Classification and regression tree analysis of acute-on-chronic hepatitis B liver failure: Seeing the forest for the trees

Acute on chronic liver failure: prognostic models and artificial intelligence applications

Critically ill patients presenting with acute on chronic liver failure (ACLF) represent a particularly vulnerable population due to various considerations surrounding the syndrome definition, lack of robust prospective evaluation of outcomes, and allocation of resources such as organs for transplantation. Ninety-day mortality related to ACLF is high and patients who do leave the hospital are frequently readmitted. Artificial intelligence (AI), which encompasses various classical and modern machine learning techniques, natural language processing, and other methods of predictive, prognostic, probabilistic, and simulation modeling, has emerged as an effective tool in various areas of healthcare. These methods are now being leveraged to potentially minimize physician and provider cognitive load and impact both short-term and long-term patient outcomes. However, the enthusiasm is tempered by ethical considerations and a current lack of proven benefits. In addition to prognostic applications, AI models can likely help improve the understanding of various mechanisms of morbidity and mortality in ACLF. Their overall impact on patient-centered outcomes and countless other aspects of patient care remains unclear. In this review, we discuss various AI approaches being utilized in healthcare and discuss the recent and expected future impact of AI on patients with ACLF through prognostic modeling and AI-based approaches.

Risk factors for pressure ulcer recurrence following surgical reconstruction: A cross-sectional retrospective analysis

Many studies on the recurrence of pressure ulcers after surgical reconstruction have focused on surgical techniques and socioeconomic factors. Herein, we aimed to identify the risk factors of the associated comorbidities for pressure ulcer recurrence. We enrolled 147 patients who underwent pressure ulcer reconstruction and were followed up for more than three years. The recurrence of pressure ulcers was defined as recurrent pressure ulcers with stage 3/4 pressure ulcers. We reviewed and analyzed systematic records of medical histories, including sex, age, associated comorbidities such as spinal cord injury (SCI), diabetes mellitus (DM), coronary artery disease, cerebral vascular accident, end-stage renal disease, scoliosis, dementia, Parkinson's disease, psychosis, autoimmune diseases, hip surgery, and locations of the primary pressure ulcer. Patients with recurrent pressure ulcers were younger than those without. Patients with SCI and scoliosis had higher odds, while those with Parkinson's disease had lower odds of recurrence of pressure ulcers than those without these comorbidities. Moreover, the decision tree algorithm identified that SCI, DM, and age < 34 years could be risk factor classifiers for predicting recurrent pressure ulcers. This study demonstrated that age and SCI are the two most important risk factors associated with recurrent pressure ulcers following surgical reconstruction.

Construction of a Compact and High-Precision Classifier in the Inductive Learning Method for Prediction and Diagnostic Problems

The study is dictated by the need to make reasonable decisions in the classification of observations, for example, in the problems of medical prediction and diagnostics. Today, as part of the digitalization in healthcare, decision-making by a doctor is carried out using intelligent information systems. The introduction of such systems contributes to the implementation of policies aimed at ensuring sustainable development in the health sector. The paper discusses the method of inductive learning, which can be the algorithmic basis of such systems. In order to build a compact and high-precision classifier for the studied method, it is necessary to obtain a set of informative patterns and to create a method for building a classifier with high generalizing ability from this set of patterns. Three optimization models for the building of informative patterns have been developed, which are based on different concepts. Additionally, two algorithmic procedures have been developed that are used to obtain a compact and high-precision classifier. Experimental studies were carried out on the problems of medical prediction and diagnostics, aimed at finding the best optimization model for the building of informative pattern and at proving the effectiveness of the developed algorithmic procedures.

Novel Prognostic Models for Predicting the 180-day Outcome for Patients with Hepatitis-B Virus-related Acute-on-chronic Liver Failure

BACKGROUND AND AIMS

It remains difficult to forecast the 180-day prognosis of patients with hepatitis B virus-acute-on-chronic liver failure (HBV-ACLF) using existing prognostic models. The present study aimed to derive novel-innovative models to enhance the predictive effectiveness of the 180-day mortality in HBV-ACLF.

