A comparison of Child-Pugh, APACHE II and APACHE III scoring systems in predicting hospital mortality of patients with liver cirrhosis

Development, validation, and feature extraction of a deep learning model predicting in-hospital mortality using Japan's largest national ICU database: a validation framework for transparent clinical Artificial Intelligence (cAI) development

OBJECTIVE

While clinical Artificial Intelligence (cAI) mortality prediction models and relevant studies have increased, limitations including the lack of external validation studies and inadequate model calibration leading to decreased overall accuracy have been observed. To combat this, we developed and evaluated a novel deep neural network (DNN) and a validation framework to promote transparent cAI development.

METHODS

Data from Japan's largest ICU database was used to develop the DNN model, predicting in-hospital mortality including ICU and post-ICU mortality by days since ICU discharge. The most important variables to the model were extracted with SHapley Additive exPlanations (SHAP) to examine the DNN's efficacy as well as develop models that were also externally validated.

MAIN RESULTS

The area under the receiver operating characteristic curve (AUC) for predicting ICU mortality was 0.94 [0.93-0.95], and 0.91 [0.90-0.92] for in-hospital mortality, ranging between 0.91-0.95 throughout one year since ICU discharge. An external validation using only the top 20 variables resulted with higher AUCs than traditional severity scores.

CONCLUSIONS

Our DNN model consistently generated AUCs between 0.91-0.95 regardless of days since ICU discharge. The 20 most important variables to our DNN, also generated higher AUCs than traditional severity scores regardless of days since ICU discharge. To our knowledge, this is the first study that predicts ICU and in-hospital mortality using cAI by post-ICU discharge days up to over a year. This finding could contribute to increased transparency on cAI applications.

Prognosis of cirrhotic patients admitted to intensive care unit: a meta-analysis

Background

The best predictors of short- and medium-term mortality of cirrhotic patients receiving intensive care support are unknown.

Methods

We conducted meta-analyses from 13 studies (2523 cirrhotics) after selection of original articles and response to a standardized questionnaire by the corresponding authors. End-points were in-ICU, in-hospital, and 6-month mortality in ICU survivors. A total of 301 pooled analyses, including 95 analyses restricted to 6-month mortality among ICU survivors, were conducted considering 249 variables (including reason for admission, organ replacement therapy, and composite prognostic scores).

Results

In-ICU, in-hospital, and 6-month mortality was 42.7, 54.1, and 75.1%, respectively. Forty-eight patients (3.8%) underwent liver transplantation during follow-up. In-ICU mortality was lower in patients admitted for variceal bleeding (OR 0.46; 95% CI 0.36–0.59; p < 0.001) and higher in patients with SOFA > 19 at baseline (OR 8.54; 95% CI 2.09–34.91; p < 0.001; PPV = 0.93). High SOFA no longer predicted mortality at 6 months in ICU survivors. Twelve variables related to infection were predictors of in-ICU mortality, including SIRS (OR 2.44; 95% CI 1.64–3.65; p < 0.001; PPV = 0.57), pneumonia (OR 2.18; 95% CI 1.47–3.22; p < 0.001; PPV = 0.69), sepsis-associated refractory oliguria (OR 10.61; 95% CI 4.07–27.63; p < 0.001; PPV = 0.76), and fungal infection (OR 4.38; 95% CI 1.11–17.24; p < 0.001; PPV = 0.85). Among therapeutics, only dopamine (OR 5.57; 95% CI 3.02–10.27; p < 0.001; PPV = 0.68), dobutamine (OR 8.92; 95% CI 3.32–23.96; p < 0.001; PPV = 0.86), epinephrine (OR 5.03; 95% CI 2.68–9.42; p < 0.001; PPV = 0.77), and MARS (OR 2.07; 95% CI 1.22–3.53; p = 0.007; PPV = 0.58) were associated with in-ICU mortality without heterogeneity. In ICU survivors, eight markers of liver and renal failure predicted 6-month mortality, including Child–Pugh stage C (OR 2.43; 95% CI 1.44–4.10; p < 0.001; PPV = 0.57), baseline MELD > 26 (OR 3.97; 95% CI 1.92–8.22; p < 0.0001; PPV = 0.75), and hepatorenal syndrome (OR 4.67; 95% CI 1.24–17.64; p = 0.022; PPV = 0.88).

Conclusions

Prognosis of cirrhotic patients admitted to ICU is poor since only a minority undergo liver transplant. The prognostic performance of general ICU scores decreases over time, unlike the Child–Pugh and MELD scores, even recorded in the context of organ failure. Infection-related parameters had a short-term impact, whereas liver and renal failure had a sustained impact on mortality.

