Establishment of Early Multi-Indicator Prediction Models of Moderately Severe Acute Pancreatitis and Severe Acute Pancreatitis

Background It is critical to accurately identify patients with severe acute pancreatitis (SAP) and moderately SAP (MSAP) in a timely manner. The study was done to establish two early multi-indicator prediction models of MSAP and SAP. Methods Clinical data of 469 patients with acute pancreatitis (AP) between 2015 and 2020, at the First Affiliated Hospital of Fujian Medical University, and between 2012 and 2020, at the Affiliated Union Hospital of Fujian Medical University, were retrospectively analyzed. The unweighted predictive score (unwScore) and weighted predictive score (wScore) for MSAP and SAP were derived using logistic regression analysis and were compared with four existing systems using receiver operating characteristic curves. Results Seven prognostic indicators were selected for incorporation into models, including white blood cell count, lactate dehydrogenase, C-reactive protein, triglyceride, D-dimer, serum potassium, and serum calcium. The cut-offs of the unwScore and wScore for predicting severity were set as 3 points and 0.513 points, respectively. The unwScore (AUC = 0.854) and wScore (AUC = 0.837) were superior to the acute physiology and chronic health evaluation II score (AUC = 0.526), the bedside index for severity in AP score (AUC = 0.766), and the Ranson score (AUC = 0.693) in predicting MSAP and SAP, which were equivalent to the modified computed tomography severity index score (AUC = 0.823). Conclusions The unwScore and wScore have good predictive value for MSAP and SAP, which could provide a valuable clinical reference for management and treatment.

In 1974, Ranson et al. [14] assessed the severity of AP for the first time. Since then, a series of clinical scoring systems have been gradually developed to stratify the severity and predict the prognosis of AP [15][16][17], including the Ranson score, the Glasgow score, the acute physiology and chronic health evaluation (APACHE) II score, the bedside index for severity in AP (BISAP), and the modified computed tomography severity index (MCTSI) [9,[17][18][19]. Nevertheless, both the Ranson score and the Glasgow score require 48 h to be determined. Therefore, the risk stratification cannot be performed at admission, leading to missing a potentially valuable and early therapeutic window. Furthermore, some parameters used in the Ranson score are not routinely collected at the time of hospitalization [9,20,21]. The APACHE II score has complex parameters, tedious operation, and poor predictive value within 24 h of onset [9,[21][22][23][24]. Wu et al. [21] pointed out that in their validation cohort, only 2.2% of cases had complete APACHE II data. Moreover, whether the APACHE II score reflects the presence of tissue necrosis remains a matter of debate [6]. Unfortunately, pancreatic parenchyma necrosis may not occur in the early stage of the disease, which restricts the early assessment of the severity of AP by MCTSI [10]. In contrast, the BISAP score, which can well predict the severity of AP, organ failure (OF), and death, is easier to calculate and can be evaluated by routine clinical data within 24 h [9,21]. However, it has also been reported that its sensitivity to mortality and SAP is not ideal [25].
Early prediction of the severity of AP remains a huge challenge. In view of the shortcomings of current risk stratification tools, the purpose of this study is to explore the prognostic factors related to the severity of AP and to establish early multi-indicator prediction models, in order to provide a reference for clinical diagnosis and treatment of AP, early intervention, and hierarchical management.

Object of Study.
Clinical data of 469 patients who met the inclusion criteria between January 1st, 2015, and June 30th, 2020, at the First Affiliated Hospital of Fujian Medical University, and between January 1st, 2012, and October 31st, 2020, at the Affiliated Union Hospital of Fujian Medical University, were retrospectively collected and analyzed, including 202 cases of mild AP (MAP), 240 cases of MSAP, and 27 cases of SAP.

Diagnostic Criteria and the Revised Atlanta
Classification of AP [26]. According to the 2012 revision of the Atlanta classification and definitions by international consensus, the diagnosis of AP requires two of the following three features: (i) abdominal pain consistent with AP (acute onset of a persistent, severe, epigastric pain often radiating to the back); (ii) serum lipase activity (or amylase activity) at least three times greater than the upper limit of normal; and (iii) characteristic findings of AP on contrast-enhanced CT and less commonly magnetic resonance imaging or transabdominal ultrasonography.
(1) MAP: MAP is characterized by the absence of local or systemic complications and OF.
(2) MSAP: MSAP is characterized by OF that resolves within 48 h (transient OF) and/or local or systemic complications without persistent OF.
(3) SAP: SAP is characterized by persistent OF (>48 h), including single OF and multiple OF.
OF is defined as a modified Marshall score ≥ 2 for the renal, respiratory, and/or cardiovascular system. Local complications are AP fluid collection, pancreatic pseudocyst, acute necrotic collection, walled-off necrosis, gastric outlet dysfunction, splenic and portal vein thrombosis, and colonic necrosis. Systemic complications are systemic inflammatory response syndrome (SIRS), OF, sepsis, intra-abdominal hypertension, abdominal compartment syndrome, and pancreatic encephalopathy.

