Natural History of Recurrent Acute Pancreatitis: Implications for Interventional Trial Design

ABSTRACT

Background

Patients with acute pancreatitis (AP) often experience recurrent acute pancreatitis (RAP), increasing the risk of chronic pancreatitis (CP) and long‐term complications. Progression along the AP–RAP–CP spectrum has been explored in epidemiological studies, but available data provide limited insight into recurrence dynamics within RAP itself. This information is crucial for the design of studies aiming to prevent RAP recurrence and disease progression.

Methods

In this nationwide cohort study, we included 45,338 Danish adults hospitalised with an incident episode of AP between January 1, 2000, and December 31, 2022, identified through national health registries. RAP was defined as recurrent hospitalisations for AP separated by ≥ 90 days. Competing risk models with pseudo‐observations estimated cumulative incidence functions for RAP recurrence stratified by the number of prior AP episodes. Inter‐episode intervals were estimated using Weibull models.

Results

During a median follow‐up of 4.5 years, 6894 patients developed RAP, corresponding to a 15.2% risk (95% confidence interval, 14.9%–15.5%). The RAP recurrence risk increased with accumulating episodes. The 12‐month cumulative incidence of RAP recurrence rose from 17% after one RAP episode to 46% after five episodes. Most recurrences occurred within 12 months, and inter‐episode intervals shortened with increasing episode number. Weibull models demonstrated a wear‐out pattern, with shape parameters increasing progressively across successive RAP episodes, indicating accelerating recurrence risk over time.

Conclusions

RAP follows an accelerating course characterised by increasing recurrence risk. Risk‐enriched trial designs targeting patients with more prior episodes may improve the feasibility and statistical efficiency of future interventional studies.

In this nationwide cohort of 6894 patients with recurrent acute pancreatitis, recurrence risk increased progressively with each episode and occurred most frequently within the first year. Successive episodes were separated by shorter intervals, indicating an accelerating disease course and highlighting the importance of early preventive interventions.

Acute pancreatitis (AP) imposes a substantial global burden on healthcare systems, with high rates of emergency admission and considerable associated costs [1, 2]. Current management remains largely supportive, as no interventions are approved for clinical use to modify the disease course once an acute episode is established. Consequently, prevention of recurrent attacks through targeting modifiable risk factors is an important clinical priority [2]. Preventive measures address specific aetiologies, such as cholecystectomy for biliary pancreatitis or lifestyle modification in alcohol‐related disease [3]. However, a significant proportion of patients lack identifiable or modifiable risk factors and face a high risk of recurrence [4]. Recurrent acute pancreatitis (RAP) is characterised by repeated episodes of pancreatic inflammation that frequently necessitate hospitalisation [5]. Patients with RAP are at increased risk of progression to chronic pancreatitis (CP), a condition associated with irreversible structural pancreatic damage and significant long‐term complications, including chronic abdominal pain, exocrine pancreatic insufficiency, and diabetes [6, 7]. These complications substantially impair quality of life and contribute to long‐term morbidity and increased mortality [8, 9, 10]. Therefore, the development of interventions that can reduce the risk of pancreatitis recurrence and halt disease progression in patients with RAP is an unmet clinical need.

From a trial‐design perspective, randomised clinical trials aiming to modify the natural history of RAP should focus on targeting underlying disease mechanisms. Recently published trials have adopted this approach, including the SHARP trial (minor papilla sphincterotomy in patients with pancreas divisum) [11], the SIMBA trial (investigating the pleiotropic anti‐inflammatory effects of simvastatin) [12], and the PAMORA‐RAP trial (evaluating a peripheral μ‐opioid receptor antagonist to mitigate pancreatic injury from endogenous and exogenous opioids) [13]. Although these studies aimed to reduce the risk of RAP by targeting presumed pathogenic pathways, none provided definitive evidence of efficacy in preventing further attacks. Moreover, all trials encountered substantial feasibility challenges, most notably difficulties with participant recruitment and lower‐than‐anticipated event accrual, leading to premature trial termination [11, 12, 13]. These limitations underscore gaps in the evidence base informing trial design in RAP. While progression along the AP–RAP–CP spectrum has been explored in epidemiological studies, available data provide limited insight into recurrence dynamics within RAP itself [14, 15]. In particular, estimates of recurrence risk, inter‐episode intervals, and their relationship to prior RAP burden remain sparse. Such parameters are critical for defining eligible populations, selecting clinically meaningful endpoints, and determining appropriate sample size and follow‐up duration for adequately powered interventional trials.

