Sage Journals: Discover world-class research

Abstract

Introduction:

Very little artificial intelligence (AI) work has been performed to investigate acetaminophen-associated hepatotoxicity. The objective of this study was to develop an AI algorithm for analyzing weighted features for toxic hepatitis after acetaminophen poisoning.

Methods:

The medical records of 187 patients with acetaminophen poisoning treated at Chang Gung Memorial Hospital were reviewed. Patients were sorted into two groups according to their status of toxic hepatitis. A total of 40 clinical and laboratory features recorded on the first day of admission were selected for algorithm development. The random forest classifier (RFC) and logistic regression (LR) were used for artificial intelligence algorithm development.

Results:

The RFC-based AI model achieved the following results: accuracy = 92.5 ± 2.6%; sensitivity = 100%; specificity = 60%; precision = 92.3 ± 3.4%; and F1 = 96.0 ± 1.8%. The area under the receiver operating characteristic curve (AUROC) was approximately 0.98. The LR-based AI model achieved the following results: accuracy = 92.00 ± 2.9%; sensitivity = 100%; specificity = 20%; precision = 92.8 ± 3.4%; recall = 98.8 ± 3.4%; and F1 = 95.6 ± 1.5%. The AUROC was approximately 0.68. The weighted features were calculated, and the 10 most important weighted features for toxic hepatitis were aspartate aminotransferase (ALT), prothrombin time, alanine aminotransferase (AST), time to hospital, platelet count, lymphocyte count, albumin, total bilirubin, body temperature and acetaminophen level.

Conclusion:

The top five weighted features for acetaminophen-associated toxic hepatitis were ALT, prothrombin time, AST, time to hospital and platelet count.

Keywords

Artificial intelligence machine learning acetaminophen poisoning toxic hepatitis

Introduction

The popular phrase “artificial intelligence” (AI), first introduced in 1956 as “the science and engineering of making intelligent machines,” is applicable to a variety of items in medicine, such as the automated readings of chest radiographs and the detection of arrhythmias from wearable devices. What is so fascinating about AI is its potential to apply logical reasoning to data that is too large for the human mind to comprehend. An example of AI in toxicology is the prediction of irinotecan toxicity in metastatic colorectal cancer patients using a machine-learning approach with pharmacokinetic parameters.¹

Acetaminophen is one of the best-selling over-the-counter analgesic agents in the world. Although it is harmless at therapeutic doses, overdoses of acetaminophen can cause severe hepatic damage. Acetaminophen poisoning is the leading cause of acute liver injury and acute liver failure in the USA.² Most of the ingested acetaminophen is metabolized through glucuronidation and sulfation. Some, however, undergoes oxidative conversion through hepatic cytochromes, mainly cytochrome P450 2E1, generating the toxic metabolite N-acetyl-p-benzoquinone imine (NAPQI).³ Normally, NAPQI is converted rapidly to nontoxic metabolites by glutathione. However, under circumstances where glutathione is depleted, such as acetaminophen overdose and severe malnutrition, NAPQI could produce liver injury.⁴

The rationale for this study was based on an important but unsatisfactorily answered question that arose for many acetaminophen-poisoned patients treated at our hospital. Some patients suffered from severe toxic hepatitis and progressed rapidly to liver failure, whereas other patients suffered mild toxic hepatitis without overt clinical symptoms. Therefore, this raises the question of what are the best clinical predictors of toxic hepatitis? In addition, very little AI work has been performed to address this issue, which initiated our interest in this project. Although there have been studies on predictors for acetaminophen poisoning using statistical analysis (Table 1),^5–15 investigations on the predictors of toxic hepatitis using the AI approach are relatively lacking. Therefore, we aimed to develop an AI algorithm for analyzing weight features for toxic hepatitis after acetaminophen poisoning. This machine-learning platform might help provide an alternative way to identify risk factors for toxic hepatitis after acetaminophen poisoning.

