Development of a Novel Nomogram Incorporating Red Blood Cell Distribution Width-Albumin Ratio for the Prediction of 30-day Mortality in Acute Pancreatitis Patients

Purpose The available nomograms used to predict acute pancreatitis (AP) are not comprehensive. We sought to investigate the effect of red blood cell distribution width (RDW)-albumin ratio (RA) on prognosis of patients with AP and develop a new nomogram to identify AP patients at high risk for mortality. Methods We used data from the Medical Information Mart for Intensive Care IV version 2.0 (MIMIC-IV v2.0). A total of 487 patients with acute pancreatitis were included. Patients enrolled in the study were randomly assigned to the training set and validation set at a 7 : 3 ratio. According to the 30-day mortality rate, the data were divided into a survival group and a death group. Multivariate logistic regression was used to establish a prognostic nomogram for predicting the 30-day mortality in AP patients. The area under the receiver operating characteristic curve (AUC), calibration curve, the net reclassification improvement (NRI), the integrated discrimination improvement (IDI), and a decision curve analysis (DCA) are used to verify the overall performance of the model. Results Among 487 patients, 54 patients died (11.1%). 338 patients were assigned to the training cohort and 149 were assigned to the validation cohort. The multivariate analysis results showed that RA, age, heart rate, temperature, AST/ALT, BUN, hemoglobin, potassium, and bilirubin were independent risk factors. The prediction performance of the newly established nomogram was better than those of other common scoring systems (including SOFA, OASIS, and APSIII). The nomogram suggests that RA (OR = 1.706, 95% CI: 1.367–2.185) is the most significant laboratory test indicator influencing prognosis. Conclusion The new nomogram incorporating RA performed well in predicting AP short-term mortality. A prospective study with a larger sample is needed to validate our findings.


Introduction
Acute pancreatitis is a complicated gastrointestinal-related disease with considerable morbidity and mortality and a fuctuating course that is frequently difcult to predict in its early stages [1,2]. In particular cases of severe acute pancreatitis (SAP), pancreatic tissue releases activated trypsin and a large number of infammatory factors, aggravating the infammatory necrosis of the pancreas and even involving other organs and tissues, with a mortality rate of up to 20% [3]. Although approximately, 80% of these patients eventually develop either mild or moderate symptoms, one in fve patients will develop severe pancreatitis. Terefore, improving the ability to diagnose severe pancreatitis early is essential for lowering the death rate from acute pancreatitis.
Tere are already a number of prediction models or scoring systems available for doctors to use in assessing the severity of AP in its early stages, but none are ideal. Te Ranson score, for instance, requires lactate dehydrogenase (LDH) at admission, which is not routinely evaluated in the emergency department of many hospitals and is frequently missed. Too many parameters are included in the Acute Physiology and Chronic Health Evaluation II (APACHE II) score [4], preventing its broad adoption by doctors. Terefore, a more convenient, rapid, and efective assessment method is still needed.
By quantifying the size of peripheral red blood cells, the red blood cell distribution width (RDW), a common blood test item, can indicate the heterogeneity of red blood cell volume [5]. It is widely employed in the early diferential diagnosis of anemia [6]. Recent research has revealed a strong correlation between a high RDW and the prognosis of a variety of serious diseases, including cancer [7], cardiovascular diseases [8], ischemic stroke [9], and acute respiratory distress syndrome (ARDS) [10]. RA is the ratio of RDW to albumin, and a study showed that the predictive value of RA in stroke patients is even greater than that of RDW [11]. However, few studies have been conducted on the ratio of RDW to albumin in infammatory diseases.
A nomogram is a helpful mathematical tool for forecasting certain endpoints, such as the evolution of disease progression or mortality, depending on a number of critical parameters. Researchers frequently use nomograms to calculate the risk of a clinical event occurrence [12]. A total of 487 patients with acute pancreatitis were selected from the MIMIC-IV database for this retrospective study, which aimed to determine the capacity of RA to predict mortality in patients with acute pancreatitis. Simultaneously, multiple independent risk factors, including RA, were integrated to build a prognostic nomogram to improve the prediction of overall survival in AP patients.