In this population‐based study, we characterise the natural history of RAP, with specific focus on parameters relevant to the design of future interventional trials. We estimated the cumulative incidence of pancreatitis recurrence as a function of the number of prior RAP episodes, quantified the time intervals between successive episodes, and assessed how recurrence risk evolves over the disease course. Based on these estimates, we simulated sample size requirements for future trials. These data are intended to provide empirically grounded estimates to inform feasibility assessments, endpoint selection, and statistical power calculations for future randomised clinical trials aimed at preventing RAP recurrence and progression to CP.

1. Methods

1.1. Study Design and Data Sources

This was a Danish nationwide retrospective cohort study. The study cohort comprised all Danish adults (≥ 18 years) hospitalised with an incident episode of AP between January 1, 2000, and December 31, 2022. Anonymised data were provided by Statistics Denmark (Project Identifier 708466). Access to these data is restricted to authorised research institutions approved by the Danish Health Data Authority.

Data for the primary analyses were obtained from the Danish National Patient Registry (DNPR) and the Danish Civil Registration System. Information on AP, RAP, CP and pancreatic cancer was obtained from the DNPR. This is a nationwide registry that records dates and discharge diagnoses for all inpatient and outpatient hospital contacts in Denmark. Diagnoses in the DNPR are coded according to the International Classification of Diseases, 10th Revision (ICD‐10), which has been used in Denmark since 1994 [16]. Demographic characteristics, migration history, and vital status were obtained from the Danish Civil Registration System, an administrative registry established in 1968 and updated daily, ensuring virtually complete follow‐up of all Danish residents [17]. To further characterise the study cohort at baseline, including comorbidities and proxies for alcohol and smoking history, we used data from the DNPR and the Danish National Prescription Registry. The latter includes information on the date of dispensing, product name, and Anatomical Therapeutic Chemical (ATC) classification code for all prescription medications dispensed at Danish pharmacies [18].

Individual‐level linkage across registries was achieved using the unique personal identification number assigned to all Danish residents at birth or upon immigration [17].

1.2. Study Cohort

The study cohort comprised patients with incident AP (ICD‐10: K85.x). The index date was defined as the date of hospitalisation for the incident AP episode. We excluded individuals younger than 18 years of age and patients with a history of pancreatic disease. The latter included patients with a diagnosis of acute pancreatitis, recurrent acute pancreatitis, chronic pancreatitis, or pancreatic cancer prior to cohort inclusion (i.e., prevalent pancreatic disease). Prevalent disease was excluded within the look‐back period (earliest January 1, 1994) prior to the index date using the relevant ICD‐10 codes, as outlined below.

1.3. Outcomes

The primary outcome was recurrence of AP during follow‐up. AP was identified in the DNPR using ICD‐10 code K85.x, and recurrent episodes were defined as hospitalisations for AP separated by a minimum interval of 90 days to ensure that AP episodes were independent [5, 19]. Because patients could develop other pancreatic diseases or die during follow‐up, thereby precluding the occurrence of RAP, CP (ICD‐10: K86.0 and K86.1), pancreatic cancer (ICD‐10: C25.x), and death were treated as competing events. Figure 1 shows a swimmer plot of 50 randomly selected patients illustrating the study design and timing of events.

FIGURE 1.

Swimmer plot depicting the disease course of 50 patients with incident acute pancreatitis. Patients were randomly selected from each stratum defined by the number of prior recurrent acute pancreatitis (RAP) episodes. Each horizontal line represents a single patient, with markers indicating recurrent AP events and competing events (chronic pancreatitis, pancreatic cancer, or death) over time. AP, acute pancreatitis; CP, chronic pancreatitis; PDAC, pancreatic ductal adenocarcinoma.

