Clinical outcomes following endoscopic retrograde cholangiopancreatography-based therapy compared with surgery in chronic pancreatitis: evidence from a multicentric cohort

Introduction

Chronic pancreatitis (CP) is a progressive fibroinflammatory disorder characterized by irreversible pancreatic injury, recurrent or persistent abdominal pain, and progressive exocrine and endocrine dysfunction [1]. It imposes a substantial clinical burden, with reported prevalence ranging from 13.5-52.4 per 100,000 persons and an incidence of approximately 8.6 per 100,000 person-years [2-4]. Disease expression is heterogeneous, reflecting variable contributions from environmental, genetic, obstructive and recurrent inflammatory factors [5].

Pain is the dominant driver of morbidity in CP and contributes substantially to impaired quality of life, recurrent healthcare utilization and opioid exposure [1,2,5]. Progressive pancreatic injury may also result in exocrine pancreatic insufficiency (EPI) and diabetes during the disease course [5-7]. Management, therefore, requires treatment of pain and ductal obstruction, nutritional support, screening for pancreatic functional decline, and attention to opioid stewardship throughout follow up.

For symptomatic obstructive CP, endoscopic therapy is commonly used as a less invasive strategy, particularly when pancreatic duct drainage, stent therapy, stone management or sphincter intervention is considered appropriate [1,2]. Surgical options, including drainage and resection-based approaches, are generally considered for refractory pain, complications, or disease anatomy less likely to respond durably to endoscopic therapy [5,8]. Randomized evidence in selected patients with painful obstructive CP and a dilated pancreatic duct has suggested better pain outcomes with early surgery than with an endoscopy-first strategy [9].

However, randomized trials include selected populations and may not fully represent routine clinical practice, where treatment decisions are influenced by comorbidity, disease phenotype, local expertise, surgical candidacy and patient preference. Real-world comparative analyses can complement trial data by describing outcomes across broader populations, while remaining limited by residual confounding and coding-based outcome ascertainment [10]. Such data are useful for hypothesis generation, service planning, and identifying outcomes requiring prospective evaluation. This study was conducted in accordance with the Revised STROCSS 2025 guideline for reporting cohort, cross-sectional, and case-control studies in surgery [11].

The present study aimed to compare electronic health record (EHR)-based clinical outcomes after endoscopic retrograde cholangiopancreatography (ERCP)-based endoscopic therapy vs. pancreatic surgery in adults with CP, using propensity score-matched data from the TriNetX US Collaborative Network. Outcomes of interest included chronic opioid prescriptions, opioid use disorder, celiac plexus block/neurolysis, pain codes, acute pancreatitis, exocrine and endocrine pancreatic insufficiency, emergency visits, all-cause mortality, and repeat intervention.

Materials and methods

Study population and cohort selection

Patients were included if they met the following criteria: (1) age ≥18 years; (2) diagnosis of CP (ICD-10-CM codes K86.0 [alcohol-induced] or K86.1 [other chronic]); and (3) an endoscopic or surgical intervention defined below, where any instance of the intervention occurred within 6 months during or after any instance of CP. This 6-month window was used to temporally relate the CP diagnosis and intervention. Patients were excluded if they had a diagnosis of pancreatic malignancy (C25), cystic fibrosis (E84), or secondary malignancy (C77, C78, C79).

Cohort A (endoscopic intervention)

Included 11,203 patients who underwent ERCP-based endoscopic intervention. A total of 56 providers from 67 responding healthcare organizations contributed patients to this cohort (Fig. 1).

Figure 1 Study flow diagram showing cohort selection and propensity score matching

Cohort B (surgical intervention)

Included 2252 patients who underwent surgical interventions, including distal subtotal pancreatectomy, proximal subtotal pancreatectomy (Whipple or pylorus-sparing Whipple-type procedures), or ampulla of Vater excision. A few procedures (e.g., Whipple’s/excision of ampulla) are not specific to CP-directed surgery; however, they were retained in the surgical code list because they can be used in specific cases of CP, and we could not confirm the indications for these procedures. A total of 47 providers from 70 responding healthcare organizations contributed patients to this cohort.

Propensity score matching

Propensity score matching was performed to balance baseline characteristics between cohorts before the outcome analysis. Matching was performed on demographics (age, sex, race/ethnicity), body mass index (BMI) categories, comorbid diagnoses, and baseline medication variables (including opioid medications). Available proxies for smoking status and alcohol-related CP included tobacco use and alcohol-related disorders. Pancreatic duct diameter, ductal stricture, intraductal stone burden, inflammatory mass, disease duration and detailed disease anatomy were not available as matching variables in this output and, therefore, could not be included in the propensity model.