METHODS

The present cohort study examined 171 HBV-ACLF patients (non-survivors, n=62; survivors, n=109). The 27 retrospectively collected parameters included the basic demographic characteristics, clinical comorbidities, and laboratory values. Backward stepwise logistic regression (LR) and the classification and regression tree (CART) analysis were used to derive two predictive models. Meanwhile, a nomogram was created based on the LR analysis. The accuracy of the LR and CART model was detected through the area under the receiver operating characteristic curve (AUROC), compared with model of end-stage liver disease (MELD) scores.

RESULTS

Among 171 HBV-ACLF patients, the mean age was 45.17 years-old, and 11.7% of the patients were female. The LR model was constructed with six independent factors, which included age, total bilirubin, prothrombin activity, lymphocytes, monocytes and hepatic encephalopathy. The following seven variables were the prognostic factors for HBV-ACLF in the CART model: age, total bilirubin, prothrombin time, lymphocytes, neutrophils, monocytes, and blood urea nitrogen. The AUROC for the CART model (0.878) was similar to that for the LR model (0.878, p=0.898), and this exceeded that for the MELD scores (0.728, p<0.0001).

CONCLUSIONS

The LR and CART model are both superior to the MELD scores in predicting the 180-day mortality of patients with HBV-ACLF. Both the LR and CART model can be used as medical decision-making tools by clinicians.

Distance-based Classification and Regression Trees for the analysis of complex predictors in health and medical research

Predicting patient outcomes based on patient characteristics and care processes is a common task in medical research. Such predictive features are often multifaceted and complex, and are usually simplified into one or more scalar variables to facilitate statistical analysis. This process, while necessary, results in a loss of important clinical detail. While this loss may be prevented by using distance-based predictive methods which better represent complex healthcare features, the statistical literature on such methods is limited, and the range of tools facilitating distance-based analysis is substantially smaller than those of other methods. Consequently, medical researchers must choose to either reduce complex predictive features to scalar variables to facilitate analysis, or instead use a limited number of distance-based predictive methods which may not fulfil the needs of the analysis problem at hand. We address this limitation by developing a Distance-Based extension of Classification and Regression Trees (DB-CART) capable of making distance-based predictions of categorical, ordinal and numeric patient outcomes. We also demonstrate how this extension is compatible with other extensions to CART, including a recently published method for predicting care trajectories in chronic disease. We demonstrate DB-CART by using it to expand upon previously published dose-response analysis of stroke rehabilitation data. Our method identified additional detail not captured by the previously published analysis, reinforcing previous conclusions. We also demonstrate how by combining DB-CART with other extensions to CART, the method is capable of making predictions about complex, multifaceted outcome data based on complex, multifaceted predictive features.

Derivation and Validation of a Nomogram for Predicting 90-Day Survival in Patients With HBV-Related Acute-on-Chronic Liver Failure

Background: Conventional prognostic models do not fully reflect the severity of hepatitis B virus (HBV)-related acute-on-chronic liver failure (ACLF). This study aimed to establish an effective and convenient nomogram for patients with HBV-related ACLF.

Methods: A nomogram was developed based on a retrospective cohort of 1,353 patients treated at the Third Affiliated Hospital of Sun Yat-sen University from January 2010 to June 2016. The predictive accuracy and discriminatory ability of the nomogram were determined by a concordance index (C-index) and calibration curve, and were compared with current scoring systems. The results were validated using an independent retrospective cohort of 669 patients consecutively treated at the same institution from July 2016 to March 2018. This study is registered at ClinicalTrials.gov (NCT03992898).

