Development And Preliminary Validation Of Risk Prediction Model In Patients With Decompensated Cirrhosis Based On Big Data

PARTⅠ DEVELOPMENT AND VALIDATION OF A PREDICTION MODEL FOR PROLONGED HOSPITAL STAY IN PATIENTS WITH ACUTE DECOMPENSATED CIRRHOSIS BASED ON A BIG DATA PLATFORMBackground and Aims: Patients with acute decompensated(AD)cirrhosis have complex conditions and different treatment modalities,resulting in prolonged hospitalization days,which not only leads to rising hospitalization costs,but also causes shortage of medical resources.The purpose of this study is to explore the factors affecting the length of stay in hospital by using multicenter big data from the condition,medication,treatment process and laboratory indexes,and to establish a prediction model for prolonged hospitalization days to help clinicians to quickly identify high-risk factors and take preventive and curative measures at an early stage.It also provides a reference for the government,health insurance and medical institutions to develop strategies to control the average hospitalization days and hospitalization costs.Methods: This is a retrospective study in four tertiary hospitals in Chongqing,China,which included 8605 patients with AD cirrhosis who were hospitalized between September 2012 and June 2021.Demographic,disease,medication,treatment course and laboratory indices of all patients were collected for comparative analysis.The least absolute shrinkage and selection operator(LASSO)regression model was used for feature selection using the indicators with significant differences obtained by univariate analysis,and logistic regression-based prediction model was used to identify risk factors and construct a nomogram.The discrimination and calibration of the model were evaluated based on the receiver operating characteristic(ROC)curve and area under the ROC curve(AUC),calibration curve and brier score.Bootstrap method and internal-external cross-validation were used to adjust the model predictive performance and generalizability.The nomogram was compared with the model for end-stage liver disease(MELD),MELD-Na,and chronic liver failure-consortium acute decompensation(CLIF-C AD)models to further assess the clinical net benefit of the model.Results: A total of 8605 patients with 48 indicators were included in the study,and the average hospital stay was 13.7 days,of which 3194 patients had prolonged hospital stay,accounting for 37.1% of the total.The mean age of the patients was 58.5±12.3 years,of which 5996(69.7%)were male.The main cause of liver cirrhosis was hepatitis B virus infection(65.8%),followed by alcoholic liver disease(15.6%),other causes(10.9%),autoimmune liver disease(9.8%),and hepatitis C virus infection(5.1%).The main reason for admission was bacterial infection(26.2%),followed by spontaneous bacterial peritonitis(23.3%),gastrointestinal bleeding(18.4%),hepatic encephalopathy(6.8%),and ascites(4.6%).The final indicators included in the nomogram included:cardiovascular drugs [OR: 4.106,95% confidence interval(CI)3.626-4.65,P<0.001],hypoglycemic drugs(OR: 1.219,95%CI 1.06-1.401,P=0.005),proton pump inhibitors(OR: 1.346,95%CI 1.157-1.566,P<0.001),parenteral nutrition(OR: 1.588,95%CI 1.406-1.794,P<0.001),human albumin infusion(OR: 6.469,95%CI 5.645-7.412,P<0.001),antitussive and expectorant drugs(OR: 3.431,95%CI 2.912-4.042,P<0.001),therapeutic procedures(OR: 3.297,95%CI 2.858-3.805,P<0.001),spontaneous bacterial peritonitis(OR: 2.61,95%CI 2.268-3.004,P<0.001),total bilirubin(OR: 1.228,95%CI 1.14-1.322,P<0.001),alanine aminotransferase(OR: 1.717,95%CI 1.603-1.839,P<0.001)and white blood cell counts(OR: 1.185,95%CI 1.069-1.314,P=0.001).The AUC value of the model was 0.888,95% CI: 0.881-0.895,specificity of 72.3%,sensitivity of 93.7%.The brier score of the model was 0.132.A calculator tool was constructed on the web using shinyapp to facilitate clinical use(https://cytjt007.shinyapps.io/dynnomapp_hospitalization/).After internal-external cross-validation,the model had a mean AUC value of 0.764 and a mean brier score of 0.155.The new model was significantly more predictive than MELD,MELD-Na and CLIF-C AD,with AUC values of0.888,0.675,0.686 and 0.502,respectively.Conclusions: Patients with AD cirrhosis have complex conditions with multiple comorbidities and complications that predispose them to prolonged hospital stays.This study found that the use of cardiovascular drugs,hypoglycemic agents,proton pump inhibitors,parenteral nutrition,human albumin,antitussive and expectorant drugs,therapeutic operations,spontaneous peritonitis,elevated levels of total bilirubin,alanine aminotransferase and white blood cells were predictors of prolonged hospital stay.By establishing a clinical prediction model and a network APP,we provided clinicians with a tool to quickly calculate the probability of prolonged hospital stay of patients,which provides ideas for the development of clinical treatment plans and care concerns.In addition,the data from multiple centers is representative,and it is expected to provide reference for Diagnosis Related Groups/Prospective Payment System(DRG/PPS)policy development.PARTⅡ RISK STRATIFICATION SCORE TO PREDICT READMISSION OF PATIENTS WITH ACUTE DECOMPENSATED CIRRHOSIS WITHIN 90 DAYSBackground and Aims: Patients with AD cirrhosis are at risk of multiple complications requiring readmission,which can impose a heavy burden on patients,families,and healthcare systems.Doctors should not only identify high-risk patients who may be readmitted when they are discharged from the hospital,but also should take effective measures to avoid readmission during hospitalization and after discharge.This study aims to establish a predictive model to identify patients at high risk of readmission based on clinical big data,which will help clinicians formulate effective interventions