Electronic supplementary material

The online version of this article (doi:10.1186/s13613-017-0249-6) contains supplementary material, which is available to authorized users.

Statistical analysis of a low cost method for multiple disease prediction

Early identification of individuals at risk for chronic diseases is of significant clinical value. Early detection provides the opportunity to slow the pace of a condition, and thus help individuals to improve or maintain their quality of life. Additionally, it can lessen the financial burden on health insurers and self-insured employers. As a solution to mitigate the rise in chronic conditions and related costs, an increasing number of employers have recently begun using wellness programs, which typically involve an annual health risk assessment. Unfortunately, these risk assessments have low detection capability, as they should be low-cost and hence rely on collecting relatively few basic biomarkers. Thus one may ask, how can we select a low-cost set of biomarkers that would be the most predictive of multiple chronic diseases? In this paper, we propose a statistical data-driven method to address this challenge by minimizing the number of biomarkers in the screening procedure while maximizing the predictive power over a broad spectrum of diseases. Our solution uses multi-task learning and group dimensionality reduction from machine learning and statistics. We provide empirical validation of the proposed solution using data from two different electronic medical records systems, with comparisons over a statistical benchmark.

Acute-on-chronic liver failure: consensus recommendations of the Asian Pacific Association for the Study of the Liver (APASL) 2014

The first consensus report of the working party of the Asian Pacific Association for the Study of the Liver (APASL) set up in 2004 on acute-on-chronic liver failure (ACLF) was published in 2009. Due to the rapid advancements in the knowledge and available information, a consortium of members from countries across Asia Pacific, "APASL ACLF Research Consortium (AARC)," was formed in 2012. A large cohort of retrospective and prospective data of ACLF patients was collated and followed up in this data base. The current ACLF definition was reassessed based on the new AARC data base. These initiatives were concluded on a 2-day meeting in February 2014 at New Delhi and led to the development of the final AARC consensus. Only those statements which were based on the evidence and were unanimously recommended were accepted. These statements were circulated again to all the experts and subsequently presented at the annual conference of the APASL at Brisbane, on March 14, 2014. The suggestions from the delegates were analyzed by the expert panel, and the modifications in the consensus were made. The final consensus and guidelines document was prepared. After detailed deliberations and data analysis, the original proposed definition was found to withstand the test of time and identify a homogenous group of patients presenting with liver failure. Based on the AARC data, liver failure grading, and its impact on the "Golden therapeutic Window," extra-hepatic organ failure and development of sepsis were analyzed. New management options including the algorithms for the management of coagulation disorders, renal replacement therapy, sepsis, variceal bleed, antivirals, and criteria for liver transplantation for ACLF patients were proposed. The final consensus statements along with the relevant background information are presented here.

APACHE II score is superior to SOFA, CTP and MELD in predicting the short-term mortality in patients with acute-on-chronic liver failure (ACLF)

OBJECTIVE

The aim of the study was to assess the performance of various prognostic scores including the acute physiology and chronic health evaluation (APACHE II), sequential organ failure assessment (SOFA), Child-Turcotte-Pugh (CTP) and model for end-stage liver disease (MELD) scores in predicting short-term mortality in patients with acute-on-chronic liver failure (ACLF).

METHODS

Altogether 100 consecutive patients with ACLF were evaluated prospectively. The diagnosis of ACLF was based on the Asian-Pacific Association for the Study of the Liver criteria except for the inclusion of non-hepatic insults as acute events. Sensitivity, specificity, positive and negative predictive values, and diagnostic accuracy for predicting short-term mortality was calculated for APACHE II, SOFA, CTP and MELD in all patients and Maddrey's discriminant function (DF) and Glasgow alcoholic hepatitis scores (GAHS) for patients with alcoholic hepatitis only.

RESULTS

Most patients had alcohol-related cirrhosis and alcoholic hepatitis as acute insults for ACLF. A total of 53 patients either died or left hospital in very sick status and were confirmed to have died the same day after leaving hospital. Overall, the area under the receiver operating characteristic curve of APACHE II was higher than those of MELD, SOFA and CTP scores for predicting short-term mortality. Even for patients with alcoholic hepatitis, APACHE II performed better than DF and GAHS.

CONCLUSIONS

Short-term mortality is high in patients with ACLF. APACHE II scoring system is superior to other prognostic scores in predicting its short-term mortality.