1.4. Baseline Characteristics

Patients' baseline characteristics were ascertained using all available information preceding each patient's index date until study entry. Information on age and sex was obtained from the Danish Civil Registration System. Comorbidity burden was assessed using the Charlson Comorbidity Index (CCI), calculated based on ICD‐10 diagnoses recorded in the DNPR (Table S1). Smoking status, defined as heavy smoking, was identified using tobacco‐related ICD‐10 codes and prescriptions for tobacco‐dependence treatments or obstructive pulmonary disease medications among individuals aged 40 years or older, in accordance with established methodologies [20]. Likewise, alcohol abuse was identified based on diagnoses or prescription records indicative of alcohol‐related diseases or disorders [20]. Detailed definitions and corresponding ICD‐10 codes are provided in Table S2. History of cholecystectomy prior to cohort entry was identified using procedure codes from the DNPR (JKA20 for laparoscopic cholecystectomy and JKA21 for open cholecystectomy).

1.5. Statistical Analysis

We applied competing risk models to estimate the cumulative incidence function of pancreatitis recurrence, accounting for competing events, including other pancreatic diseases and all‐cause mortality. Analyses were stratified by the number of prior RAP episodes. To directly model the cumulative incidence function in the presence of competing risks, we employed the pseudo‐observation approach, which accommodates censoring and allows estimation of absolute risks at the patient‐level [21]. Cumulative incidence estimates were reported with 95% confidence intervals (CIs).

To analyse the temporal dynamics of RAP, we modelled the time intervals between RAP episodes using Weibull modelling [22]. Models were stratified according to the number of prior RAP episodes. In this context, the Weibull shape parameter ($k$) was used to assess changes in hazard over time with cumulative RAP events: $k>1$ indicates an increasing hazard (‘wear‐out’ effect), $k=1$ a constant hazard, and $k<1$ a decreasing hazard.

Simulated sample size requirements were estimated for Cox proportional hazards models using the Schoenfeld formula to determine the number of events needed to detect a specified treatment effect. Calculations assumed equal allocation between groups, 80% power, and a two‐sided significance level of 0.05. Under these assumptions, 246 events were required to detect a hazard ratio of 0.7 (30% relative risk reduction), and 65 events were required to detect a hazard ratio of 0.5 (50% relative risk reduction). These effect sizes are consistent with those targeted in recent randomised trials aiming to modify the natural history of RAP [11, 12, 13]. Hypothetical total sample sizes were then derived by dividing the required number of events by the observed cumulative incidence within each stratum of prior RAP episodes. The 12‐month cumulative incidence estimates were used for this analysis.

A sensitivity analysis was performed using an alternative definition of RAP requiring a 30‐day (rather than 90‐day) separation between episodes. Cumulative incidence functions and Weibull regression models were re‐estimated under this alternative definition. Finally, we performed stratified analyses based on alcohol abuse and prior cholecystectomy using the same analytical approach as in the primary analysis.

All data management and statistical analyses were performed using Stata 18.0 (StataCorp, College Station, TX, USA).

1.6. Declaration of Generative AI and AI‐Assisted Technologies

During the preparation of this work, the authors used ChatGPT (OpenAI) to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and took full responsibility for the content of the publication.

2. Results

Among 45,338 individuals with incident AP who had no prior history of CP or pancreatic cancer and were followed for a median of 4.5 years (interquartile range [IQR], 1.8–9.8 years), 6894 patients experienced one or more RAP events and were therefore included in the RAP cohort. This corresponds to a RAP risk of 15.2% (95% CI, 14.9%–15.5%) among patients with incident AP. Among the 6894 patients with RAP, 4999 (72.5%) had one RAP episode, 1189 (17.2%) two episodes, and 706 (10.2%) three or more RAP episodes during follow‐up (Table 1).

TABLE 1.

Demographic and clinical characteristics of the total recurrent acute pancreatitis (RAP) cohort and stratified by number of RAP episodes during follow‐up.