Sources of Data.
All the data for this study were derived from MIMIC-IV (2.0), a public critical care database developed by MIT. Te database collected data on 315,640 inpatients at Beth Israel Deaconess Medical Center (BIDMC) from 2008 to 2019. Laboratory tests, treatment records, demographic information, and so on are all examples of data. Te database has been thoroughly deprivatized of all personal information. Te researcher has completed an ethical training program and has been granted access to the database by the MIT Institutional Review Board (Certifcate number: 43700334).

Case Inclusion Criteria.
Te data were collected from the database using PostgreSQL tools and SQL structured query language, and patients with acute pancreatitis were identifed using the ICD-9 diagnosis code (5770). Te following patients were omitted from the study: (1) Patients younger than 18 years of age. (2) Patients' hospital stay was less than 24 hours. (3) Red blood cell distribution width and albumin data were missing. (4) Patients with missing information rate >10%. If the patient was hospitalized more than once, only data from the frst hospitalization were evaluated.

Data Collection.
Te extracted data included demographic information, prognostic scoring systems, vital signs, comorbidities, and laboratory fndings within 24 hours. Demographic information includes sex and age. Te prognostic scoring systems included Sequential Organ Failure Score Assessment (SOFA), Acute Physiology Score III (APS III), and Oxford Acute Disease Severity Score (OASIS). Vital signs include heart rate, mean blood pressure, breath rate, and temperature. Laboratory fndings included red blood cell distribution width (RDW), albumin, red blood cell distribution width-albumin ratio (RA), aspartate aminotransferase-alanine aminotransferase (AST/ALT), blood urea nitrogen (BUN), creatinine, blood glucose, hemoglobin, white blood cell count (WBC), platelets, blood calcium, blood sodium, blood potassium, and total bilirubin. Comorbidities include chronic heart failure (CHF), chronic pulmonary disease, liver disease, malignant cancer, diabetes, and sepsis. Te primary outcome of this study was 30-day mortality.

Statistical Analysis.
Tis study included variables with a missing rate of less than 10% to reduce the possibility of information bias. In variables that do not ft a normal distribution, we use the median to fll in the missing values. We use the mean to fll in the missing values in variables that ft a normal distribution. Te included cases were then randomly assigned in a 7 : 3 ratio to the training cohort and the validation cohort. Continuous variables with a normal distribution were expressed as mean standard deviation (Mean ± SD), and group comparisons were made using the independent sample T test. Continuous variables that did not ft the normal distribution were expressed as medians (quartiles), and group comparisons were made using the Mann-Whitney U tests. Categorical variables were written as frequency (%), and the chi-square or Fisher's exact test was used to compare groups.
In the training cohort, we utilized univariate and multivariate logistic regression to identify characteristics associated with 30-day mortality in individuals with acute pancreatitis. Variables with a P of value less than 0.1 obtained from univariate regression were included in multivariate regression for further analysis. Te backward stepwise regression method was applied to a multivariate logistic regression model to select variables in the training cohort. For continuous variables that comprised all independent risk factors in the fnal model, we tested for multicollinearity amongst prognostic components using variance infation factor (VIF), and an arithmetic square root of VIF ≥2 was considered as collinearity.
Te independent risk factors in the multivariate logistic model were used to construct a nomogram to predict the 30day survival rate of patients with acute pancreatitis. Te area under the receiver operating characteristic curve (ROC-AUC) was used as the evaluation index of the nomogram. Te calibration curve was used to evaluate the consistency between the predicted probabilities and the actual results. If the calibration curve is close to the diagonal, the calibration is considered good.
Additionally, net reclassifcation improvement (NRI) and integrated discrimination improvement (IDI) were utilized to evaluate this nomogram. NRI and IDI refect the improved performance of the new model in comparison to the old model. NRI was utilized to compare the accuracy of two models. IDI was utilized to assess the efectiveness of the improvements. NRI and IDI are more sensitive to diferences in discrimination between two models than AUC.
Using a decision curve analysis (DCA), we evaluated the clinical value of the prediction model. DCA shows the clinical validity of the prognostic model by showing the net beneft of clinical action guided by the prognostic model. In this study, SPSS (22.0) and R (4.0.5) were used for statistical analysis and mapping. P < 0.05 was considered statistically signifcant.