Predictors of the outcomes of acute-on-chronic hepatitis B liver failure

AIM: To identify the risk factors in predicting the outcome of acute-on-chronic hepatitis B liver failure patients.

METHODS: We retrospectively divided 113 patients with acute-on-chronic liver failure-hepatitis B virus (ACLF-HBV) and without concurrent hepatitis C or D virus infection and hepatocellular carcinoma into two groups according to their outcomes after anti-HBV therapy. Their demographic, clinical, and biochemical data on the day of diagnosis and after the first week of treatment were analyzed using the Mann-Whitney U test, Fisher’s exact test, and a multiple logistic regression analysis.

RESULTS: The study included 113 patients (87 men and 26 women) with a mean age of 49.84 years. Fifty-two patients survived, and 61 patients died. Liver failure (85.2%), sepsis (34.4%), and multiple organ failure (39.3%) were the main causes of death. Multivariate analyses showed that Acute Physiology and Chronic Health Evaluation (APACHE) II scores ≥ 12 [odds ratio (OR) = 7.160, 95% CI: 2.834-18.092, P < 0.001] and positive blood culture (OR = 13.520, 95% CI: 2.740-66.721, P = 0.001) on the day of diagnosis and model for end-stage liver disease (MELD) scores ≥ 28 (OR = 8.182, 95% CI: 1.884-35.527, P = 0.005) after the first week of treatment were independent predictors of mortality.

CONCLUSION: APACHE II scores on the day of diagnosis and MELD scores after the first week of anti-HBV therapy are feasible predictors of outcome in ACLF-HBV patients.

The utility of acute physiology and chronic health evaluation II scores for prediction of mortality among intensive care unit (ICU) and non-ICU patients with methicillin-resistant Staphylococcus aureus bacteremia

OBJECTIVE

Bloodstream infections due to methicillin-resistant Staphylococcus aureus (MRSA) have been associated with significant risk of in-hospital mortality. The acute physiology and chronic health evaluation (APACHE) II score was developed and validated for use among intensive care unit (ICU) patients, but its utility among non-ICU patients is unknown. The aim of this study was to determine the ability of APACHE II to predict death at multiple time points among ICU and non-ICU patients with MRSA bacteremia.

DESIGN

Retrospective cohort study.

PARTICIPANTS

Secondary analysis of data from 200 patients with MRSA bacteremia at 2 hospitals.

METHODS

Logistic regression models were constructed to predict overall in-hospital mortality and mortality at 48 hours, 7 days, 14 days, and 30 days using APACHE II scores separately in ICU and non-ICU patients. The performance of APACHE II scores was compared with age adjustment alone among all patients. Discriminatory ability was assessed using the c-statistic and was compared at each time point using χ(2) tests. Model calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test.

RESULTS

APACHE II was a significant predictor of death at all time points in both ICU and non-ICU patients. Discrimination was high in all models, with c-statistics ranging from 0.72 to 0.84, and was similar between ICU and non-ICU patients at all time points. APACHE II scores significantly improved the prediction of overall and 48-hour mortality compared with age adjustment alone.

CONCLUSIONS

The APACHE II score may be a valid tool to control for confounding or for the prediction of death among ICU and non-ICU patients with MRSA bacteremia.

Development of quality measures for monitoring and improving care in gastroenterology

Over the past decade, most quality assurance (QA) efforts in gastroenterology have been aimed at endoscopy. Endoscopic quality improvement was the rational area to begin QA work in gastroenterology due to the relatively acute nature of complications and the high volume of procedures performed. While endoscopy is currently the focus of most quality assurance (QA) measures in gastroenterology, more recent efforts have begun to address clinical gastroenterology practices both in the outpatient and inpatient settings. Clinical outpatient and inpatient gastroenterology is laden with areas where standardization could benefit patient care. While data and experience in clinical gastroenterology QA is relatively limited, it is clear that inconsistent use of guidelines and practice variations in gastroenterology can lead to lower quality care. In this review, we review a variety of areas in clinical gastroenterology where existing guidelines and published data suggest both the need and practicality of active QA measures.

