosinski M, Biecek P. et al. Drawing survival curves using ‘ggplot2’. R package version 0.4.9. 2021.
• 25 R Core Team. R: A language and environment for statistical computing. 2010. https://lib.stat.cmu.edu/R/CRAN/doc/manuals/r-devel/fullrefman.pdf
• 26 Acehan F, Çamlı H, Kalkan C. et al. Characteristics and clinical outcomes of acute cholangitis in older patients. Eur Geriatr Med 2023; 14: 263-273
• 27 Inamdar S, Sejpal DV, Ullah M. et al. Weekend vs. weekday admissions for cholangitis requiring an ERCP: Comparison of outcomes in a national cohort. Am J Gastroenterol 2016; 111: 405-410
• 28 Li F, Thomas LE, Li F. Addressing extreme propensity scores via the overlap weights. Am J Epidemiol 2019; 188: 250-257
• 29 Wang M, Wadhwani SI, Cullaro G. et al. Racial and ethnic disparities among patients hospitalized for acute cholangitis in the United States. J Clin Gastroenterol 2023; 57: 731-736