In this double-center retrospective study, we determined the threshold of 112.5 of admission mPASS in predicting severe AP and constructed a combined model based on CECT radiomics and WBC, providing a new simple method for predicting the activity of acute pancreatitis. The AUC values of the combined model for predicting the activity of AP were 0.84, 0.77 and 0.80 in the training, validation and test group, respectively. This study is the first attempt to predict the activity of AP using CECT radiomics and laboratory parameters.
The AUC of admission mPASS scores in predicting severe AP was 0.90 in the main cohort, which was similar to a previous study [8]. In the present study, the length of stay, severity and complication rate of AP patients with high level of disease activity at admission were higher than those of AP patients with low level of disease activity (p < .05), which showed no difference with previous studies [4]. The follow-up results showed that patients with high level of disease activity at admission were more likely to show imaging findings and clinical manifestations with no alleviation. These results verified the accuracy of mPASS > 112.5 at admission in predicting important clinical outcomes and imaging with no remission.
In the present study, significant differences were found in the APACHE II and BISAP between the high and low level of disease activity groups, which may be because of the overlapping components of severity and mPASS, such as organ failure and SIRS. WBC count was the only independent risk factor to predict the activity of AP. Previous research has shown that trypsin activation and neutrophil infiltration are mutually reinforcing [20]. WBC is associated with SIRS, and their toxic mediators can cause tissue injury [21]. Although ALB levels have been used to predict organ failure and high CRP levels are associated with severe AP [22, 23], additional researches are required to judge whether ALB and CRP can be used for predicting the activity of AP.
The definition of disease activity is reversible manifestations during the course while severity is a fixed state or outcome[3]. In the present study, there were significant differences in the CTSI and EPIC scores between the high and low level of disease activity groups, which suggested that imaging may be able to identify an association between the severity and activity. However, conventional imaging performance often lags behind disease progression. Minimal activity changes may cause severe fluctuations in radiomics scores, while conventional imaging scoring systems may change a little or not at all.
Radiomics plays a vital role in the study of AP including predicting the severity of AP, the recurrence of AP and the development of peripancreatic necrosis [11, 24, 25]. Recently, Zhao Y et al [26] emphasized the application of radiomics in AP severity, but studies have not reported whether CECT radiomics can be used to predict AP activity. In the present study, the CECT radiomics features showed the ability to predict the activity of AP. Compared to CTSI and EPIC, radiomics model had a superior value. It suggested that radiomics features revealed some of the differences between high and low level of disease activity, which was possibly due to some small morphological changes in pancreatic parenchyma caused by hemoconcentration, decreased blood flow, decreased tissue oxygenation decreased, vasoconstriction or increased permeability [27, 28].
The effectiveness of the radiomics model was improved by incorporating WBC count. The compositions of the combined model were easily acquired from CECT and routine blood tests, thus avoiding additional medical tests. The follow-up results showed that patients with high total points based on the nomogram were more likely to have a high risk of high activity level at admission and show an increasing trend in imaging and clinical performance, which may be associated with the development of irreversible liquefaction and necrosis from transient ischemia and hypoxia of pancreatic parenchyma [29]. The follow-up results showed that the ability of the combined model in predicting the clinical and imaging with no remission is comparable to the admission mPASS, and confirmed the potential clinical application value of the combined model as a quantitative instrument in predicting the activity and prognosis of AP.
The model in the present study showed a stable performance. The performance in the test group from another tertiary referral center showed that the model had a certain generalization ability. Resampling is used to diminish the unsatisfactory reproducibility of radiomics features. The radiomics features are linearly separable, and multiple logistic regression has a good stability, direct calculation process and good performance.
There were several limitations in the present study. First, more clinical characteristics and laboratory parameters that may be highly associated with AP activity must be considered in a comprehensive evaluation. However, conventional laboratory parameters related to AP severity were used in the present study to avoid additional costs. Second, although 305 cases are sufficient for radiomics, more cases were needed to be brought in for increasing universality and credibility. Third, there may be some confounding factors and bias in this retrospective study. We have taken some measures such as developing specific criteria and using external validation from institution 2 to ensure the reliability of the model.
In conclusion, the mPASS score of 112.5 at admission is a meaningful threshold in predicting severe AP, and the radiomics features may reflect the differences between AP with high and low level of disease activity, which are hidden in the pancreatic parenchyma. WBC may reflect the differences between inflammatory processes of AP with distinct activity. The combined model based on CECT radiomics provided a new method to predict the activity of AP, and had a potential value in predicting clinical or imaging with no remission. It expected to be a promising tool for predicting the activity and prognosis of AP, which could contribute to further management.