Evaluation of the respiratory muscle strength of cirrhotic patients: relationship with Child-Turcotte-Pugh scoring system

Pulmonary abnormalities are observed in chronic hepatopathy. The measurement of the maximum inspiratory and expiratory pressure may evaluate lung function and the risks associated with hepatic transplantation. Thus, the present work sought to evaluate the respiratory muscle strength of 29 patients between 17 and 63 years old who were enrolled for liver transplantation. The patients were classified according to Child-Turcotte-Pugh score as A, B, or C, and also according to a physiotherapeutic evaluation, which included measurement of respiratory muscle strength by means of a digital manovacuometer, which determines the maximum inspiratory pressure (MaxIP) and the maximum expiratory pressure (MaxEP). The tests were performed with seated individuals having their nostrils obstructed by a nasal clip. The MaxIP was measured during the effort initiated in the residual volume, whereas the MaxEP was measured during the effort initiated in the total pulmonary capacity, keeping pressures stable for at least 1 second. The statistical analysis was performed through using the Mann-Whitney test with a 5% level of significance. The MaxIP values of Child A 95.5 +/- 40.507 cm H(2)O (average +/- DP) and Child B 87.2 +/- 35.02 patients were higher than those for Child C patients (34.83 +/- 3.68; P < .05). Similar results were observed for the MaxEP of Child A and B groups (116.25 +/- 31.98 and 97.28 +/- 31.08, respectively; P < .05), versus the Child C group (48.16 +/- 22.60). Between groups A and B, the MaxEP were similar (P > .05). We concluded that Child C patients display muscle weakness significantly greater than that of subjects classified as Child A or B.

[Mortality and prognostic factors of the cirrhotic patients with hepatic encephalopathy admitted to medical intensive care unit]

OBJECTIVE

To assess mortality and to identify variables that could predict it in cirrhotic patients hospitalized to the medical intensive care unit (MICU) for hepatic encephalopathy (HE).

STUDY DESIGN

Retrospective cohort study.

PATIENTS AND METHODS

From January 1995 to December 2004, the cirrhotic patients admitted consecutively in MICU were screened and those with altered level of consciousness were included. The MICU mortality rate was assessed. Nearly 80 variables were analyzed and compared between survivors and non-survivors.

STATISTICAL ANALYSIS

t test, chi(2) or Fisher exact tests, Kaplan-Meier and log rank, Cox regression analysis.

RESULTS

A total of 180 patients (42 women-138 men, mean age: 59+/-10 years) were admitted (incidence: 2.6%). The SAPS II was 30.1+/-11, Acute Physiology Age and Chronic Health Evaluation II (APACHE II): 16.5+/-5.3, Child-Pugh score: 9.1+/-1.9 and GCS: 11+/-2.8. The causes of liver cirrhosis was identified in 41.2% of cases (viral: 35.6%, alcohol: 5.6%). Nearly 18% of patients had an antecedent of HE. The causes of HE were: infection (65.6%), upper gastrointestinal bleeding (32.2%), drugs (5%) and metabolic cause (5%). MICU mortality rate was 33.3% and seemed higher in gastrointestinal bleeding. Eighteen variables were significantly associated with poor prognosis in univariate analysis. Only three variables remained significant in multivariate analysis: systolic blood pressure<90 mmHg (RR=4; IC 95%=2-8.1), total WBC>12000 n/mm(3) (RR=3.1; IC 95%=1.8-5.3) and use of mechanical ventilation (RR=3.1; IC 95%=1.7-5.6).

CONCLUSION

The MICU mortality of cirrhotic patients with HE was high and significantly associated with haemodynamic instability, hyperleucocytosis and mechanical ventilation.

Prediction of target range of intact parathyroid hormone in hemodialysis patients with artificial neural network

The application of artificial neural network (ANN) to predict outcome and explore potential relationships among clinical data is increasing being used in many clinical scenarios. The aim of this study was to validate whether an ANN is a useful tool for predicting the target range of plasma intact parathyroid hormone (iPTH) concentration in hemodialysis patients. An ANN was constructed with input variables collected retrospectively from an internal validation group (n = 129) of hemodialysis patients. Plasma iPTH was the dichotomous outcome variable, either target group (150 ng/L<or= iPTH <or=300 ng/L) or non-target group (iPTH< 150 ng/L or iPTH hormone >300 ng/L). After internal validation, the ANN was prospectively tested in an external validation group (n = 32) of hemodialysis patients. The final ANN was a multilayer perceptron network with six predictors including age, diabetes, hypertension, and blood biochemistries (hemoglobin, albumin, calcium). The externally validated ANN provided excellent discrimination as appraised by area under the receiver operating characteristic curve (0.83 +/- 0.11, p = 0.003). The Hosmer-Lemeshow statistic was 5.02 (p= 0.08 > 0.05) which represented a good-fit calibration. These results suggest that an ANN, which is based on limited clinical data, is able to accurately forecast the target range of plasma iPTH concentration in hemodialysis patients.