Our review followed the Cochrane Handbook for Systematic Reviews (16) and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guideline (Supplementary Table S1) (17). The study protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO, CRD42021281983).

Our study aimed to investigate risk factors for developing PD and DM following AP, via analyzing all factors assessed during the hospitalization with AP, that were compared between new-onset PD or DM and normal glucose regulation groups. To establish the eligibility criteria, we used the PECOTS framework.

Population (P): adult AP patients without confirmed DM at discharge. Exposure and comparator (E): any factor assessed at the time of hospitalization with AP and (C) its control group, such as severe vs. non-severe AP, necrosis vs. absence of necrosis, smoking vs. not smoking, male vs. female. Classification of AP severity has changed over the years. Our study group’s data in two ways firstly comparing severe AP (SAP) vs. moderately severe and mild AP as one group and alternatively comparing SAP and moderately severe AP as one group vs. mild AP. Some studies applied classification criteria with only two categories: severe and non-severe AP. These studies defined SAP based on 1992 Atlanta criteria (18–21), Scoring ≥8 on APACHE II (22), ≥3 Ranson score (23), and ≥ 2 Japanese severity score (24). We analyzed the findings of these studies using the categories of SAP vs. moderate and mild AP as one group.

Outcome (O): Number of AP patients who developed, after hospital discharge: DM or PD (impaired fasting glucose, impaired glucose tolerance, and HbA1c ≥5.7 and < 6.5%) as reported by the study authors. Multiple studies provided the number of patients who developed PD or DM combined; we included this composite outcome in our analyses as PD/DM. In case of studies providing incomplete or no definition for glycemic outcomes or not stating explicitly that preexisting DM was excluded from the cohort, this uncertainty was taken into account during the risk of bias assessment.

Timing (T): Initially, we planned to include studies assessing the outcome at least 3 months after hospital discharge. However, we decided to deviate and include all studies that reported on the relevant outcomes after hospital discharge because of the limited and heterogeneous data on follow-up and diagnosis time intervals.

Study design (S): The analysis included interventional and observational studies that met the criteria of our review’s PECO framework. Case reports, case series, and studies with less than 10 participants per outcome group or less than 10 participants in the exposed or comparator group were excluded. Conference abstracts were retained.

The systematic search was carried out in three databases: MEDLINE (via PubMed), Embase, and Cochrane Central Register of Controlled Trials (CENTRAL) from database inception to September 04, 2023 without any filters or restrictions. The main concepts in the search strategy were prediabetes, diabetes, acute, and pancreatitis. See Supplementary Table S2 for the detailed search key and selection process.

Data collection process is detailed in Supplementary Table S2. Data on the following variables were collected when available: country, year of publication, study period, follow-up time, name and the number of centers, study design, sample size, age, sex and weight of participants, inclusion and exclusion criteria of participants, classification of AP severity, outcome domains reported and their assessment method, and risk factors during the initial AP episode and their definitions. For a complete list of the risk factors investigated in relation to new-onset PD, DM, or PD/DM by the included studies, see Supplementary Table S3.

We calculated odds ratios (ORs) with 95% confidence intervals (CI). Refer to Supplementary Table S4 for detailed description.

Two independent reviewers (OZ and AK) assessed each study for risk of bias using the Quality In Prognosis Studies (QUIPS) tool (25). Disagreements were resolved by discussion until reaching a consensus. Risk of bias analyses were conducted for each outcome and prognostic factor separately. To simplify and ease the interpretation of these results, three summary Risk of bias assessments were created for the three main outcomes (DM, PD, and PD/DM), taking into account the worst possible scenario for each study and each domain.

To assess the possibility of publication bias (small study effect), we created and visually assessed funnel plots for every analysis where at least six studies were included. Harbord modified Egger’s test was performed in the case of 10 or more included studies (26), with a p < 0.1 indicating statistical significance for funnel plot asymmetry.