Table 1.

Published literature on risk factors for toxic hepatitis after acetaminophen poisoning using different analytical approaches.

Study	Year	Sample size	Rate of toxic hepatitis, %	Risk factor or weight feature for toxic hepatitis	Analytical methods	Mortality rate, %
Current study	2021	187	11.8	ALT; prothrombin time; AST; time to hospital; platelet; lymphocyte; albumin; total bilirubin; body temperature; acetaminophen level	Artificial intelligence and statistical analysis	0.5
Hidaka et al.⁵	2020	102	19.6	Child-Pugh score ≥ B; age ≥ 70 years	Statistical analysis
Chomchai and Chomchai⁶	2018	197	20.3	BMI ≥ 25 kg/m²; Acetaminophen psi parameter ≥ 5 mM·hour	Statistical analysis
Marks et al.⁷	2017	545	3.7	Acetaminophen_pl:acetaminophen_t ratio ≥ 3	Statistical analysis
Levine et al.⁸	2016	304	31.6	Initial prothrombin time; Initial AST > 50 U/L	Statistical analysis
Wong et al.⁹	2015	3823	0.9	Acetaminophen level·ALT > 1500 mg/LU/L	Statistical analysis
Al-Hourani et al.¹⁰	2013	410	3.9	Initial ALT > 50 U/L	Statistical analysis
Chomchai and Chomchai¹¹	2014	255	12.5	Acetaminophen psi parameter ≥ 5 mM·hour; Acetaminophen level·AT > mg/LU/L	Statistical analysis
Nguyen et al.¹²	2008	210436	7.2	Hepatitis C virus; nonalcoholic fatty liver disease; alcoholic liver disease; malnutrition	Statistical analysis	8.6
Tsai et al.¹³	2004	75	12.0	Age; time to hospital	Statistical analysis	1.3
Alander et al.¹⁴	2000	322	8.4	Time to hospital > 24 hours; age 10–17 years; intentional ingestion; acetaminophen dose > 150 mg/kg; white race	Statistical analysis	0.3
Wang et al.¹⁵	1999	71	26.8	Alcohol consumption; acetaminophen dose	Statistical analysis	2.8

Note: Acetaminophen psi parameter = 11.542[APAP]4 h (e-0.173ti - e-0.173tf) - [APAP]threshold (tf - ti), [APAP]4 h = serum acetaminophen concentration (mg/L) at 4 hour postingestion, [APAP]threshold = 45 mg/L, tf = time (hour) of initiation of n-acetylcysteine, ti = 6 hours; Acetaminophen_pl plasma acetaminophen level; Acetaminophen_t threshold acetaminophen level at that time point on the treatment nomogram (based on a line through 100 mg/L at 4 hour) above which n-acetylcysteine would be administered; AST aspartate aminotransferase; ALT alanine aminotransferase; ALP alkaline phosphatase; AT aminotransferase; BMI body mass index; ICD-9-CM Classification of Diseases, Ninth Revision.

Materials and methods

Ethics statement

This retrospective observational study complied with the guidelines of the Declaration of Helsinki and was approved by the Medical Ethics Committee of Chang Gung Memorial Hospital, Linkou, Taiwan. The institutional review board number was 201800245B0.

Patients

We examined the medical records of 187 patients with intentional acetaminophen poisoning treated at Chang Gung Memorial Hospital between 2000 and 2011. Demographic, clinical, and laboratory data were collected for analysis. All patients were followed to investigate the risks of toxic hepatitis. The diagnoses of acetaminophen intoxication were based on clinical, physical and laboratory examinations and confirmed by blood acetaminophen levels (enzymatic method).

Inclusion and exclusion criteria

All adult patients aged 18 years and above who were diagnosed with acetaminophen poisoning at Chang Gung Memorial Hospital were eligible for inclusion in this study. Patients were excluded from analysis if they were younger than 18 years or did not have detectable acetaminophen levels in their blood despite a history of drug ingestion.