After propensity score matching, the following patient numbers were available for analysis:

Cohort A (endoscopic intervention): 1451 patients (matched from 8843 pre-matching)

Cohort B (surgical intervention): 1451 patients (matched from 1494 pre-matching)

Follow-up time after matching

Cohort A mean 771.494 days (standard deviation [SD] 401.864), median 1095 days (interquartile range [IQR] 725); Cohort B mean 827.285 days (SD 391.715), median 1095 days (IQR 569) (Table 1). The median exceeded the mean because at least half of each matched cohort was followed to the 1095-day end of the analysis window, while patients with shorter available observation reduced the mean follow up; Kaplan-Meier methods were used to account for variable follow up and censoring.

Table 1 Patient characteristics before and after propensity score matching

Outcome definitions and analyses

The analysis evaluated the following primary and secondary outcomes:

Chronic opioid prescriptions: Receipt of any of the following opioid medications: morphine, oxycodone, methadone, buprenorphine, oxymorphone, fentanyl, tapentadol, hydromorphone or hydrocodone. This outcome captured post-index prescription records, but did not measure duration, dose, morphine milligram equivalents or persistence of use.

Opioid use disorder (OUD): Diagnosis of opioid abuse (F11.1), opioid dependence (F11.2) or opioid use unspecified (F11.9).

Celiac plexus block/neurolysis: Procedures for pain management including celiac plexus injection with anesthetic agent (CPT 64530) or destruction by neurolytic agent (CPT 64680).

All-cause mortality: Recorded demographic status of deceased.

Emergency room (ER) visits: Visit type recorded as emergency (HL7 visit type: EMER).

Acute pancreatitis: Diagnosis of acute pancreatitis (ICD-10-CM K85).

EPI: Diagnosis of exocrine pancreatic insufficiency (ICD-10-CM K86.81).

Endocrine pancreatic insufficiency: Diagnosis of diabetes mellitus due to underlying condition (E08), Type 2 diabetes mellitus (E11), or other specified diabetes mellitus (E13).

Need for repeat intervention: Occurrence of any of the following procedure codes: 43274, 43276, 43277, 48140, 48145, 48148, 48150, 48153, 48152, 48154.

Pain codes: Documentation of abdominal and pelvic pain (R10), pain not elsewhere classified (G89), or pain unspecified (R52).

Results

After propensity score matching, 1451 patients were included in each cohort, reduced from 8843 in the endoscopic cohort and 1494 in the surgical cohort before matching. Mean follow up after matching was 771.494 days (SD 401.864; median 1095 days, IQR 725) in the endoscopic cohort and 827.285 days (SD 391.715; median 1095 days, IQR 569) in the surgical cohort. Key comorbidities relevant to surgical selection were balanced after matching, including chronic ischemic heart disease (standardized difference 0.049), type 2 diabetes mellitus (standardized difference 0.047), and obesity/BMI variables (standardized differences ≤0.059).

For pain-related outcomes, new chronic opioid prescriptions occurred in 146/1092 patients in the endoscopic cohort and 79/1032 in the surgical cohort (risk 0.134 vs. 0.077), corresponding to a risk difference of 0.057 (95%CI 0.031-0.083; P<0.001), a RR of 1.747 (95%CI 1.346-2.266), and an OR of 1.862 (95%CI 1.396-2.484). Kaplan-Meier analysis showed survival probabilities at the end of the time window of 83.71% and 91.36%, respectively, with log-rank P<0.001 and a hazard ratio (HR) of 1.861 (95%CI 1.415-2.448). OUD occurred in 75/1362 vs. 52/1278 patients (risk 0.055 vs. 0.041), with a risk difference of 0.014 (95%CI -0.002 to 0.031; P=0.084), a RR of 1.353 (95%CI 0.958-1.912), and an OR of 1.374 (95%CI 0.956-1.974). Kaplan-Meier analysis showed survival probabilities of 92.60% and 95.10%, with a log-rank P-value of 0.058 and an HR of 1.406 (95%CI 0.987-2.003). Celiac plexus block/neurolysis was recorded in 42/1439 and 11/1421 patients (risk 0.029 vs. 0.008), yielding a risk difference of 0.021 (95%CI 0.012-0.031; P<0.001), a RR of 3.770 (95%CI 1.949-7.293), and an OR of 3.854 (95%CI 1.976-7.516). Corresponding Kaplan-Meier survival probabilities were 96.35% and 99.03%, with log-rank P<0.001 and an HR of 3.986 (95%CI 2.052-7.744). Pain codes were documented in 201/452 patients in the endoscopic cohort and 152/325 in the surgical cohort (risk 0.445 vs. 0.468), with a risk difference of -0.023 (95%CI -0.094 to 0.048; P=0.525), a RR of 0.951 (95%CI 0.814-1.110), and an OR of 0.911 (95%CI 0.685-1.213). Kaplan-Meier analysis showed median survival of 824 days in the endoscopic cohort and 872 days in the surgical cohort, with survival probabilities of 46.55% and 49.18%, log-rank P=0.049, and an HR of 0.806 (95%CI 0.652-0.995).