	Total cohort	Cohort stratified by number of RAP episodes
		1 Episode	2 Episodes	3 Episodes	4 Episodes	≥ 5 Episodes
Number of patients (%)	6894 (100)	4999 (72.5)	1189 (17.2)	383 (5.6)	165 (2.4)	158 (2.3)
Time from index AP, median years (IQR)	1.2 (2.6)	0.7 (1.7)	2.1 (2.1)	3.3 (3.9)	3.7 (4.7)	5.9 (5.0)
Mean age (SD)	55.9 (17.2)	57.5 (17.2)	53.5 (16.3)	50.4 (16.9)	47.6 (16.3)	46.0 (15.0)
Age, n (%)
	1379 (20.0)	875 (17.5)	266 (22.4)	118 (30.8)	61 (37.0)	59 (37.3)
40–60 years	2624 (38.1)	1830 (36.6)	507 (42.6)	156 (40.7)	62 (37.6)	69 (43.7)
	2891 (41.9)	2294 (45.9)	416 (35.0)	109 (28.5)	42 (25.5)	30 (19.0)
Male sex, n (%)	3959 (57.4)	2794 (55.9)	733 (61.6)	229 (59.8)	93 (56.4)	110 (69.6)
CCI group, n (%)
0	1875 (27.2)	1338 (26.8)	310 (26.1)	127 (33.2)	55 (33.3)	45 (28.5)
1	1278 (18.5)	890 (17.8)	236 (19.8)	71 (18.5)	42 (25.5)	39 (24.7)
	3741 (54.3)	2771 (55.4)	643 (54.1)	185 (48.3)	68 (41.2)	74 (46.8)
Alcohol abuse, n (%)	1545 (22.4)	1090 (21.8)	292 (24.6)	97 (25.3)	30 (18.2)	36 (22.8)
Heavy smoking, n (%)	2168 (31.4)	1604 (32.1)	378 (31.8)	97 (25.3)	44 (26.7)	45 (22.8)
Cholecystectomy, n (%)	456 (6.6)	343 (6.9)	76 (6.4)	18 (4.7)	9 (5.5)	10 (6.3)

Abbreviations: AP, acute pancreatitis; CCI, Charlson comorbidity index.

2.1. Patient Characteristics

Baseline demographic and clinical characteristics of the RAP cohort are summarised in Table 1. The mean age at inclusion was 55.9 years, and 57.4% of patients were men. Alcohol abuse was observed in 22.4% of patients, 31.4% were heavy smokers, and 72.8% had one or more comorbidities (i.e., Charlson Comorbidity Index [CCI] score ≥ 1). Cholecystectomy had been performed in 6.6% of RAP patients prior to inclusion. An age gradient was observed across patients stratified by the number of RAP episodes, with decreasing mean age and an increasing proportion of younger patients (< 40 years) as the number of RAP episodes increased. Otherwise, no major differences were observed across groups defined by the number of RAP episodes.

2.2. Risk of Pancreatitis Recurrence

Figure 2 illustrates the cumulative incidence function of pancreatitis recurrence, stratified by the number of prior RAP episodes. The corresponding estimates at 6, 9, 12, and 24 months are presented in Table 2. Recurrence risk increased progressively with each additional prior episode. For example, at 12 months, the cumulative incidence was 0.17 (95% CI, 0.16–0.18) for patients with a single prior RAP episode, rising to 0.46 (95% CI, 0.38–0.55) for those with five or more prior RAP episodes, highlighting a marked escalation in recurrence risk with accumulating RAP events.

FIGURE 2.

Cumulative incidence function (CIF) of pancreatitis recurrence, stratified by the number of prior recurrent acute pancreatitis (RAP) episodes. Each curve represents the estimated risk of recurrence over time for patients with a specific number of prior episodes, accounting for competing events such as chronic pancreatitis, pancreatic cancer and death.

TABLE 2.

Cumulative incidence of pancreatitis recurrence at 6, 9, 12 and 24 months stratified by the number of previous recurrent acute pancreatitis (RAP) episodes.

Transition	6 months	9 months	12 months	24 months
RAP 1 ➔ RAP 2	0.10 (0.09–0.10)	0.14 (0.14–0.15)	0.17 (0.16–0.18)	0.24 (0.22–0.25)
RAP 2 ➔ RAP 3	0.14 (0.12–0.16)	0.21 (0.19–0.23)	0.25 (0.23–0.27)	0.34 (0.32–0.37)
RAP 3 ➔ RAP 4	0.18 (0.15–0.21)	0.30 (0.27–0.34)	0.36 (0.33–0.40)	0.46 (0.42–0.50)
RAP 4 ➔ RAP 5	0.22 (0.17–0.26)	0.34 (0.29–0.39)	0.43 (0.37–0.48)	0.55 (0.49–0.61)
RAP 5 ➔ RAP ≥ 6	0.20 (0.14–0.50)	0.35 (0.27–0.43)	0.46 (0.38–0.55)	0.56 (0.48–0.65)