Results: Multivariable analysis of the derivation cohort found that independent predictors of 90-day survival were age, white blood cell (WBC) count, hemoglobin (Hb), aspartate aminotransferase (AST), total bilirubin (TBil), international normalized ratio, serum creatinine (Cr), alpha fetoprotein (AFP), serum sodium (Na), hepatic encephalopathy (HE), pre-existing chronic liver disease(PreLD), and HBV DNA load. All factors were included in the nomogram. The nomogram calibration curve for the probability of 90-day survival indicated that nomogram-based predictions were in good agreement with actual observations. The C-index of the nomogram was 0.790, which was statistically significantly greater than those for the current scoring systems in the derivation cohort (P < 0.001). The results were confirmed in the validation cohort.

Conclusions: The proposed nomogram is more accurate in predicting the 90-day survival of patients with HBV-related ACLF than current commonly used methods.

Proposed clinical indicators for efficient screening and testing for COVID-19 infection using Classification and Regression Trees (CART) analysis

The introduction and rapid transmission of SARS-CoV-2 in the United States resulted in methods to assess, mitigate, and contain the resulting COVID-19 disease derived from limited knowledge. Screening for testing has been based on symptoms typically observed in inpatients, yet outpatient symptoms may differ. Classification and regression trees recursive partitioning created a decision tree classifying participants into laboratory-confirmed cases and non-cases. Demographic and symptom data from patients ages 18-87 years enrolled from March 29-June 8, 2020 were included. Presence or absence of SARS-CoV-2 was the target variable. Of 832 tested, 77 (9.3%) tested positive. Cases significantly more often reported diarrhea (12 percentage points (PP)), fever (15 PP), nausea/vomiting (9 PP), loss of taste/smell (52 PP), and contact with a COVID-19 case (54 PP), but less frequently reported sore throat (-27 PP). The 4-terminal node optimal tree had sensitivity of 69%, specificity of 78%, positive predictive value of 20%, negative predictive value of 97%, and AUC of 76%. Among those referred for testing, negative responses to two questions could classify about half (49%) of tested persons with low risk for SARS-CoV-2 and would save limited testing resources. Outpatient symptoms of COVID-19 appear to be broader than the inpatient syndrome.Initial supplies of anticipated COVID-19 vaccines may be limited and administration of first such available vaccines may need to be prioritized for essential workers, the most vulnerable, or those likely to have a robust response to vaccine. Another priority group could be those not previously infected. Those who screen out of testing may be less likely to have been infected by SARS-CoV-2 virus thus may be prioritized for vaccination when supplies are limited.

Development and Validation of a Novel Risk Prediction Model Using Recursive Feature Elimination Algorithm for Acute-on-Chronic Liver Failure in Chronic Hepatitis B Patients With Severe Acute Exacerbation

Background: Patients with chronic hepatitis B (CHB) with severe acute exacerbation (SAE) are at a progression stage of acute-on-chronic liver failure (ACLF) but uniform models for predicting ACLF occurrence are lacking. We aimed to present a risk prediction model to early identify the patients at a high risk of ACLF and predict the survival of the patient. Methods: We selected the best variable combination using a novel recursive feature elimination algorithm to develop and validate a classification regression model and also an online application on a cloud server from the training cohort with a total of 342 patients with CHB with SAE and two external cohorts with a sample size of 96 and 65 patients, respectively. Findings: An excellent prediction model called the PATA model including four predictors, prothrombin time (PT), age, total bilirubin (Tbil), and alanine aminotransferase (ALT) could achieve an area under the receiver operating characteristic curve (AUC) of 0.959 (95% CI 0.941-0.977) in the development set, and AUC of 0.932 (95% CI 0.876-0.987) and 0.905 (95% CI 0.826-0.984) in the two external validation cohorts, respectively. The calibration curve for risk prediction probability of ACLF showed optimal agreement between prediction by PATA model and actual observation. After predictive stratification into different risk groups, the C-index of predictive 90-days mortality was 0.720 (0.675-0.765) for the PATA model, 0.549 (0.506-0.592) for the end-stage liver disease score model, and 0.648 (0.581-0.715) for Child-Turcotte-Pugh scoring system. Interpretation: The highlypredictive risk model and easy-to-use online application can accurately predict the risk of ACLF with a poor prognosis. They may facilitate risk communication and guidetherapeutic options.