Clinical management

The protocols that were used to treat patients included gastric lavage with large amounts of normal saline, which was followed by the infusion of 1 g/kg activated charcoal and 250 mL magnesium citrate through a nasogastric tube. Magnesium citrate was used to prevent constipation after charcoal administration. The indication for N-acetylcysteine therapy was based on the Rumack-Matthew nomogram for acetaminophen poisoning to assess potential liver toxicity.¹⁶

Definitions

Acetaminophen-associated toxic hepatitis was diagnosed if the ALT increase was more than 10 times the upper reference limit.¹⁷ Acute respiratory failure has been defined as a condition of respiratory insufficiency requiring intubation and mechanical ventilation for more than 24 hours, regardless of the fraction of inspired oxygen.¹⁸

Statistical analysis

Data are expressed as the mean ± standard deviation or number (percentage), unless otherwise stated. Student’s t-test was used to compare the means of continuous variables and normally distributed data. Categorical data were analyzed with the chi-square test. All statistical tests were two-tailed, with P values <0.05 considered statistically significant. Data were analyzed with SPSS 23.0 software for Windows (SPSS, Inc., Chicago, IL, USA).

Data collection and feature engineering

Demographics, vital signs, laboratory values, interventions, medications, and nurse documentation were accessed through the hospital information system of Chang Gung Memorial Hospital. In the data preprocessing, we first excluded all data containing missing values. For missing values in the other recording data, we filled them with the median values of each corresponding feature. This process guarantees the quality and consistency of the data as well as the model. Next, we conducted standard normalization of every feature, alleviating the influence of a different range of values. We selected 40 clinical and laboratory features recorded on the first day of admission for use in algorithm development.

Model selection

We applied several machine-learning methods, such as random forest classifier (RFC), K-nearest neighbor classification (KNN), logistic regression (LR), support vector machine, XGBoost, and neural network, and used fivefold cross-validation for the assessment of model performance. Cross-validation is a resampling procedure used to evaluate machine-learning models, and “5” refers to the number of groups the given data sample was split into. We found that random forest classifier, a decision-tree-based algorithm, and logistic regression had the best performance. Therefore, we adopted the random forest classifier and logistic regression for the following artificial intelligence algorithm development.

Performance evaluation

After we used random forest classifier and logistic regression to fit the training dataset and obtained the probability threshold, we computed the toxic hepatitis probability of the testing dataset as well as the accuracy, recall (sensitivity), specificity, precision and F1 score given that specific threshold. More importantly, we computed the receiver operating characteristic (ROC) curve for performance evaluation. We also performed a feature weight analysis to differentiate the contribution of each feature to the performance of the algorithm to evaluate the algorithm’s fitness for the clinical scenario. Feature weighting is closely associated with the AUROC, sensitivity, and precision of an algorithm and can be obtained during cross-validation in model training.

Results

Among the 187 patients with intentional acetaminophen poisoning, 35 (18.7%) were male, and their mean age was 28.94 ± 12.12 years (Table 2). Patients with toxic hepatitis not only had higher blood acetaminophen levels (88.14 ± 85.33 versus 44.73 ± 33.64 μg/mL, P < 0.001) and longer times to hospital (12.50 ± 11.88 versus 5.34 ± 4.10 hours, P < 0.001) but also suffered a higher incidence of respiratory failure (18.1 versus 1.8%, P = 0.004) than patients without toxic hepatitis. Laboratory analysis found higher blood concentrations of white blood cell count (10722.73 ± 7603.504 versus 8813.94 ± 3119.22 per μL, P = 0.032), AST (847.45 ± 2926.92 versus 27.67 ± 19.64 U/L, P < 0.001), ALT (450.50 ± 1014.10 versus 21.06 ± 20.83 U/L, P < 0.001), total bilirubin (1.35 ± 0.81 versus 0.88 ± 0.29 mg/dL, P < 0.001), bicarbonate (22.00 ± 4.36 versus 21.93 ± 1.25 mEq/L, P = 0.002), and prothrombin time (1.59 ± 0.80 versus 1.09 ± 0.06 INR, P < 0.001) in patients with toxic hepatitis than in those without toxic hepatitis. Furthermore, the blood concentration of platelets was lower in patients with toxic hepatitis than in those without toxic hepatitis (21.62 ± 10.70 versus 25.02 ± 5.83 10³ per μL, P = 0.024).

