Background/Objectives: Acute pancreatitis (AP) increases the risk for development of diabetes mellitus (DM); Pre-DM/DM have significant comorbidites, making early recognition through prediction models critical. There are no prediction models for the development of pre-DM/DM following a single episode of AP in children.
Methods: This was a prospective observational cohort study that enrolled patients under age 21 with an index admission for AP and a 3 and 12 months follow up period. Clinical laboratory values, imaging findings (CT or MRI performed within 90 days of AP diagnosis) course of admission, and chemokine and cytokine measures performed on plasma collected at index AP were tested for association with DM development. A multivariable regression model to predict development of pre-DM/DM following an index AP was derived from this combination of clinical, biomarker and imaging data sets by including all significant univariate predictors (p<0.2) and randomly splitting data into training and test cohort in an 8:2 fashion.
Results: From a total of 187 index AP subjects enrolled, 137 (73%) and 144 (77%) underwent DM screening at 3 and 12 months respectively. 137 (73%) had imaging available. Of those with 3 month follow up, 22/137 [NK1] (16%) had pre-DM/DM (pre DM n=21, DM n=1). Of those with 12 month follow up, 23/144 (16%) patients had preDM/DM, (PreDM n=18, DM n=5). Pre-DM/DM at any time point occurred in 40/187[NK2] (21%) of subjects (PreDM n=35, DM n=5).
Univariate analysis of associations with preDM/DM at 12 months showed a higher frequency of severe AP (SAP[AM3] [AM4] ), present in 52% with preDM/DM vs. 17% no DM (p<0.05), and higher median [IQR] levels of IL-6 (910 [618-3438] vs. 196 [71-480], p<0.05), and CRP (4.16 [1.67-10.7] vs. 1.55 [0.4-3.68], p=0.1) in patients with preDM/DM. Age (p=0.49), BMI (p=0.69), and recurrent AP (ARP) (p=0.8) had no significant association with preDM/DM. Further psoas muscle area (PSMA)[AM5] z-score (p=0.17) and subcutaneous fat area standardized to height (p=0.3) had no significant association with preDM/DM.[PD6]
The optimal multivariable model for prediction of preDM/DM at 12 months based on clinical variables included SAP, CRP, IL-6 and age, with an area under the curve (AUC) of 0.8 in both the training and test data set[GG7] s. Adding imaging variables, the model included SAP, CRP, IL-6 and subcuteanous fat area standardized to height and had a training AUC of 0.83, and test AUC 0.79.
Conclusions: Development of pre-DM and DM following an index AP episode can be predicted by attack severity, CRP and IL6 levels along with imaging markers. Future studies to validate this model through multi-center and larger cohort designs are needed.

Figure 1: Area under the receive operater curves (AUROC) of multivariable prediction model for preDM/DM with curves of the univariate variables: Panel A: Multivariable model including: SAP, CRP, IL-6 and age. AUROC of predictive model is 0.8. Panel B: Multivariable model including radiomics: SAP, CRP, IL-6 and most proximal subcutaneous fat area standardized to height (Sub-Q fat) with final model AUROC 0.8. Both AUROC of predictive model better than the AUROC of each individual parameter alone.

<b>Table 1:</b> Univariate analysis of chemokines, cytokines and radiomics at index AP episode and associated with preDM/DM development at 12 month follow up. Highlighted values have p < 0.2 and were included in optmimal multiviariable progression model.

Table 1: Univariate analysis of chemokines, cytokines and radiomics at index AP episode and associated with preDM/DM development at 12 month follow up. Highlighted values have p < 0.2 and were included in optmimal multiviariable progression model.

Pradipta Debnath

Appearances