to reduce readmission rates.Methods: This retrospective cohort study of 956 patients with AD cirrhosis was conducted at six tertiary hospitals in Chongqing between September 2012 and April 2020.Among them,705 patients admitted from September 2012 to December 2016 served as the derivation cohort,and251 patients admitted from January 2017 to April 2020 served as the temporal validation cohort.Demographics,clinical characteristics,and laboratory indicators of all patients were collected for comparative analysis.The LASSO regression model was used,and the indicators with significant differences obtained by univariate analysis were used for feature selection.Cox regression prognostic model was used to identify prognostic factors and construct nomogram.Model discrimination,calibration,and clinical net benefit were assessed based on C-index,ROC curve and AUC,calibration curve and decision curve analysis.Kaplan–Meier curves were constructed for stratified risk groups,and log-rank tests were used to determine significant differences between the curves.Results: A total of 956 patients were included,with a mean age of58.8 ± 12.6 years,of whom 653(68.31%)were male.The etiology of cirrhosis was hepatitis B virus infection(50.4%),alcoholic liver disease(10.7%),autoimmune liver disease(9.9%),hepatitis C virus infection(5.0%)and other/cryptogenic factors(19.6%).The readmission rates at 30,60 and 90 days after discharge were 24.58%,42.99% and 51.78%respectively.Bacterial infection was the main reason for index hospitalization and readmission.Independent factors in the nomogram included gastrointestinal bleeding [Hazard rate(HR): 2.787,95% CI:2.221 – 3.499,P<0.001],serum sodium(HR: 0.955;95% CI: 0.933 –0.978,P<0.001),total bilirubin(HR: 1.004,95% CI: 1.003 – 1.005,P<0.001),and international normalized ratio(HR: 1.398;95% CI: 1.126–1.734,P=0.002).For the convenience of clinicians,we provided a web-based calculator tool(https://cqykdx1111.shinyapps.io/dynnomapp/).The nomogram exhibited good discrimination ability,both in the derivation and validation cohorts.The predicted and observed readmission probabilities were calibrated with reliable agreement.The nomogram demonstrated superior predictive performance and net benefits over MELD,CLIF-C AD,Child-Turcotte-Pugh(CTP),and MELD-Na scoring models.The high-risk group(nomogram score > 56.8)was significantly likely to have higher rates of readmission than the low-risk group(nomogram score ≤ 56.8;p < 0.0001).Conclusions: Short-term readmission rates associated with AD cirrhosis are high.Bacterial infections are the leading cause of hospital admission and readmission.We developed and validated a prognostic model with 4 predictors of gastrointestinal bleeding,serum sodium,total bilirubin,and international normalized ratio,which assessed the probability of short-term readmission in patients with AD cirrhosis.This nomogram can assist clinical decision-making,and help differentiate patients who require intensive management to prevent short-term readmissions.PARTⅢ RECOMPENSATION FACTORS FOR PATIENTS WITH DECOMPENSATED CIRRHOSISBackground and Aims: After effective etiological control,treatment or prevention,some patients with decompensated cirrhosis may no longer have decompensation-related complications,for a long period of time,or even withdraw from the list of liver transplantation,called“recompensation”.The occurrence of recompensation can reduce the mortality of patients and improve the quality of life.The aim of this study was to assess the factors associated with recompensation in patients with decompensated liver cirrhosis.Methods: This study was a retrospective case-control study conducted in six tertiary hospitals in Chongqing,including 3953 consecutive follow-up patients with decompensated cirrhosis from January2014 to October 2019.Demographics,clinical characteristics,and laboratory indicators were collected from and compared between groups of patients with recompensated and AD cirrhosis.Univariable and multivariable logistic regressions were used to select indicators associated with recompensation among patients with decompensated cirrhosis with different complications.A decision tree algorithm with 10-fold cross validation was used to develop and validate the model to identify the characteristics of patients with recompensation.We followed the transparent reporting of a multivariable predictive model for individual prognosis or diagnosis(TRIPOD)guideline for development and reporting of the new model.Results: In the total sample of included patients,there were 553 patients with recompensation and 3400 patients with acute decompensation,including 1158 patients with gastrointestinal bleeding,1715 patients with a bacterial infection,104 patients with hepatic encephalopathy,and 423 patients with ascites.The most relevant indicator of recompensation selected by the decision tree model was albumin,with a threshold of 40 g/L.Total protein,hemoglobin,basophil percentage,alanine aminotransferase,neutrophil-to-lymphocyte ratio,and diabetes were also selected to subsequently distinguish patients.The terminal nodes with a probability of recompensation was 0.89.The overall accuracy rate of the model was 0.92(95% CI: 0.91-0.93),and it exhibited high specificity(86.9%)and sensitivity(92.6%).Conclusions: Albumin,total protein,hemoglobin,percentage of basophils,alanine aminotransferase,neutrophil to lymphocyte ratio and the absence of diabetes mellitus can be used as characteristics of recompensation in patients with decompensated cirrhosis.Of these,a serum albumin level of ≥40 g/L is the most distinctive feature in patients with recompensation.These variables may help clinicians to develop treatment plans to facilitate the transition of patients with decompensated cirrhosis to recompensation and improve their quality of life and survival rates.