Pancreatic outcomes also differed between groups. Acute pancreatitis occurred in 167/530 patients in the endoscopic cohort and 65/565 in the surgical cohort (risk 0.315 vs. 0.115), corresponding to a risk difference of 0.200 (95%CI 0.153-0.248; P<0.001), a RR of 2.739 (95%CI 2.110-3.555), and an OR of 3.539 (95%CI 2.578-4.858). Kaplan-Meier survival probabilities at the end of follow up were 60.21% and 86.76%, with log-rank P<0.001 and an HR of 3.341 (95%CI 2.506-4.453) (Fig. 2). EPI developed in 200/1213 vs. 132/1206 patients (risk 0.165 vs. 0.109), with a risk difference of 0.055 (95%CI 0.028-0.083; P<0.001), a RR of 1.506 (95%CI 1.227-1.849), and an OR of 1.606 (95%CI 1.269-2.034). Kaplan-Meier survival probabilities were 78.72% in the endoscopic cohort and 87.39% in the surgical cohort, with log-rank P<0.001 and an HR of 1.585 (95%CI 1.272-1.975). Endocrine pancreatic insufficiency occurred in 125/881 patients in the endoscopic cohort and 117/885 in the surgical cohort (risk 0.142 vs. 0.132), with a risk difference of 0.010 (95%CI -0.022 to 0.042; P=0.554), a RR of 1.073 (95%CI 0.849-1.356), and an OR of 1.085 (95%CI 0.827-1.424). Kaplan-Meier survival probabilities were 81.18% and 83.64%, with a log-rank P-value of 0.372 and an HR of 1.122 (95%CI 0.872-1.443).

Figure 2 Forest plot of hazard ratios for post-matching time-to-event outcomes CI, confidence interval; ER, emergency room

For healthcare utilization and survival, all-cause mortality accounted for 100/1447 patients in the endoscopic cohort and 90/1449 in the surgical cohort (risk 0.069 vs. 0.062), yielding a risk difference of 0.007 (95%CI -0.011 to 0.025; P=0.447), a RR of 1.113 (95%CI 0.845-1.465), and an OR of 1.121 (95%CI 0.835-1.505). Kaplan-Meier survival probabilities were 90.97% and 92.81%, with a log-rank P-value of 0.287 and an HR of 1.167 (95%CI 0.878-1.552). Emergency room visits occurred in 181/937 vs. 174/1016 patients (risk 0.193 vs. 0.171), with a risk difference of 0.022 (95%CI -0.012 to 0.056; P=0.210), a RR of 1.128 (95%CI 0.934-1.362), and an OR of 1.159 (95%CI 0.920-1.458). Kaplan-Meier survival probabilities were 74.94% in the endoscopic cohort and 80.13% in the surgical cohort, with a log-rank P-value of 0.082 and an HR of 1.202 (95%CI 0.976-1.481). Survival probabilities at the end of the analysis window for each outcome are summarized in Fig. 3.

Figure 3 Kaplan-Meier end-of-window survival probabilities for post-matching outcomes in the endoscopic and surgical cohorts ER, emergency room

Discussion

This comparison of patients with CP undergoing index endoscopic therapy (ERCP-based intervention) vs. pancreatic surgery showed differences across selected outcomes, without a consistent advantage for either approach. In the matched cohorts, endoscopic therapy was associated with higher rates of new chronic opioid prescriptions, celiac plexus block/neurolysis, acute pancreatitis and EPI. In contrast, OUD, endocrine pancreatic insufficiency, emergency department utilization, and all-cause mortality were not significantly different between groups on risk-based or Kaplan-Meier analyses. Pain-code documentation did not differ significantly in the risk analysis, although the Kaplan-Meier output reported a statistically significant log-rank result with an HR below 1; this coded proxy should therefore be interpreted cautiously. The main new information from this study is that a multicenter EHR cohort demonstrated higher endoscopic-associated signals for opioid prescribing, acute pancreatitis, celiac plexus block/neurolysis and EPI, while mortality, emergency department utilization, OUD, endocrine insufficiency, and primary risk-based pain-code outcomes did not differ significantly. These observations are hypothesis-generating and should not be interpreted as proof of treatment superiority.