The systematic search yielded 14,977 results (Figure 1). Overall, 61 studies with 85 reporting articles were eligible for inclusion. The meta-analysis encompassed 50 studies and 76,797 patients.

Key study characteristics are summarized in Table 1. Approximately 68% of the studies were based on the general AP population, four included only SAP patients, six focused on necrotizing AP patients and in 10 studies participants were superselected for other criteria. The outcome was reported as PD, DM, and PD/DM in 6, 43, and 22 studies, respectively. A total of 79 prognostic factors were reported on by at least one study and the unique combinations of prognostic factors and outcomes numbered 137 different comparisons. Meta-analysis was possible in the case of 34 risk factor and outcome pairs.

Our findings of the 34 meta-analyses are summarized in an aggregated forest plot, which shows the pooled OR for each risk factor and outcome pair (Figure 2). In addition, per risk factor groups we present the original forest plots or more detailed aggregated forest plots. All other individual plots can be found in the Supplementary material.

Overall, the proportion of the high risk of bias studies was notable (32–44%) for all three outcome factors (Supplementary Figures S23–S25). This was primarily due to a lack of reporting on study attrition and suboptimal definitions of outcome measurements.

High heterogeneity was noted in several of our analyses. Subgroup analysis for follow-up length significantly reduced heterogeneity only for new-onset DM in relation to alcoholic etiology. For the other prognostic factors, heterogeneity remained high even after accounting for follow-up time.

Of the 34 risk factor and outcome pairs that could be meta-analyzed, sensitivity analysis was feasible in the case of 14 analyses (Supplementary Figures S26–S34). Leave-one-out analysis identified one study (24), whose omission would make a significant difference, which we reported in paragraph 3.3.1.

Publication bias assessment was limited to six meta-analyses on new-onset DM: severe AP, moderately severe and severe AP, alcoholic and biliary etiology, recurrent AP, and male sex (Supplementary Figures S35–S40). Possible small study publication bias was detected in the case of alcoholic etiology in relation to DM development based on Egger’s test and visual inspection of the funnel plot.

Past meta-analyses applying indirect comparisons found conflicting results regarding the association of AP severity and new-onset PD or DM (7, 8). Our analysis of direct, within-study comparisons confirms a positive relationship between SAP, moderately severe AP, and new-onset DM. Classification of AP severity is based on the development of local complications (such as necrosis) and organ failure (1). Beta cell death secondary to local complications of AP is believed to be one of the possible mechanisms behind the ensuing DM (82). Our meta-analysis supports this hypothesis as necrosis was associated with significantly greater risk of developing DM and PD/DM. Moreover, patients with local complications might require interventions such as pancreatic debridement, lavage, drainage, necrosectomy, and partial pancreatectomy, during which further pancreatic tissue is lost (83).

Nevertheless, cell death is only one aspect of the complex pathomechanism of post-AP DM. It was proposed that the inflammation accompanying AP stimulates endogenous beta-cell proteins to undergo post-translational modifications (84). Such modified proteins could trigger autoimmune processes as seen in type 1 diabetes (85), which could explain the earlier and greater need for insulin therapy seen with post-pancreatitis DM compared to type 2 DM (12). The level of inflammatory cytokines correlate well with persistent organ failure, which is the hallmark of SAP (86). Our study found an increase in the odds of developing DM and PD/DM after AP with organ failure, albeit not-statistically significant. It is noteworthy that none of the included studies specified the duration of the organ failure, and few mentioned the affected organs or the number of organs affected.

We found a more pronounced association with severity in the case of PD/DM than with DM suggesting an even more substantial influence of AP severity on the development of PD. Moreover, comparing severe and moderate AP as one group vs. mild yielded a higher odds ratio than the comparison of SAP to moderate and mild AP as one group. This could imply that progresses from mild to moderate AP severity has a greater impact on PD and DM development compared to the step from moderate to severe disease progression.