2.3. Time Between Recurrent Pancreatitis Episodes

Most pancreatitis recurrence events occurred within the first 12 months after the previous episode, and intervals shortened with an increasing number of prior episodes (Figure S1). Building on this observation, Weibull models of the time intervals between pancreatitis episodes, stratified by the number of prior RAP episodes, were employed. These models revealed a wear‐out phenomenon (shape parameters > 1) with gradually increasing shape parameters with the accumulation of the number of prior RAP events (Figure 3). This indicates an increasing risk of pancreatitis recurrence and progressively shorter intervals between episodes as the number of prior RAP events accumulates.

FIGURE 3.

Weibull models of the time intervals between pancreatitis episodes, stratified by the number of prior recurrent acute pancreatitis (RAP) episodes. Models demonstrate a ‘wear‐out’ pattern (shape parameter > 1) that increases progressively (indicated by increasing shape parameter values) with accumulating RAP events. This indicates an accelerating risk of pancreatitis recurrence and progressively shorter intervals between episodes as the number of prior RAP events increases.

2.4. Sample Size Simulations

Table 3 presents hypothetical sample size simulations for two effect sizes (hazard ratios of 0.5 and 0.7) and the observed 12‐month cumulative incidence of pancreatitis recurrence. Overall, the simulations indicate that including patient populations with more prior RAP episodes reduces the required sample size.

TABLE 3.

Sample size simulations.

Number of prior RAP episodes*	Proportion of RAP cohort (%)	Cumulative incidence*	HR of interest: 0.5		HR of interest: 0.7
			Events required	Sample size estimation	Events required	Sample size estimation
1 Episode	72.5	0.17	65	382	246	1447
2 Episodes	17.2	0.25	65	260	246	984
3 Episodes	5.6	0.36	65	181	246	683
4 Episodes	2.4	0.43	65	151	246	572
≥ 5 Episodes	2.3	0.46	65	141	246	535

Abbreviations: HR, hazard ratio; RAP, recurrent acute pancreatitis.

^*The 12‐month cumulative incidence of pancreatitis recurrence.

2.5. Sensitivity Analysis

Using a 30‐day (vs. 90‐day) interval, the sensitivity analysis identified 10,487 patients with one or more RAP episodes. Corresponding cumulative incidence functions and Weibull models are provided in Supplemental Figures 2 and 3. The corresponding cumulative incidence estimates at 3, 6, 9, 12, and 24 months are presented in Table S3. The sensitivity analysis revealed slightly higher cumulative RAP incidences, with findings from Weibull models comparable to those observed in the primary analysis (i.e., accelerating recurrence risk with accumulating RAP episodes). However, the swimmer plot visualisation revealed that several RAP episodes occurred in close temporal proximity to the preceding episode, suggesting these episodes could represent readmissions for unresolved inflammation or complications of the former pancreatitis admission rather than a new and independent episode (Figure S4).

2.6. Stratified Sub‐Analyses by Aetiology

In sub‐analyses stratified by alcohol abuse and prior cholecystectomy, the observed risk of pancreatitis recurrence and the interval between recurrent episodes were largely comparable to the findings of the primary analysis (Figures [Link], [Link]).

3. Discussion

In this Danish nationwide population‐based cohort of 45,338 patients with incident AP, a significant proportion (15.2%) developed RAP during follow‐up, yielding almost 7000 cases for detailed RAP recurrence analysis. We demonstrate that pancreatitis recurrence risk increased progressively with accumulating RAP episodes. The 12‐month cumulative incidence rose from 17% after a first RAP episode to 46% among patients with five or more episodes. Recurrences clustered within the first year from the prior episode, and inter‐episode intervals shortened as prior events accumulated. Weibull modelling identified a wear‐out pattern and gradually increasing shape parameters with an increasing number of RAP episodes, indicating an accelerating rather than a constant pancreatitis recurrence risk over time. Together, these findings support a progressively intensifying disease course. Sample size simulations showed that restricting inclusion to patients with more prior RAP episodes substantially reduced required sample sizes, but at the same time reduced the number of eligible patients.