Table 2.

Clinical data of patients with acetaminophen poisoning, stratified according to the status of toxic hepatitis (n = 187).

Variable	Patients with toxic hepatitis (n = 22)	Patients without toxic hepatitis (n = 165)	P value
Baseline characteristics
Age, year	29.32 ± 10.683	28.88 ± 12.318	0.875
Male gender, n (%)	7 (31.8)	28 (17.0)	0.140
Suicide attempt, n (%)	22 (100)	165 (100)	1.000
Acetaminophen ingested dose, mg	44.73 ± 24.32	35.16 ± 23.58	0.075
Acetaminophen level, μg/mL	88.14 ± 85.33	44.73 ± 33.64	< 0.001***
Concomitant benzodiazepine ingestion, n (%)	3 (13.6)	37 (22.4)	0.420
Smoking habit, n (%)	4 (18.2)	25 (15.2)	0.754
Alcohol consumption, n (%)	4 (18.2)	31 (18.8)	0.605
Hypertension, n (%)	2 (9.1)	4 (2.4)	0.148
Diabetes mellitus, n (%)	0 (0)	3 (1.8)	1.000
Depressive disorder, n (%)	14 (63.6)	111 (67.3)	0.810
Time to hospital, hour	12.50 ± 11.88	5.34 ± 4.10	<0.001***
Clinical manifestations
Body temperature, °C	36.66 ± 0.81	36.95 ± 4.04	0.741
Systolic blood pressure, mmHg	127.68 ± 23.06	119.42 ± 15.79	0.031*
Diastolic blood pressure, mmHg	76.88 ± 14.35	71.61 ± 11.55	0.053
Pulse rate, per minute	81.50 ± 16.91	81.75 ± 14.88	0.942
Consciousness disturbance, n (%)	6 (27.3)	35 (21.2)	0.583
Respiratory failure, n (%)	4 (18.2)	3 (1.8)	0.004**
Laboratory findings
White blood cell count, per μL	10722.73 ± 7603.504	8813.94 ± 3119.22	0.032*
Hemoglobin, g/dL	13.62 ± 1.71	13.36 ± 1.38	0.434
Platelet count, 10³ per μL	21.62 ± 10.70	25.02 ± 5.83	0.024*
Neutrophil count, per μL	77.68 ± 13.11	73.17 ± 12.75	0.122
Lymphocyte count, per μL	16.93 ± 12.72	21.41 ± 10.91	0.078
Eosinophil count, per μL	0.23 ± 0.445	0.56 ± 1.162	0.192
Glucose, mg/L	123.77 ± 24.19	125.79 ± 22.00	0.690
Urea nitrogen, mg/dL	8.87 ± 3.91	8.26 ± 2.77	0.358
Creatinine, mg/dL	0.79 ± 0.22	0.770 ± 0.20	0.587
Sodium, mEq/L	140.45 ± 2.96	140.08 ± 2.84	0.569
Potassium, mEq/L	3.36 ± 0.37	3.56 ± 1.51	0.530
Bicarbonate, mEq/L	22.00 ± 4.36	21.93 ± 1.25	0.002**
AST, U/L	847.45 ± 2926.92	27.67 ± 19.64	<0.001***
ALT, U/L	450.50 ± 1014.10	21.06 ± 20.83	<0.001***
Total bilirubin, mg/dL	1.35 ± 0.81	0.88 ± 0.29	<0.001***
Albumin, g/dL	3.85 ± 0.51	3.93 ± 0.82	0.244
Prothrombin time, INR	1.59 ± 0.80	1.09 ± 0.06	<0.001***
Treatment
Gastric lavage and active charcoal therapy, n (%)	10 (45.5)	118 (71.5)	0.025*
N-acetylcysteine therapy, n (%)	21 (95.5)	134 (81.2)	0.132