Baseline Characteristics of Patients.
A total of 487 eligible patients were screened for this research, comprising 272 males (55.9%) and 215 females (44.1%), with a 30-day mortality rate of 11.1%. 338 patients were assigned to the training cohort and 149 to the validation cohort, and none of the characteristics were signifcantly diferent between the two cohorts. Te training cohort consisted of 194 (57.4%) males and 144 (42.6%) females, with a median age of 59 years. Te validation cohort consisted of 78 (52.3%) males and 71 (47.7%) females with a median age of 58 years. Table 1 summarizes the baseline characteristics of patients with acute pancreatitis.

Univariate and Multivariate Logistic Regression Analyses.
Using univariate logistic regression analysis, we discovered that age, heart rate, MBP, breath rate, temperature, RA, AST/ ALT, BUN, creatinine, hemoglobin, WBC, platelet, potassium, bilirubin, CHF, and sepsis were risk variables for 30day death in patients with acute pancreatitis. We integrated these variables into multivariate logistic analysis, and the results indicated that age, heart rate, temperature, RA, AST/ ALT, BUN, hemoglobin, potassium, and bilirubin remained risk factors for mortality in acute pancreatitis patients. Among them, we discovered that RA was strongly associated with 30-day mortality, with an OR of 1.706, implying that every one-point increase in RA increases the probability of death by 70.6%. Te VIF was computed, and none of the variables listed above had an arithmetic square root of VIF greater than 2, suggesting that multicollinearity was not present in the model. Table 2 summarizes the results of univariate and multivariate logistic regression analyses.

Prognostic Nomogram for 30-Day Mortality and the
Nomogram's Performance. We establish a nomogram prediction model based on the results of a multiple logistic regression model. Figure 2 depicts the nomogram containing all the independent risk factors. Te nomogram suggests that RA (OR = 1.706, 95%CI: 1.367-2.185) is the most signifcant indicator, along with AST/ALT, age, temperature, BUN, bilirubin, heart rate, potassium, and hemoglobin. Based on the fndings of nomogram analysis, the area under the receiver operating characteristic curve (AUC) of the training model for predicting the probability of death within 30 days of acute pancreatitis was 0.891, with a 95% confdence interval of 0.846-0.936 (Figure 3(a)). Compared with SOFA, OASIS, and APS III, the AUC value of the training cohort was signifcantly improved. Table 3 summarizes the results of the received operating characteristics curve of the nomogram.
Te bootstrap method was used to generate the internal calibration curve, and the calibration curve nearly matched the reference curve, showing that the nomogram's capacity for diferentiation and prediction was excellent (Figure 4(a)). We put the verifcation cohort's information into the nomogram for verifcation. Results indicated that the AUC for predicting the risk of death was 0.898, with a 95% confdence interval of 0.814-0.982 (Figure 3(b)). Te calibration curve for the validation cohort stayed close to the standard curve ( Figure 4(b)), which shows that the nomogram is good at making predictions.
After comparing the nomogram with SOFA, OASIS, and APS III, the NRI of the training cohort was 0.35, 0.407, and 0.25, respectively. Te NRI of the validation cohort was 0.473, 0.515, and 0.372, respectively. In addition, the IDI of the training cohort was 0.199, 0.249, and 0.159, respectively. Te IDI of the validation cohort was 0.355, 0.382, and 0.37, respectively. Table 3 summarizes the NRI and IDI of the training and validation cohorts. We discovered that both NRI and IDI were positive, indicating that the nomogram had a greater capacity for discrimination than the other three scoring systems. In the DCA curve of Figure 5, when the threshold probability is between 0.4 and 0.8, we fnd that the DCA curve of the nomogram is higher than that of the three scoring systems. Tis indicates that clinical interventions guided by nomograms could achieve greater net benefts.