We found that alcoholic AP was associated with an increased risk of developing DM. Alcohol has a toxic effect on the pancreas. Its metabolites elicit sustained intracellular calcium overload, which disrupts beta-cell functioning and insulin secretion while also leading to oxidative stress (87), to which beta-cells are especially vulnerable due to their low antioxidation capacity (88).

The most likely explanation for the tendency seen with hypertriglyceridemic etiology is that hypertriglyceridemia itself is associated with DM (89). The two conditions often coexist in metabolic syndrome. Analysis of the Hungarian Study Group’s registry data shows that 69% of the non-diabetic hypertriglyceridemic AP patients present with at least two factors of the metabolic syndrome on admission and they are at an increased risk of developing post-AP DM (90). Therefore, the development of DM might be a natural progression of the disease, possibly quickened by the AP episode.

Acute pancreatitis tends to be more severe if caused by excessive alcohol consumption or hypertriglyceridemia (91) and if metabolic syndrome is present (90). Toxic factors (e.g., alcohol and fatty acids) play a role in the development and severity of pancreatitis when they accumulate (92). This aligns with the multiple hits theory of AP severity documented for smoking, drinking (93), obesity, hypertension, and hyperlipidemia (90). The risk factors we identified—local complications, severity, alcoholic, and hypertriglyceridemic AP—often coexist (91, 94, 95). Suggesting that the development of post-AP DM might work on a similar multiple hits theory basis.

Both alcoholic and hypertriglyceridemic etiologies are linked to poor dietary habits that are difficult to change and the ongoing exposure conveys a high risk for RAP, progression of the disease, and development of complications (96). On the contrary, the recurrence of biliary AP is often prevented by cholecystectomy after the index episode (64). Without identifying a treatable or preventable etiology, there is a risk for RAP. However, we found no association between idiopathic AP and DM development, possibly due to the control group—containing alcoholic and hypertriglyceridemic etiology—demonstrating a positive association with DM.

Recurrent acute pancreatitis conveys repeated pancreatic inflammation and cellular insult or loss, leading to an assumed association with developing pancreatic endocrine dysfunction (96). Our analysis found a tendency of increased odds for new-onset DM and PD/DM with RAP, which neared statistical significance. It should be pointed out that there was considerable heterogeneity in the study designs. Some studies excluded patients presenting with RAP at the index AP episode while others included them. Importantly, those who had RAP and developed DM by the index AP episode were excluded from the analysis based on the premise of pre-existing DM. Moreover, 60% of the analyzed studies had a relatively short follow-up of less than 3 years. Finally, different distributions of the etiological factors among the included studies might influence the observed association between RAP and DM or PD/DM, as alcoholic and hypertriglyceridemic APs are associated with a greater risk of RAP (91). All four factors could influence the true relationship between disease recurrence and PD/DM development.

Our study found that obesity was associated with a significantly greater risk of new-onset DM. Some of the other risk factors we identified for new-onset DM after AP (hypertriglyceridemic AP, AP-related complications, and SAP) tend to occur more frequently in obese individuals (90, 97). Moreover, excess weight is a known independent risk factor for type 2 DM. Therefore, it could be a natural progression of the disease or AP might even trigger DM in genetically or metabolically predisposed patients (7). At present, there is still a lack of consensus on differentiating type 2 DM from post-AP DM in patients who had an AP episode (7, 98). Most studies define post-AP DM as new-onset of hyperglycemia (using the standard cutoff values for DM as per the World Health Organization or American Diabetes Association recommendations) following an AP episode (12, 13, 99). The prospective, multi-center DREAM study (Diabetes RElated to Acute Pancreatitis and Its Mechanisms) was recently designed to characterize the DM phenotypes after AP and their pathomechanism (100, 101).