Note: AST aspartate aminotransferase; ALT alanine aminotransferase; *P < 0.05; ** P < 0.01; *** P < 0.001.

The algorithm development process included three steps (Table 3). During feature engineering, the 187 patients were divided into training data (n = 149) and testing data (n = 38). The RFC-based AI model was developed using 80% of the records as the training cohort and 20% of the records as the testing cohort. The established algorithm achieved the following results: accuracy = 92.5 ± 2.6%, sensitivity = 100%, specificity = 60%, precision = 92.3 ± 3.4%, and F1 = 96.0 ± 1.8%. The AUROC was approximately 0.98 (Figure 1).

Figure 1.

The receiver operating characteristic (ROC) curve. The random forest classifier (RFC)-based artificial intelligence (AI) model achieved the following results: accuracy = 92.5 ± 2.6%; sensitivity = 100%; specificity = 60%; precision = 92.3 ± 3.4%; and F1 = 96.0 ± 1.8%. The area under the receiver operating characteristic curve (AUROC) was approximately 0.98.

Table 3.

Identification of the top 10 weighted features for toxic hepatitis using an artificial intelligence approach.

Feature variable	Weight
ALT	0.10989196
Prothrombin time	0.10960985
AST	0.10497889
Time to hospital	0.05078998
Platelet count	0.05067652
Lymphocyte count	0.0420866
Albumin	0.04114908
Total bilirubin	0.04077905
Body temperature	0.03726995
Acetaminophen level	0.03409645

Note: AST aspartate aminotransferase; ALT alanine aminotransferase.

To compare the diagnostic performance of the RFC-based AI algorithm to the LR-based AI algorithm, the same 187 patients were used to evaluate the performance of toxic hepatitis using logistic regression. The LR-based AI model was developed using 80% of the records as the training cohort and 20% of the records as the testing cohort. The established algorithm achieved the following results: accuracy = 92.00 ± 2.9%, sensitivity = 100%, specificity = 20%, precision = 92.8 ± 3.4%, recall = 98.8 ± 3.4%, and F1 = 95.6 ± 1.5%. The AUROC was approximately 0.68. The weight features were calculated, and the top 10 were listed according to their categories (Table 3). The weight represents the influence of each feature. ALT, prothrombin time, AST, time to hospital and platelet count were the most highly weighted.

Discussion

These data are important because this is one of the few studies where an AI algorithm has been applied to analyze weight features for acetaminophen-associated toxic hepatitis. Using a binary mixed model tree artificial intelligence approach,¹⁹ it was demonstrated that the three most important weight features for the daily outcome of patients with acetaminophen-induced acute liver failure were the use of pressors and bilirubin and creatinine levels. On the other hand, our analysis showed that the 10 most important weighted features for toxic hepatitis were ALT, prothrombin time, AST, time to hospital, platelet count, lymphocyte count, albumin, total bilirubin, body temperature and acetaminophen level. As shown in Table 1,^5–15 several previous studies have identified risk factors for acetaminophen-associated toxic hepatitis using a traditional statistical analysis approach. In contrast, our data were analyzed using an AI algorithm to identify the top 10 weighted features for toxic hepatitis.