3.1. Linking Empirical Observations With Mechanistic Insights in RAP

A salient finding of this study was the accelerating risk of pancreatitis recurrence with an accumulating number of prior RAP episodes. We observed not only a progressive rise in cumulative recurrence risk, but also shortening intervals between episodes and a wear‐out pattern in Weibull models, indicating that the risk of recurrence intensifies over time. This suggests that RAP does not follow a static or random course; rather, the disease appears to evolve dynamically, with each episode conferring greater vulnerability to subsequent events. Mechanistic frameworks further underscore the importance of these observations. The sentinel acute pancreatitis event (SAPE) hypothesis posits that an initial inflammatory insult primes the pancreas through immunological sensitization, thereby lowering the threshold for future attacks and promoting progression along the AP–RAP–CP continuum [23, 24]. Within this conceptual model, recurrence frequency and inter‐episode timing are not merely descriptive clinical metrics, but potential markers of underlying disease activity and progression. The accelerating recurrence pattern observed in our data is consistent with such an inflammatory sensitization process and provides empirical support for this pathophysiological framework [23, 24].

3.2. Implications for Trial Design in RAP

The observation of an accelerating recurrence risk with accumulating RAP episodes has several important implications for the design of interventional trials aimed at reducing pancreatitis recurrence in RAP [11, 12, 13]. First, patient selection should account for prior disease burden. Because recurrence risk increases with each additional RAP episode, and intervals between episodes shorten, enrolling patients with more prior RAP events will enrich the study population for near‐term outcomes. This strategy increases event rates, reduces required follow‐up time, and markedly lowers the sample size needed to detect clinically meaningful effect sizes. In contrast, trials enrolling patients after only one or two RAP episodes may face low short‐term event rates, prolonged follow‐up requirements, and feasibility challenges, as reflected in previous studies [11, 12, 13]. However, the inclusion of patients with more prior RAP episodes significantly reduces the pool of eligible patients. Accordingly, the majority of patients (72.5%) only had one RAP episode, and patients with three or more RAP episodes comprised only 10.2% of our population‐based RAP cohort. Second, the non‐constant hazard of recurrence suggests that time‐to‐event assumptions in trial planning should move beyond models that assume proportional and stable risk over time. An accelerating hazard supports the use of flexible survival models and careful consideration of how prior RAP episode number is incorporated, either as a stratification factor, an inclusion criterion, or a covariate in the primary analysis. Failure to account for this dynamic risk structure may lead to underpowered studies or misinterpretation of treatment effects. Third, recurrence frequency and inter‐episode intervals may serve not only as binary endpoints (event versus no event), but as quantitative measures of disease modification. If RAP reflects an inflammatory sensitization process consistent with the SAPE framework, then an effective intervention might be expected to lengthen inter‐episode intervals or attenuate the acceleration in recurrence risk over time. Incorporating recurrent‐event analyses or modelling changes in recurrence dynamics could therefore provide a more sensitive assessment of therapeutic effect than a single first event endpoint alone. Fourth, follow‐up duration can be more precisely aligned with underlying risk. Given that most recurrences occur within 12 months, and particularly among patients with multiple prior episodes, primary endpoints at 6 or 12 months may be sufficient in enriched high‐risk populations. This has direct implications for feasibility, cost, and participant retention. Finally, these findings support a shift toward mechanism‐informed enrichment strategies. If disease progression accelerates after repeated inflammatory insults, interventions targeting inflammatory amplification, immune modulation, or fibrosis pathways may be most impactful when deployed in patients who have already demonstrated a recurrent phenotype. Aligning biological rationale with empirically defined high‐risk groups may increase the likelihood of detecting clinically meaningful benefits. Collectively, these considerations suggest that future RAP trials should integrate cumulative RAP episode burden into eligibility criteria, statistical design, endpoint definition, and follow‐up planning. Doing so may improve feasibility and enhance the probability of successfully identifying therapies capable of altering the natural history of RAP.