Artificial liver support system therapy in acute-on-chronic hepatitis B liver failure: Classification and regression tree analysis

Artificial liver support systems (ALSS) are widely used to treat patients with hepatitis B virus-related acute-on-chronic liver failure (HBV-ACLF). The aims of the present study were to investigate the subgroups of patients with HBV-ACLF who may benefit from ALSS therapy, and the relevant patient-specific factors. 489 ALSS-treated HBV-ACLF patients were enrolled, and served as derivation and validation cohorts for classification and regression tree (CART) analysis. CART analysis identified three factors prognostic of survival: hepatic encephalopathy (HE), prothrombin time (PT), and total bilirubin (TBil) level; and two distinct risk groups: low (28-day mortality 10.2–39.5%) and high risk (63.8–91.1%). The CART model showed that patients lacking HE and with a PT ≤ 27.8 s and a TBil level ≤455 μmol/L experienced less 28-day mortality after ALSS therapy. For HBV-ACLF patients with HE and a PT > 27.8 s, mortality remained high after such therapy. Patients lacking HE with a PT ≤ 27.8 s and TBil level ≤ 455 μmol/L may benefit markedly from ALSS therapy. For HBV-ACLF patients at high risk, unnecessary ALSS therapy should be avoided. The CART model is a novel user-friendly tool for screening HBV-ACLF patient eligibility for ALSS therapy, and will aid clinicians via ACLF risk stratification and therapeutic guidance.

A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models

OBJECTIVES

The objective of this study was to compare performance of logistic regression (LR) with machine learning (ML) for clinical prediction modeling in the literature.

STUDY DESIGN AND SETTING

We conducted a Medline literature search (1/2016 to 8/2017) and extracted comparisons between LR and ML models for binary outcomes.

RESULTS

We included 71 of 927 studies. The median sample size was 1,250 (range 72-3,994,872), with 19 predictors considered (range 5-563) and eight events per predictor (range 0.3-6,697). The most common ML methods were classification trees, random forests, artificial neural networks, and support vector machines. In 48 (68%) studies, we observed potential bias in the validation procedures. Sixty-four (90%) studies used the area under the receiver operating characteristic curve (AUC) to assess discrimination. Calibration was not addressed in 56 (79%) studies. We identified 282 comparisons between an LR and ML model (AUC range, 0.52-0.99). For 145 comparisons at low risk of bias, the difference in logit(AUC) between LR and ML was 0.00 (95% confidence interval, -0.18 to 0.18). For 137 comparisons at high risk of bias, logit(AUC) was 0.34 (0.20-0.47) higher for ML.

CONCLUSION

We found no evidence of superior performance of ML over LR. Improvements in methodology and reporting are needed for studies that compare modeling algorithms.

Nomogram prediction of individual prognosis of patients with acute-on-chronic hepatitis B liver failure

BACKGROUND

The current definitions and etiologies of acute-on-chronic liver failure (ACLF) are clearly very different between East and West.

AIMS

This study aimed to develop an effective prognostic nomogram for acute-on-chronic hepatitis B liver failure (ACHBLF) as defined by the Asia Pacific Association for the Study of the Liver (APASL).

METHODS

The nomogram was based on a retrospective study of 573 patients with ACHBLF, defined according to the APASL, at the Beijing Ditan Hospital. The results were validated using a bootstrapped approach to correct for bias in two external cohorts, including an APASL ACHBLF cohort (10 hospitals, N = 329) and an EASL-CLIF ACHBLF cohort (Renji Hospital, N = 300).