Liver disease is often reflected by abnormalities in liver function tests. While aminotransferase and bilirubin represent hepatocyte integrity and cholestasis, respectively, albumin and prothrombin time indicate liver function mass.¹⁷ Several biomarkers, such as ALT, prothrombin time, AST, albumin and total bilirubin, ranked in the top 10 feature weights in our study. Aminotransferases, which are highly concentrated in the liver, are enzymes that catabolize amino acids, allowing them to enter the citric acid cycle. While AST is also represented in kidneys, brain, skeletal muscle and red blood cells, ALT has a lower concentration in skeletal muscle and kidneys. In one retrospective study including 304 subjects,⁸ initial AST > 50 U/L predicted hepatic injury in patients with acetaminophen overdose with a sensitivity of 79.5% and a specificity of 82.6%. In another study including 410 subjects,¹⁰ initial ALT > 50 U/L predicted hepatic injury in patients with acetaminophen overdose with a sensitivity of 100% and a specificity of 86%.

Prothrombin time measures the extrinsic pathway of coagulation and represents the synthetic capacity of the liver. In one retrospective study including 304 subjects,⁸ prothrombin time predicted hepatic injury in patients with acetaminophen overdose with a sensitivity of 82.1% and a specificity of 63.6%.⁸

Albumin is synthesized by the hepatocytes. A variety of factors regulate the synthesis of albumin, including nutritional status, cytokines and hormones.²⁰ In one study,²¹ short-term fasting increased acetaminophen exposure in healthy subjects administered oral acetaminophen at 1000 mg. However, in another study including 197 subjects,⁶ overweight-obesity status was a significant risk factor for acute liver injury secondary to acute acetaminophen overdose.

Bilirubin is the product of heme metabolism. It binds to albumin in the plasma and is transported throughout the body. In hepatocytes, bilirubin is conjugated with glucuronic acid by glucuronosyltransferase. In one animal study,²² inhibition of hepatic bilirubin conjugation and excretion of conjugated bilirubin into the bile was observed after acetaminophen administration to rats.

Platelet count ranked as the fifth feature weight in this study. The maintenance of hematopoietic stem cells requires hepatic thrombopoietin.²³ Through a series of divisions, hematopoietic stem cells yield each of the hematopoietic lineages, including megakaryocytes, which ultimately give rise to platelets. Thrombopoietin is also involved in megakaryocyte development.²⁴ In one animal study,²⁵ acetaminophen poisoning caused the depletion of blood platelets and the accumulation of platelets in the liver. Notably, platelets have been found to contribute to the progression of acetaminophen-induced liver injury.²⁵

Time to hospital ranked as the fourth feature weight in our study. In one study including 322 pediatric subjects,¹⁴ hepatocellular injury was associated with a presentation longer than 24 hours after ingestion. In another study including 75 subjects,¹³ time to hospital was a risk factor for acetaminophen-induced hepatotoxicity. The acetaminophen psi parameter takes into account the acetaminophen level at 4 hours postingestion and the time to initiation of n-acetylcysteine therapy. Chomchai and Chomchai demonstrated in a retrospective study of 255 subjects¹¹ that the acetaminophen psi parameter predicts hepatotoxicity in overdose patients with a sensitivity of 96.9% and a specificity of 91.5%.

Lymphocyte count ranked as the sixth feature weight in our study. This is reasonable because acetaminophen overdose is always linked with blood toxicity. Rat lymphocytes carry cofactors and enzymes involved in the mechanism of acetaminophen hepatotoxicity,^26,27 and furthermore, a rat study also demonstrated that blood lymphocyte count decreases during acetaminophen hepatotoxicity.²⁸

Body temperature ranked as the ninth feature weight in our study. Acetaminophen is widely used to treat fever. It is assumed that the antipyretic action of acetaminophen is mediated through the cyclooxygenase pathway.²⁹ In one animal study,³⁰ it was demonstrated that mild hypothermia attenuates liver injury and benefits survival in mice with acetaminophen toxicity. A previous study³¹ revealed that hypothermia favors functional mitochondria and decreases oxidative stress to fight acetaminophen-induced hepatotoxicity.