Discussion
Te essence of acute pancreatitis is autodigestion of the pancreatic parenchyma, which means that there is a local infammatory response but also a systemic infammatory damage, which can be secondary to serious bacterial infection and multiple organ damage [13]. Te severity can range from mild self-limited disease to severe acute necrotizing pancreatitis, which is characterized by systemic complications and multiple organ failure [14]. Simultaneously, individuals with acute pancreatitis frequently advance rapidly, making early misdiagnosis common. Tus, early, accurate, and prompt assessment is critical for patients with acute pancreatitis, since it is the key to reducing mortality. Currently, there is still a dearth of efcient clinical markers for swiftly and sensitively assessing the severity of acute pancreatitis. Even while physical examination, imaging features, blood laboratory testing, and scoring systems such as the Bedside Index for Severity in the Acute Emergency Medicine International Pancreatitis (BISAP) score are currently popular ways to diferentiate the severity of acute pancreatitis [15]; these methods have their own set of shortcomings. For instance, abdominal pain is not typical in patients; premature CT examination often cannot confrm the diagnosis and may even lead to a misdiagnosis. Serum amylase and serum calcium have low specifcity and are frequently inconsistent with the true severity of the patient's condition. Te scoring system is still not sufciently standardized, which restricts its clinical utility to some extent. Terefore, RA, as a straightforward and easily obtainable static indicator, is an excellent supplement to existing evaluation approaches.
In recent years, a number of diagnostic and prognostic markers, both alone and in combination, have been tested on individuals diagnosed with AP [16][17][18]. After elevated RDW was established as a predictor of poor outcomes in chronic heart failure in 2007, multiple studies have shown that it is signifcantly related to short-term mortality outcomes in a range of infammatory disorders [19,20]. Additionally, Jin [21] and Zhang's [22] studies demonstrated a strong correlation between RDW and acute pancreatitis mortality. Admittedly, RA's outstanding predictive value is in part due to RDW's excellent forecasting ability. However, whether RA is a more accurate predictor of short-term mortality in acute pancreatitis than RDW remains unknown. As a result, we compared the AUC values of RA, RDW, and albumin in this study and were pleasantly surprised to discover that RA had a signifcantly better predictive performance than RDW and albumin, indicating that RA is a more optimized predictor than RDW, which may be inextricably linked to albumin's critical protective role in a variety of infammatory diseases [23,24]. Tis result is also similar to that of Zhao's study [11]. Simultaneously, univariate analysis revealed that a decrease in albumin was substantially linked with mortality in this study (Table 1), implying that serum albumin may possibly have a protective efect in patients with pancreatitis. However, the mechanisms underlying the association between RDW and mortality from infammatory diseases remain unclear. It has been shown in one study that there is a strong, positive, and independent correlation between RDW and traditional infammatory biomarkers. Tis may be because infammation reduces red blood cell survival, resulting in diferences in red blood cell volumes and an increase in size heterogeneity among red blood cells [25]. Additional evidence suggests that oxidative stress enhances cellular heterogeneity by afecting erythropoiesis and reducing the half-life of red blood cells in circulation, ultimately resulting in elevated RDW levels [26]. However, it remains challenging to reliably estimate the prognosis of a disease based on a single laboratory marker. Increasing the number of indicators used in a diagnostic procedure can improve either its sensitivity or its specifcity, and a nomogram is a visual graph of the fndings of a multivariate regression analysis [27]. Because the complicated regression equation is converted into a visual graph by the nomogram, the fndings of the regression analysis model are made easier to comprehend, are more intuitive, and are simpler to evaluate for the patient. Tese are advantages of the nomogram. Terefore, nomograms are accepted by a signifcant number of researchers involved in clinical medical studies and they are widely applied as a visual statistical model for the purpose of determining the prognosis and risk of disease [28,29].