Inclusion and Exclusion
Criteria. Inclusion criteria were as follows: (1) patients who met the diagnostic criteria of AP, and had complete clinical data, (2) patients who were more than 18 years old, (3) patients in whom systemic examination and abdominal CT scanning were completed, and (4) initial onset.
Exclusion criteria were as follows: (1) patients who were less than 18 years old, (2) acute attack or recurrence of chronic pancreatitis, (3) pregnant or lactating patients, (4) AP caused by malignant tumor or abdominal space occupying lesion, (5) patients who had severe mental/neurological disorders or lack of self-awareness, and (6) whose clinical data were incomplete.

Outcome
Measurements. Secondary outcome measurements are as follows: the patient data were subjected to univariate analysis and binary logistic regression analysis to obtain independent prognostic indicators of the severity of AP.
Primary outcome measurements are as follows: using the above indicators, unweighted predictive score (unwScore) and weighted predictive score (wScore) for MSAP and SAP were established. The receiver operator characteristic 2 Gastroenterology Research and Practice (ROC) curves of independent prognostic factors, 4 existing systems, and prediction models were produced. The area under the curves (AUC) were compared while the cutoff values, the sensitivity, the specificity, the positive predictive values (PPV), the negative predictive values (NPV), and the accuracy rates were calculated to verify the predictive efficiency of new models.
2.6. Statistical Analysis. SPSS 24 software and MedCalc v19.3.0 software were used for statistical analysis, and Graph-Pad Prism 8 was used for drawing. The chi-square test or the Fisher exact test was used for countable data. Normally distributed data are expressed as the mean ± standard deviation ( x ± S), and the independent sample t-test was used for the comparison between groups. Nonnormally distributed data are expressed as the median (quartile spacing), and the Mann-Whitney rank sum test was used for comparison between groups. Binary logistic regression analysis was used to analyze the influencing factors. P < 0:05 was considered to indicate statistical significance. ROC curves were used to determine the best critical value of each prognostic index and prediction model and their prediction values.

Results of Univariate Analysis of Influencing Factors in the MAP and MSAP+SAP Groups
3.1.1. Comparison of General Data between the MAP and MSAP+SAP Groups. There was no significant difference in sex, vital signs (systolic blood pressure, diastolic blood pressure, and mean arterial pressure), APACHE II score, and hypertension between the two groups. There was a significant difference in age, etiology, vital signs (temperature, pulse rate, and respiratory rate), severity score (BISAP score, MCTSI score, and Ranson score), underlying diseases (diabetes, fatty liver, and hyperlipidemia), hospitalization days, and hospitalization expenses (P < 0:05) ( Table 1).

Comparison of Clinical Indexes between the MAP and MSAP+SAP Groups.
There was no significant difference in PLT, RDW/PLT, TBIL, BUN, sCr, and BUN/sCr between the two groups (P > 0:05). There was a significant difference in WBC, HB, HCT, RDW, GLU, AST, LDH, ALB, TCHOL, TG, D-dimer, HCO 3 − , CRP, PCT, serum calcium, corrected  , CRP, PCT, serum calcium, corrected calcium, serum phosphorus, serum sodium, and serum potassium were included in the binary logistic regression analysis. The results showed that WBC, LDH, CRP, sCr, TG, D-dimer, and serum potassium were independent prognostic risk factors for MSAP and SAP, while serum calcium was an independent prognostic protective factor for MSAP and SAP (Table 3).

Establishment of Multi-Index Joint
Prediction Models  Figures 1 and 2), which were included in the unwScore model and the wScore model.
The unwScore model was superior to the APACHE II score, the BISAP score, and the Ranson score in predicting the occurrence of MSAP and SAP in patients with AP, which was similar to that of MCTSI (Table 5, Figure 3).
The wScore model was superior to the APACHE II score, the BISAP score, and the Ranson score in predicting the occurrence of MSAP+SAP in patients with AP, which was similar to the MCTSI score and the unwScore model (Table 5, Figure 3).