RESULTS

Multivariate analysis of the derivation cohort for survival analysis helped identify the independent factors as age, total bilirubin, albumin, international normalized ratio, and hepatic encephalopathy, which were included in the nomogram. The predictive value of nomogram was the strongest compared with CLIF-C ACLF, MELD and MELD-Na and similar to COSSH-ACLF in both the derivation and prospective validation cohorts with APASL ACHBLF, but the CLIF-C ACLF was better in the EASL-CLIF ACHBLF cohort.

CONCLUSIONS

The proposed nomogram could accurately estimate individualized risk for the short-term mortality of patients with ACHBLF as defined by APASL.

Identification of Pancreatic Injury in Patients with Elevated Amylase or Lipase Level Using a Decision Tree Classifier: A Cross-Sectional Retrospective Analysis in a Level I Trauma Center

Background: In trauma patients, pancreatic injury is rare; however, if undiagnosed, it is associated with high morbidity and mortality rates. Few predictive models are available for the identification of pancreatic injury in trauma patients with elevated serum pancreatic enzymes. In this study, we aimed to construct a model for predicting pancreatic injury using a decision tree (DT) algorithm, along with data obtained from a population-based trauma registry in a Level I trauma center. Methods: A total of 991 patients with elevated serum levels of amylase (>137 U/L) or lipase (>51 U/L), including 46 patients with pancreatic injury and 865 without pancreatic injury between January 2009 and December 2016, were allocated in a ratio of 7:3 to training (n = 642) or test (n = 269) sets. Using the data on patient and injury characteristics as well as laboratory data, the DT algorithm with Classification and Regression Tree (CART) analysis was performed based on the Gini impurity index, using the rpart function in the rpart package in R. Results: Among the trauma patients with elevated amylase or lipase levels, three groups of patients were identified as having a high risk of pancreatic injury, using the DT model. These included (1) 69% of the patients with lipase level ≥306 U/L; (2) 79% of the patients with lipase level between 154 U/L and 305 U/L and shock index (SI) ≥ 0.72; and (3) 80% of the patients with lipase level <154 U/L with abdomen injury, glucose level <158 mg/dL, amylase level <90 U/L, and neutrophil percentage ≥76%; they had all sustained pancreatic injury. With all variables in the model, the DT achieved an accuracy of 97.9% (sensitivity of 91.4% and specificity of 98.3%) for the training set. In the test set, the DT achieved an accuracy of 93.3%, sensitivity of 72.7%, and specificity of 94.2%. Conclusions: We established a DT model using lipase, SI, and additional conditions (injury to the abdomen, glucose level <158 mg/dL, amylase level <90 U/L, and neutrophils ≥76%) as important nodes to predict three groups of patients with a high risk of pancreatic injury. The proposed decision-making algorithm may help in identifying pancreatic injury among trauma patients with elevated serum amylase or lipase levels.

Derivation and validation of different machine-learning models in mortality prediction of trauma in motorcycle riders: a cross-sectional retrospective study in southern Taiwan

OBJECTIVES

This study aimed to build and test the models of machine learning (ML) to predict the mortality of hospitalised motorcycle riders.

SETTING

The study was conducted in a level-1 trauma centre in southern Taiwan.

PARTICIPANTS

Motorcycle riders who were hospitalised between January 2009 and December 2015 were classified into a training set (n=6306) and test set (n=946). Using the demographic information, injury characteristics and laboratory data of patients, logistic regression (LR), support vector machine (SVM) and decision tree (DT) analyses were performed to determine the mortality of individual motorcycle riders, under different conditions, using all samples or reduced samples, as well as all variables or selected features in the algorithm.

PRIMARY AND SECONDARY OUTCOME MEASURES

The predictive performance of the model was evaluated based on accuracy, sensitivity, specificity and geometric mean, and an analysis of the area under the receiver operating characteristic curves of the two different models was carried out.

RESULTS

In the training set, both LR and SVM had a significantly higher area under the receiver operating characteristic curve (AUC) than DT. No significant difference was observed in the AUC of LR and SVM, regardless of whether all samples or reduced samples and whether all variables or selected features were used. In the test set, the performance of the SVM model for all samples with selected features was better than that of all other models, with an accuracy of 98.73%, sensitivity of 86.96%, specificity of 99.02%, geometric mean of 92.79% and AUC of 0.9517, in mortality prediction.