In this work, we established a nomogram for the early prediction of death in AP patients based on a large population gathered from a critical disease database. Te variables included in the nomogram were RA, age, heart rate, temperature, AST/ALT, BUN, hemoglobin, potassium, and bilirubin. In addition, the clinical indicators and laboratory test data included in this nomogram can be easily obtained. Te model was validated, and its predictive performance was comparable to that of current scoring systems.
Te roles of BUN and creatinine in predicting the prognosis of AP are still debatable. Both renal damage induced by infammation and renal hypoperfusion produced by hypovolemia can elevate BUN and creatinine levels. Yang et al. [30] revealed that the predictive value of creatinine was much higher than that of BUN in predicting the short-term mortality of severe acute pancreatitis; however, our results demonstrated that the predictive value of BUN was even higher, which is consistent with Jiang's fndings [31]. Te relatively plausible explanation for the early increase in BUN is insufcient resuscitation resulting in prerenal azotemia [32]. Multiple prior studies have indicated that high bilirubin is a substantial risk factor for early mortality in AP [33,34]. When AP develops, bile ducts are clogged, bile cannot be excreted, and bilirubin is retained in liver cells, disrupting the normal metabolism of liver cells and reducing liver function [35]. In our study, in addition to the bilirubin level, we also found that the AST/ALT ratio was an independent risk factor. AST and ALT are typically used to   In both fgures, RA had higher AUC values than RDW and albumin. Te AUC value of RA is lower than SOFA and APS III but higher than OASIS. ROC, receiver operating characteristic; RA, red blood cell distribution width-albumin ratio; SOFA, sequential organ failure assessment; OASIS, Oxford acute disease severity score; APS III, acute physiology score III.
signal liver damage, and alcohol consumption can lead to the development of alcoholic cirrhosis and pancreatitis. Studies have shown that the AST/ALT ratio is closely related to the degree of cirrhosis [36]. Terefore, the elevated AST/ALT ratio in a high number of AP patients may be due to excessive alcohol use. In addition to the abovementioned independent risk variables, blood potassium and heart rate were also incorporated into the prediction model, and an increase in these two indicators suggests that the patient may have had severe internal environment abnormalities, metabolic acidosis, or even shock. Our study has several strengths. First, we found that RA is a signifcant independent risk factor that was absent from earlier AP models. Second, we constructed a nomogram with readily accessible parameters. Tis nomogram presents the probability of AP short-term mortality outcomes and allows practitioners to analyze patient outcomes. Te performance of our model is far superior to that of traditional disease severity scores.
However, it is vital to point out some of the study's shortcomings, which are as follows: We should note that because this was a retrospective observational study, there is still the possibility of selection bias and confusion bias in our fndings. In regard to the second issue, this study's conclusions are not supported by multicenter clinical data, which raises questions about the validity of these fndings. Ideally, we should perform additional multicenter research in the future using a large number of study sites.

Conclusion
Our study reveal the predictive values of RA in patients with AP. Concurrently, the new nomogram demonstrated good performance in predicting short-term mortality in AP by constructing a nomogram employing RA and a number of other easily accessible clinical parameter relevant to AP prognosis. However, additional confrmation of the present fndings requires the conduct of larger prospective investigations with a greater duration of follow-up.

Conflicts of Interest
Te authors declare that they have no conficts of interest.

Authors' Contributions
Li-na Pan and Shen-ao Pan contributed equally to this work.