Blood acetaminophen level ranked as the 10th feature weight in our study. Both the acetaminophen psi parameter and acetaminophen-aminotransferase multiplication product take the acetaminophen level into account. Several studies have demonstrated that they can predict acetaminophen hepatotoxicity.^6,9,11 Another retrospective observational study included 545 subjects,⁷ where acetaminophen_pl was defined as the acetaminophen level and acetaminophen_t was defined as the threshold acetaminophen level at that time point on the treatment nomogram (based on a line through 100 mg/L at 4 hours) above which n-acetylcysteine was administered. An acetaminophen_pl:acetaminophen_t ratio ≥3 predicted acetaminophen hepatotoxicity.

This AI algorithm has many potential applications in analyzing cases of poisoning. The machine-learning platform offers the following advantages over traditional statistical analysis. First, the variables are readily available in clinical practice and can be obtained upon admission. Second, the outcome can be predicted within a very short period of time using a simple laptop. Third, since the AI algorithm uses feature weighting to approximate the influence of individual features, it is not necessary to discard any variables as in statistical analysis because no null hypothesis is assumed. Fourth, the receiver operating characteristic curve obtained by artificial intelligence is based on multiple parameters, rather than a single parameter as in statistical analysis.

Finally, the limitations of this study are its small sample size and short follow-up duration. Further studies are warranted.

In summary, this study developed an AI algorithm to predict weighted features for toxic hepatitis after acetaminophen poisoning. The top five weighted features for toxic hepatitis are ALT, prothrombin time, AST, time to hospital and platelet count.

Footnotes

Author contributions

Data collection and manuscript writing JSY; Data analysis CCH, WHH; Patient care and management WHH, CWH, THY; Study design and supervision CHW.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by research grants from Chang Gung Memorial Hospital (CMRPG5J0201, CORPG3K0191).

ORCID iDs

T-H Yen

C-H Weng

References

Oyaga-Iriarte

Insausti

Sayar

, et al. Prediction of irinotecan toxicity in metastatic colorectal cancer patients based on machine learning models with pharmacokinetic parameters. J Pharmacol Sci 2019; 140: 20–25.

Larson

Polson

Fontana

, et al.

Acetaminophen-induced acute liver failure: results of a United States multicenter, prospective study

Hepatology. 2005; 42: 1364–1372.

Larson

. Acetaminophen hepatotoxicity. Clin Liver Dis 2007; 11: 525–548, vi.

Jaeschke

Williams

Ramachandran

, et al. Acetaminophen hepatotoxicity and repair: the role of sterile inflammation and innate immunity. Liver Int 2012; 32: 8–20.

Hidaka

Kaji

Takatori

, et al. Risk factors for acetaminophen-induced liver injury: a single-center study from Japan. Clin Ther 2020; 42: 704–710.

Chomchai

. Being overweight or obese as a risk factor for acute liver injury secondary to acute acetaminophen overdose. Pharmacoepidemiol Drug Saf 2018; 27: 19–24.

Marks

DJB

Dargan

Archer

JRH

, et al. Outcomes from massive paracetamol overdose: a retrospective observational study. Br J Clin Pharmacol 2017; 83: 1263–1272.

Levine

O’Connor

Padilla-Jones

, et al. Comparison of prothrombin time and aspartate aminotransferase in predicting hepatotoxicity after acetaminophen overdose. J Med Toxicol 2016; 12: 100–106.

Wong

Sivilotti

Dargan

, et al. External validation of the paracetamol-aminotransferase multiplication product to predict hepatotoxicity from paracetamol overdose. Clin Toxicol (Phila) 2015; 53: 807–814.

10.

Al-Hourani

Mansi

Pettie

, et al. The predictive value of hospital admission serum alanine transaminase activity in patients treated for paracetamol overdose. QJM 2013; 106: 541–546.

11.