Pain in CP is multifactorial, and may reflect inflammatory, obstructive and neuropathic mechanisms; accordingly, guideline-based care emphasizes multidimensional assessment and multimodal management rather than reliance on any single surrogate, such as a diagnostic code or prescription pattern [12]. Within that context, the higher incidence of new chronic opioid prescriptions after endoscopic therapy may indicate less durable symptom control in some patients, recurrent inflammatory episodes prompting escalation of analgesia, or differences in disease phenotype not fully captured by EHR-based covariates. However, the opioid endpoint captured prescription occurrence, rather than duration, dose, morphine milligram equivalents, or patient-reported analgesic response. Celiac plexus block/neurolysis was also more frequent after endoscopic therapy, whereas OUD and pain-code documentation on risk analysis were not significantly different.

Randomized data in selected patients with painful obstructive CP have suggested that early surgery can provide better pain control than an endoscopy-first approach [9]. Longer-term follow up from the same comparison also indicates that the 2 strategies may diverge over time in clinically meaningful outcomes [13]. Our findings are directionally consistent with those data with respect to chronic opioid exposure, but they do not demonstrate a clear difference across all pain-related endpoints. These findings should be interpreted in the context of disease phenotype, symptom drivers and timing of intervention.

Despite higher chronic opioid prescribing after endoscopic therapy, OUD was not significantly different between groups. This finding should be interpreted carefully, because OUD may be under-recognized or inconsistently coded in routine practice, and because opioid exposure in CP may reflect both disease burden and prescribing culture. Nonetheless, CP remains a clinical setting in which opioid stewardship, longitudinal monitoring and attention to non-opioid pain strategies are important, irrespective of treatment pathway [14].

Acute pancreatitis occurred more frequently after endoscopic therapy than after surgery. This association is clinically plausible, as ERCP-based pancreatic duct manipulation is associated with procedure-related pancreatitis risk, and repeated instrumentation or incomplete control of ductal obstruction may contribute to recurrent inflammatory events in some patients [15]. However, we cannot reliably distinguish post-ERCP pancreatitis from disease-related flares; this outcome is not directly comparable between cohorts, and should be interpreted as an observed association, rather than a direct mechanistic effect of the procedure itself.

Recent randomized data evaluating endoscopic duct decompression strategies, including extracorporeal shock-wave lithotripsy plus endoscopy, suggest that the benefit may be modest and not universal, again emphasizing the heterogeneity in response [16]. These findings support selective use of endoscopic therapy, with early surgical review when disease characteristics suggest limited endoscopic durability.

EPI was more frequent after endoscopic therapy. In CP, EPI reflects cumulative parenchymal injury and progressive loss of exocrine function, and current guidance emphasizes early recognition and treatment, because maldigestion and malnutrition have important clinical consequences [17]. One possible interpretation is that the higher burden of acute pancreatitis observed in the endoscopic cohort may be accompanied by greater ongoing pancreatic injury in some patients. However, residual confounding, differential clinical surveillance and coding practices may also have contributed to this difference.

Endocrine pancreatic insufficiency did not differ significantly between the endoscopic and surgical cohorts in our study. Although pancreatic surgery, particularly when resection is involved, may reduce endocrine reserve in a gland that is already functionally compromised by CP, the present EHR-based comparison did not demonstrate a statistically significant difference in endocrine insufficiency during the analysis window. The definition included diabetes mellitus due to underlying condition, type 2 diabetes mellitus, and other specified diabetes mellitus; therefore, it does not isolate pancreatogenic diabetes and may misclassify diabetes etiology. Pancreatogenic diabetes remains a recognized complication of both CP and pancreatic surgery, and consensus guidance supports systematic screening and structured management in this setting [18].

Emergency department utilization was not significantly different between groups on either risk-based analysis or Kaplan-Meier analysis. Given the higher rates of acute pancreatitis and chronic opioid prescriptions after endoscopic therapy, a signal toward greater unscheduled care might be clinically plausible, but it was not statistically demonstrated. Healthcare utilization remains influenced by local practice patterns, access to care, and EHR capture.