We found no association between sex, smoking, alcohol consumption, and DM or PD/DM. The lack of association with alcohol consumption is paradoxical in light of the increased risk of DM with AP of alcoholic etiology. However, the included studies mostly compared alcohol consumption to not-drinking, not taking into account the amount and duration of alcohol consumption. Studies with follow-up length over 4 years were more likely to favor an association between alcohol consumption and new-onset DM or PD/DM (29, 64, 67) compared to shorter studies (31, 36, 37). Also, the analysis included only four and three studies for DM and PD/DM, respectively, with two of the studies containing unusually low proportions of alcoholic etiology (4 and 14%) and three studies including only patients with AP of alcoholic etiology.

Additionally, we observed a clinically relevant odds ratio for post-AP DM with liver disease and dyslipidemia. We believe that statistical significance was not achieved due to the low number of studies investigating these risk factors and their heterogeneous nature. In our analysis, chronic kidney disease was associated with a significantly higher risk of post-AP DM. It is notable that the analysis was based on three studies, of which Firkins et al. (43) accounted for 99.4% of the pooled results due to the large sample size. This is a retrospective nationwide database analysis, where only patients with a second hospital admission within one calendar year were included. Patients with chronic kidney disease are admitted more frequently to hospitals (102); thus, they were likely over-represented in the study by Firkins et al. (43).

Timely translation of scientific data to clinical practice has crucial importance in healthcare (103, 104). Long-term complications of AP (exocrine and endocrine insufficiency) were documented as early as 1941 (105); nonetheless, the Chinese guideline in 2021 was the first to recommend follow-up visits after AP (59). They recommend that all AP patients should be monitored after rehabilitation, however, for different lengths of time depending on severity. They rated the strength of recommendation and supporting evidence weak.

While all AP patients should be followed up for the development of long-term complications after AP, financial and human resources are often limited in healthcare. Our study highlights the sub-populations of AP patients who are at a higher risk for developing PD or DM. Therefore, more frequent follow-ups of these patients increase the likelihood of preventing and reducing post-AP diabetes-related morbidity, mortality, and healthcare costs.

Due to the broad search strategy and lack of constraints on the results, this is the first comprehensive systematic analysis of potential risk factors for new-onset PD/DM following AP, with the largest number of included studies (50 in total) covering 76,797 participants in the meta-analysis. Our study was based on direct, within-study comparisons; therefore, it is more representative of the true effect of risk factors compared to previous meta-analyses (7, 8). Due to the inclusive nature of our research, there was substantial heterogeneity between the studies, which we attempted to reduce by performing separate analyses for PD/DM and DM and conducting subgroup analysis for follow-up length. Almost a third of the meta-analyses were based on three studies. For these risk factor and outcome pairs, conclusions should be cautiously handled.

All patients require medical follow-up for endocrine and exocrine insufficiency after AP. Our results show that patients who have suffered severe or moderately severe AP, alcoholic or hypertriglyceridemic AP, develop pancreatic necrosis or organ failure, had multiple AP episodes, are obese or have pre-existing chronic kidney disease, liver disease or dyslipidemia are at a greater risk for developing PD or DM. Therefore, closer monitoring is warranted in these high-risk groups.

Further long-term follow-up studies of AP patients are needed to observe morbidity and mortality following single and multiple AP episodes as well. High-quality well-controlled observational studies with long follow-up duration are needed to establish an evidence-based follow-up schedule after AP to help identify patients early in a prediabetic state, where interventions could still prevent DM. Future studies should also explore interventions for preventing post-pancreatitis DM. In 2022 the Hungarian Pancreatic Study Group launched two longitudinal randomized controlled trials on dietary intervention (106) and smoking and alcohol cessation following hospitalization for AP (107).

We found that AP severity, alcoholic and hypertriglyceridemic etiologies, pancreatic necrosis, organ failure, RAP and comorbidities of obesity, chronic kidney disease, liver disease, and dyslipidemia are associated with a higher risk of developing PD or DM following AP. Glucose homeostasis should be regularly monitored in high-risk populations after hospital discharge. Further research is needed to establish an appropriate follow-up schedule and interventions for preventing DM after AP.