3.3. Validity of the 90‐Day Threshold for RAP Definition

The sensitivity analysis using a 30‐day separation interval between RAP episodes revealed important limitations that support the appropriateness of the 90‐day definition. Hence, the swimmer plot visualisation demonstrated that several RAP episodes classified under the 30‐day threshold occurred in close temporal proximity to the preceding episode, raising concern for misclassification of readmissions due to persistent inflammation or unresolved complications as distinct recurrent episodes [3, 5, 19]. This finding aligns with the established understanding of the natural history of acute pancreatitis: while mild disease typically resolves within 1 week, moderately severe and severe cases can develop local complications that persist for weeks to months [3]. Pancreatic and peripancreatic fluid collections typically evolve over a 4‐week period, with acute necrotic collections maturing into walled‐off necrosis typically after 4 weeks [3]. The 90‐day threshold used in the primary analysis provides adequate temporal separation to ensure complete resolution between episodes, consistent with the American Gastroenterological Association definition of RAP as ‘2 or more distinct episodes of AP with complete intervening resolution, separated by at least 3 months’ [19].

3.4. Strengths and Limitations

Key strengths include the nationwide, population‐based design, large sample size, long observation period, and virtually complete follow‐up enabled by Danish health registries [17]. These features allowed robust modelling of recurrence dynamics. The use of competing risk methods strengthens validity by accounting for death, CP, and pancreatic cancer. Finally, we translated epidemiological findings into actionable trial‐design parameters through formal sample size simulations and outlined considerations for the design of future trials.

Limitations should also be acknowledged. Diagnoses were based on ICD‐10 coding in administrative data. Although misclassification is possible, AP coding validity in Danish health registries is high, with a positive predictive value of 97.3% [25]. We did not conduct sensitivity analyses using alternative ICD‐10 definitions, as AP is typically coded consistently once the diagnosis has been clinically established. Therefore, we considered substantial differential misclassification across recurrent episodes to be unlikely. Additionally, the presented data likely reflect predominantly severe cases of RAP requiring hospitalisation. Many patients have presumably experienced milder RAP episodes managed at home without hospital admission, which are therefore not captured in the present study. In the context of a randomised controlled trial, such episodes would more likely be identified during follow‐up, potentially influencing power calculations depending on the selected endpoint [26]. Our data do not reliably distinguish between inpatient and outpatient encounters; therefore, all AP events were included irrespective of care setting. An episode was defined as the first AP event followed by a 90‐day window, with any subsequent events within this period classified as part of the same episode. A new episode was defined by the first event occurring after 90 days. This approach introduces a potential risk of misclassification, whereby short, distinct episodes may be merged, and prolonged courses may be split across episodes. Sensitivity analyses using a 30‐day window, however, produced similar results, suggesting that misclassification is unlikely to have materially affected the findings. An additional important limitation of our study is the limited availability of detailed patient‐level characteristics, which prevented stratified analyses by specific aetiologies, risk factors, and other clinically relevant variables. As a partial mitigation, we performed subgroup analyses using available proxies for aetiology (alcohol abuse and prior cholecystectomy), and these yielded results comparable to the primary analysis. However, these proxies should be interpreted cautiously. Information on alcohol abuse was derived through indirect assessment methods, and cholecystectomy is not a definitive indicator of biliary aetiology in the context of RAP, as this procedure has historically been offered to patients despite a lack of definitive evidence for biliary aetiology [27]. Comorbidities were also ascertained indirectly through health registries, which may introduce bias, although previous studies from Scandinavia have demonstrated close agreement between CCI scores derived from health registries and those obtained from medical record review [28]. Future studies are needed to better characterise patient profiles and to further elucidate their relationship with RAP trajectories. Finally, generalisability to healthcare systems with different coding practices or admission thresholds warrants consideration.

4. Conclusions

RAP follows a progressively accelerating course, characterised by increasing recurrence risk and shortening inter‐episode intervals as prior pancreatitis episodes accumulate. These findings support a dynamic disease model consistent with immunological sensitization and provide an empirical framework for designing future trials. Risk‐enrichment strategies based on cumulative RAP episode burden may enhance feasibility and statistical efficiency, strengthening the foundation for interventional studies aimed at preventing pancreatitis recurrence and disease progression.

Author Contributions

Søren S. Olesen: conceptualization, methodology, writing – original draft, formal analysis. Dhiraj Yadav: writing – review and editing. Asbjørn M. Drewes: writing – review and editing. Srdan Novovic: writing – review and editing. Stig B. Laursen: writing – review and editing. Niels Henrik Bruun: methodology, formal analysis, writing – review and editing, data curation. Enrique de‐Madaria: writing – review and editing. Line Davidsen: writing – review and editing.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supplemental Figure S1: Histograms of time intervals between recurrent acute pancreatitis (RAP) episodes, stratified by the number of prior episodes. Each histogram illustrates the distribution of inter‐episode intervals within each stratum.