CONCLUSION

ML can provide a feasible level of accuracy in predicting the mortality of motorcycle riders. Integration of the ML model, particularly the SVM algorithm in the trauma system, may help identify high-risk patients and, therefore, guide appropriate interventions by the clinical staff.

Prediction of Mortality in Patients with Isolated Traumatic Subarachnoid Hemorrhage Using a Decision Tree Classifier: A Retrospective Analysis Based on a Trauma Registry System

Background: In contrast to patients with traumatic subarachnoid hemorrhage (tSAH) in the presence of other types of intracranial hemorrhage, the prognosis of patients with isolated tSAH is good. The incidence of mortality in these patients ranges from 0–2.5%. However, few data or predictive models are available for the identification of patients with a high mortality risk. In this study, we aimed to construct a model for mortality prediction using a decision tree (DT) algorithm, along with data obtained from a population-based trauma registry, in a Level 1 trauma center. Methods: Five hundred and forty-five patients with isolated tSAH, including 533 patients who survived and 12 who died, between January 2009 and December 2016, were allocated to training (n = 377) or test (n = 168) sets. Using the data on demographics and injury characteristics, as well as laboratory data of the patients, classification and regression tree (CART) analysis was performed based on the Gini impurity index, using the rpart function in the rpart package in R. Results: In this established DT model, three nodes (head Abbreviated Injury Scale (AIS) score ≤4, creatinine (Cr) <1.4 mg/dL, and age <76 years) were identified as important determinative variables in the prediction of mortality. Of the patients with isolated tSAH, 60% of those with a head AIS >4 died, as did the 57% of those with an AIS score ≤4, but Cr ≥1.4 and age ≥76 years. All patients who did not meet the above-mentioned criteria survived. With all the variables in the model, the DT achieved an accuracy of 97.9% (sensitivity of 90.9% and specificity of 98.1%) and 97.7% (sensitivity of 100% and specificity of 97.7%), for the training set and test set, respectively. Conclusions: The study established a DT model with three nodes (head AIS score ≤4, Cr <1.4, and age <76 years) to predict fatal outcomes in patients with isolated tSAH. The proposed decision-making algorithm may help identify patients with a high risk of mortality.

Gamma-glutamyl transpeptidase-to-platelet ratio predicts the prognosis in HBV-associated acute-on-chronic liver failure

BACKGROUND

The gamma-glutamyl transpeptidase-to-platelet ratio (GPR) is a new noninvasive marker for assessing liver fibrosis. We aimed to evaluate the performance of GPR for prediction of 90-day mortality in patients with acute-on-chronic liver failure (ACLF).

METHODS

A total of 355 patients with HBV-associated ACLF were enrolled from two clinical centers and divided into training group (n=210) and validation group (n=145). Potential risk factors for 90-day mortality were analyzed.

RESULTS

Age, MELD score and GPR were independent risk factors associated with ACLF prognosis. A new scoring system (MELD-GPR) was developed. MELD-GPR=9.211-0.029×age-0.290×MELD-0.460×GPR. For ACLF patients with liver cirrhosis, the area under the receiver operating characteristic curve (AUROC) of MELD-GPR was 0.788, which was significantly higher than that of MELD and MELD-Na (0.706 and 0.666, respectively). Patients were stratified into three groups according to MELD-GPR scores (high risk: <-0.19, intermediate risk: -0.19-0.95, and low risk: >0.95), and the high-risk group (MELD-GPR<-0.19) had a poor prognosis (P<0.01). For ACLF patients without liver cirrhosis, MELD-GPR<0.95 predicted a poor prognosis.

CONCLUSIONS

Incorporating GPR into MELD may provide more accurate survival prediction in patients with HBV-ACLF.