Risk factors based prediction model for new-onset diabetes mellitus after first-attack acute pancreatitis

Abstract To construct a practical prediction model for the risk of new-onset diabetes mellitus (DM) in patients with first-attack acute pancreatitis (AP) based on risk factors derived from demographic and clinical data. A total of 780 patients diagnosed with AP were retrospectively enrolled in the Second Affiliated Hospital of Nanchang University from June 2016 to June 2017. A series of exclusion criteria were applied and 396 patients were finally included. With a ratio of 2:1, patients were randomly divided into two groups named training (n = 264) and validation set (n = 132). Demographic and clinical data that may be risk factors of new-onset DM after first-attack AP were collected. Univariate and multivariate analyses were used to determine potential risk factors in the training set, and a predictive nomogram was constructed. Nomogram performance was determined in the training and validation sets concerning discrimination and calibration capabilities. Finally, clinical applicability of the nomogram was assessed in the validation set by decision curve analysis. The morbidity rate of new-onset DM after first-attack AP was 8.6% (34/396) in the included patient cohort. Hyperlipemia (OR = 6.87, 95%CI = 2.33 ~ 20.26, p = 0.000), GGT ≥ 40U/L (OR = 0.07, 95%CI = 0.03 ~ 0.27, p = 0.008), serum glucose ≥ 6.1mmol/L (OR = 7.73, 95%CI = 1.89 ~ 31.64, p = 0.004), CT grade ≥ 2 or 4 points (OR = 3.16 or 4.95, 95%CI = 1.05 ~ 9.45 or 1.12 ~ 21.89, p = 0.039 or 0.035) and APACHE II grade ≥ 8 points (OR = 3.82, 95%CI = 1.19 ~ 12.27, p = 0.024) were independent risk or protective factors and were assembled for nomogram construction. Internal and external validations showed good discrimination (Area under the receiver operating characteristic curve = 0.884 and 0.770) and calibration capabilities. The decision curve analysis showed good clinical applicability. We have developed a practical nomogram to predict the risk of new-onset DM after first-attack AP based on risk factors derived from demographic and clinical data, which would contribute to the identification and management of these high-risk patients.