Supplemental Figure S2: Cumulative incidence function (CIF) of pancreatitis recurrence, stratified by the number of prior recurrent acute pancreatitis (RAP) episodes using an alternative definition of RAP based on a 30‐day inter‐episode interval (vs. 90 days in the primary analysis). Each curve represents the estimated risk of recurrence over time for patients with a specific number of prior episodes, accounting for competing events such as chronic pancreatitis, pancreatic cancer, and death.

Supplemental Figure S3: Weibull models of the time intervals between pancreatitis episodes, stratified by the number of prior recurrent acute pancreatitis (RAP) episodes, using an alternative definition of recurrent acute pancreatitis based on a 30‐day inter‐episode interval (vs. 90 days in the primary analysis). Models demonstrate a ‘wear‐out’ pattern (shape parameter > 1) that increases progressively (indicated by increasing shape parameter values) with accumulating RAP events. This indicates an accelerating risk of pancreatitis recurrence and progressively shorter intervals between episodes as the number of prior RAP events increases.

Supplemental Figure S4: Swimmer plot depicting the disease course of 50 patients with incident acute pancreatitis using an alternative definition of recurrent acute pancreatitis (RAP) based on a 30‐day inter‐episode interval (vs. 90 days in the primary analysis). Patients were randomly selected from each stratum defined by the number of prior RAP episodes. Each horizontal line represents a single patient, with markers indicating recurrent AP events and competing events (chronic pancreatitis, pancreatic cancer, or death) over time. AP; acute pancreatitis, CP; chronic pancreatitis, PDAC; pancreatic ductal adenocarcinoma.

Supplemental Figure S5: Cumulative incidence function (CIF) of pancreatitis recurrence in patients with alcohol abuse (n = 1545), stratified by the number of prior recurrent acute pancreatitis (RAP) episodes. Each curve represents the estimated risk of recurrence over time for patients with a specific number of prior episodes, accounting for competing events such as chronic pancreatitis, pancreatic cancer, and death.

Supplemental Figure S6: Weibull models of the time intervals between pancreatitis episodes in patients with alcohol abuse (n = 1545), stratified by the number of prior recurrent acute pancreatitis (RAP) episodes. Models demonstrate a ‘wear‐out’ pattern (shape parameter > 1) that increases progressively (indicated by increasing shape parameter values) with accumulating RAP events. This indicates an accelerating risk of pancreatitis recurrence and progressively shorter intervals between episodes as the number of prior RAP events increases.

Supplemental Figure S7: Cumulative incidence function (CIF) of pancreatitis recurrence in patients with prior cholecystectomy (n = 456), stratified by the number of prior recurrent acute pancreatitis (RAP) episodes. Each curve represents the estimated risk of recurrence over time for patients with a specific number of prior episodes, accounting for competing events such as chronic pancreatitis, pancreatic cancer, and death.

Supplemental Figure S8: Weibull models of the time intervals between pancreatitis episodes in patients with prior cholecystectomy (n = 456), stratified by the number of prior recurrent acute pancreatitis (RAP) episodes. Models demonstrate a ‘wear‐out’ pattern (shape parameter > 1) that increases progressively (indicated by increasing shape parameter values) with accumulating RAP events. This indicates an accelerating risk of pancreatitis recurrence and progressively shorter intervals between episodes as the number of prior RAP events increases.

Supplemental Table S1: Charlson comorbidity index presented with score and ICD‐10 codes for each comorbidity.

Supplemental Table S2: ICD‐10 and ATC codes used for the definition of alcohol abuse and heavy smoking.

Supplemental Table S3: Cumulative incidence of pancreatitis recurrence at 3, 6, 9, 12 and 24 months stratified by number of previous recurrent acute pancreatitis (RAP) episodes. A 30‐day lag period between RAP episodes was applied in this sensitivity analysis.

Olesen S. S., Bruun N. H., Davidsen L., et al., “Natural History of Recurrent Acute Pancreatitis: Implications for Interventional Trial Design,” Alimentary Pharmacology & Therapeutics 64, no. 7 (2026): 976–984, 10.1111/apt.70820.

Data Availability Statement

Research data are not shared.