All-cause mortality also did not differ significantly between groups on risk-based analysis or Kaplan-Meier analysis. Mortality was defined as a recorded demographic status of “deceased” and was not adjudicated as CP-attributable, procedure-related, or due to other causes. This should not be taken as evidence of equivalence, because the cause of death was not adjudicated, and important factors such as frailty, nutritional status, alcohol and smoking exposure, disease morphology, surgical candidacy, early post-procedural complications, and center-level practice variation are not fully captured in this type of dataset. The study therefore does not support a statistically significant all-cause mortality difference during the reported analysis window, but it cannot determine comparative procedural safety or early complication-related mortality.

The strengths of this study include a large multicenter real-world dataset and the use of propensity score matching to reduce measured baseline imbalance between treatment groups. However, several limitations remain and should be considered for interpretation. First, residual confounding by indication is a major limitation: treatment allocation in CP is influenced by ductal anatomy, stone burden, inflammatory mass, disease duration, symptom severity, nutritional status, surgical candidacy, alcohol and smoking intensity, center expertise, and patient preference, many of which were not available in the database. Matching therefore reduced the measured imbalance, but cannot establish causality or treatment superiority. Second, exposure misclassification remains possible. The endoscopic cohort was restricted to ERCP-based stent and ductal dilation/sphincteroplasty codes, but surgical codes still included heterogeneous resection procedures and CPT 48148, which is not specific to CP-directed treatment. Third, outcome misclassification is possible, because diagnoses, procedures, medications, emergency department visits and death status were identified from structured EHR codes. Pain was assessed using coded proxies, rather than validated patient-reported instruments such as visual analog, numeric rating or Izbicki scores. Chronic opioid prescriptions did not capture dose, duration, morphine milligram equivalents or persistent use. Acute pancreatitis could not be separated into post-ERCP pancreatitis and disease-related flares, which may bias interpretation of this endpoint against endoscopic therapy. Diabetes definitions included type 2 diabetes, and therefore did not isolate pancreatogenic diabetes. Fourth, although patients with prior recorded outcomes were excluded from outcome-specific analyses, no fixed washout period was applied; some post-index outcomes may represent previously undocumented disease progression rather than definite treatment-induced complications. Fifth, the 6-month CP diagnosis-to-intervention window may have introduced heterogeneity and time-window bias. Sixth, early procedural complications severe enough to limit subsequent follow up may be under-captured, and all-cause mortality cannot be attributed to CP or the intervention. Finally, we cannot provide the total number of endoscopic procedures per patient; after excluding patients with prior repeat-intervention records, the repeat-intervention endpoint was not estimable, limiting interpretation of comparative treatment durability.

Taken together, these findings support a pragmatic and individualized approach to CP management. Early multidisciplinary review, involving endoscopy, surgery, pain management, nutrition and endocrinology, remains appropriate. The present data suggest that patients managed with an endoscopy-first strategy may warrant close attention to recurrent pancreatitis, exocrine functional decline, celiac plexus block/neurolysis and escalation of opioid exposure, but these associations should be interpreted alongside the limitations of EHR coding and residual confounding by indication. Surveillance for diabetes remains important after both endoscopic and surgical treatment pathways. These observations also reinforce the importance of structured opioid stewardship and proactive screening for both EPI and diabetes as part of longitudinal CP care [12,17,18].

In conclusion, in this study comparing endoscopic therapy to pancreatic surgery for CP, endoscopic therapy was associated with higher rates of new chronic opioid prescriptions, celiac plexus block/neurolysis, acute pancreatitis and EPI during follow up. No significant between-group differences were observed for opioid use disorder, endocrine pancreatic insufficiency, emergency department utilization or all-cause mortality. Pain documentation was not significantly different on risk analysis, although the Kaplan-Meier analysis for this coded proxy reached statistical significance and should be interpreted cautiously. The need-for-repeat-intervention endpoint was not estimable after exclusion of patients with prior outcome records. Overall, these results highlight clinically relevant tradeoffs between endoscopic and surgical strategies in real-world practice, but they remain associative and hypothesis-generating. Careful multidisciplinary patient selection, structured opioid stewardship, and surveillance for pancreatic functional decline remain important. Prospective studies with granular disease-phenotype data, procedure counts, early complication capture and patient-reported outcomes are needed to better define which patients benefit most from early surgery vs. an endoscopy-first approach.