Discussion
AP is an inflammatory disease of highly variable severity, ranging from mild cases with low mortality to severe cases with high mortality [27]. Early severity stratification becomes critical, especially on the day of admission, as this period is regarded as a window of opportunity to prevent pancreatic necrosis and OF [9]. Currently, there are four frequently used AP scoring systems to assist clinicians to iden-tify SAP and MSAP, namely, the APACHE II score, the Ranson score, the BISAP score, and the MCTSI score. Each scoring system has specific applications and advantages, but each also has limitations.
In the present study, we constructed early multiindicator prediction models that comprise only seven prognostic indexes: WBC, LDH, CRP, TG, D-dimer, serum potassium, and serum calcium. Compared with the traditional scoring systems, the advantages of the unwScore and wScore models constructed in this study lie in their simplicity, safety, objectivity, low cost, early risk stratification, high predictive value, dynamic monitoring, and quantitative prediction.
In 1974, Ranson et al. [14] first put forward the Ranson score, which is a milestone in the assessment of AP severity. However, the Ranson score is based on 11 parameters identified as important prognostic factors, requiring 48 hours for a complete evaluation [28]. Similarly, it is only applicable within 48 hours of admission, and it is difficult to track and evaluate the changes of the disease in real time [29]. Different from the Ranson score, the unwScore and wScore models could be used simultaneously by using only seven variables within 24 hours after admission. Similarly, the unwScore model could also quantitatively predict the occurrence probability of MSAP and SAP. With the increasing model score, the probability of occurrence of MSAP and SAP increased.
In 2004, Mortele et al. [30] revised and simplified the CTSI score and put forward the MCTSI score, which correlated more closely with the severity of AP and the following parameters: the length of the hospital stay, the need for surgical or percutaneous procedures, and the occurrence of infection. However, the ideal time for CT scans is at least 72 hours after the onset of symptoms [31], and it is not recommended to conduct routine CT scans on admission purely for the purpose of severity assessment [32]. Enhanced CT scans are also not recommended when there is significant renal damage (usually creatinine levels higher than 1.5 mg/dL) or a history of obvious allergy to contrast media [33]. Compared with the MCTSI score, the advantages of our models are that they are not only simple and fast but also does not need imaging examination, thus reducing unnecessary radiation and economic burden. Especially when severe and unstable patients are not suitable for CT enhanced scanning, it is not conducive to the early evaluation of MCTSI score.
The APACHE score was originally designed to assess the severity of acute illness patients who entered the ICU in the 1970s [9]. In 1985, Knaus et al. [22] simplified the APACHE score and proposed the APACHE II score, which is the most commonly used. Still, the APACHE II score takes into account various parameters, including acute physiological variables, age, and chronic health conditions, some of which may not be relevant to AP prognosis, whereas other important measures, such as pancreatic injury and significant regional complications, are missed [21,33,34]. Unlike the APACHE II score, the seven prognostic indicators included in the unwScore and wScore models all play a key role in the occurrence and development of AP or have clinical   Gastroenterology Research and Practice significance [9,20,27,[35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52]. These indexes are easy to obtain in the clinic, and the detection method is convenient. Furthermore, the model algorithm is simple, rapid, feasible, and easy to popularize and monitor dynamically and is also suitable for grass-roots hospitals. Beyond that, the unwScore and wScore models can also avoid unnecessary pain and financial expenditure to patients because blood gas analysis is not necessary. In 2008, Wu et al. [21] proposed the BISAP score. The detail of "SIRS" in the system is mainly based on the values of vital signs, which makes the score unstable so that evalu-ation has to be conducted repeatedly. And similar to APACHE II score, the criterion of "impaired mental status" in this system is subjective. Fortunately, all the variables included in the unwScore and wScore models are objective routine clinical indicators, thus reducing the subjective bias of evaluators, which could be used between different centers.
Compared to the four existing scoring systems, the performance of new models was good for the prediction of AP severity. The overall sensitivity, specificity, PPV, and NPV for the prediction of disease severity with the Ranson score were 75%, 77%, 49%, and 91%, respectively, according to a meta-analysis. At admission, the sensitivity of the APACHE II score > 7 to predict SAP is 65%, with a specificity of 76%, a PPV of 43%, a NPV of 89%, and an accuracy of 75%. At 48 hours, the sensitivity of the APACHE II score > 7 to predict SAP is 76%, with a specificity of 84%, a PPV of 54%, a NPV of 93%, and an accuracy of 70-80% [28,53]. Though the NPVs of our scoring systems were not satisfactory (66.92% and 67.48%), they had similar sensitivity (67.42% and 70.41%) and accuracy (75.91% and 75.05%) and slightly higher specificity (87.13% and 81.19%) and PPV (87.38% and 83.19%). Moreover, compared with the traditional score, although the BISAP score is a big step forward in simplicity, it is not an advancement in accuracy [29]. A prospective study by Papachristou et al. [34] showed that the BISAP score (AUC = 0:81, 95% CI: 0.74-0.87) was similar to the Ranson score (AUC = 0:94, 95% CI: 0.89-0.97), the APACHE II score (AUC = 0:78, 95% CI: 0.71-0.84), and the MCTSI score (AUC = 0:84, 95% CI: 0.76-0.89) in predicting SAP. While preserving their advantages of simplicity and efficiency, the unwScore (AUC = 0:854) and wScore (AUC = 0:837) models constructed in this study could still predict the severity of AP well, and their predictive values were even superior to those of the BISAP score (AUC = 0:766), the APACHE II score (AUC = 0:526), and the Ranson score (AUC = 0:693) and equivalent to that of the MCTSI score (AUC = 0:823).
In summary, we have derived two new scoring systems for predicting severity of AP based on commonly used clinical indexes. All seven variables included in this scoring system can be easily measured within 24 hours of admission. Compared to four existing scoring systems, the new  Figure 3: ROC curves of the unweighted predictive model, the weighted predictive model, and the severity scores of AP. Note: AP: acute pancreatitis; ROC: receiver operating characteristic; unwScore: unweighted predictive score; wScore: weighted predictive score; APACHE II: acute physiology and chronic health evaluation II; BISAP: bedside index for severity in acute pancreatitis; MCTSI: modified computed tomography severity index.