Chomchai

. Predicting acute acetaminophen hepatotoxicity with acetaminophen-aminotransferase multiplication product and the Psi parameter. Clin Toxicol (Phila). 2014; 52: 506–511.

12.

Nguyen

Sam

Thuluvath

. Hepatitis C is a predictor of acute liver injury among hospitalizations for acetaminophen overdose in the United States: a nationwide analysis. Hepatology 2008; 48: 1336–1341.

13.

Tsai

Chang

Weng

, et al. Acute acetaminophen intoxication in Taiwan: outcomes and risk factors. J Formos Med Assoc 2004; 103: 830–835.

14.

Alander

Dowd

Bratton

, et al. Pediatric acetaminophen overdose: risk factors associated with hepatocellular injury. Arch Pediatr Adolesc Med 2000; 154: 346–350.

15.

Wang

Huang

Deng

, et al. Characteristics and risk factors of acetaminophen-induced hepatitis in Taiwan. Zhonghua Yi Xue Za Zhi (Taipei). 1999; 62: 369–375.

16.

Wolf

Heard

Sloan

, et al. Clinical policy: critical issues in the management of patients presenting to the emergency department with acetaminophen overdose. Ann Emerg Med 2007; 50: 292–313.

17.

Giannini

Testa

Savarino

. Liver enzyme alteration: a guide for clinicians. CMAJ. 2005; 172: 367–379.

18.

Luhr

Antonsen

Karlsson

, et al. Incidence and mortality after acute respiratory failure and acute respiratory distress syndrome in Sweden, Denmark, and Iceland. The ARF Study Group. Am J Respir Crit Care Med 1999; 159: 1849–1861.

19.

Speiser

Karvellas

Wolf

, et al. Predicting daily outcomes in acetaminophen-induced acute liver failure patients with machine learning techniques. Comput Methods Programs Biomed 2019; 175: 111–120.

20.

Levitt

. Human serum albumin homeostasis: a new look at the roles of synthesis, catabolism, renal and gastrointestinal excretion, and the clinical value of serum albumin measurements. Int J Gen Med 2016; 9: 229–255.

21.

Achterbergh

Lammers

Kuijsten

, et al. Effects of nutritional status on acetaminophen measurement and exposure. Clin Toxicol (Phila) 2019; 57: 42–49.

22.

Davis

Ideo

Harrison

, et al. Early inhibition of hepatic bilirubin conjugation after paracetamol (acetaminophen) administration in the rat. Digestion 1975; 13: 42–48.

23.

Decker

Leslie

Liu

, et al. Hepatic thrombopoietin is required for bone marrow hematopoietic stem cell maintenance. Science 2018; 360: 106–110.

24.

Deutsch

Tomer

. Megakaryocyte development and platelet production. Br J Haematol 2006; 134: 453–466.

25.

Miyakawa

Joshi

Sullivan

, et al. Platelets and protease-activated receptor-4 contribute to acetaminophen-induced liver injury in mice. Blood 2015; 126: 1835–1843.

26.

Dey

Dhawan

Kishore Seth

, et al. Evidence for cytochrome P450 2E1 catalytic activity and expression in rat blood lymphocytes. Life Sci 2005; 77: 1082–1093.

27.

Hannon-Fletcher

Barnett

. Lymphocyte cytochrome P450 expression: inducibility studies in male Wistar rats. Br J Biomed Sci 2008; 65: 1–6.

28.

Masson

Peterson

Chung

, et al. Lymphocyte loss and immunosuppression following acetaminophen-induced hepatotoxicity in mice as a potential mechanism of tolerance. Chem Res Toxicol 2007; 20: 20–26.

29.

Anderson

. Paracetamol (Acetaminophen): mechanisms of action. Paediatr Anaesth 2008; 18: 915–921.

30.

Vaquero

Bélanger

James

, et al. Mild hypothermia attenuates liver injury and improves survival in mice with acetaminophen toxicity. Gastroenterology. 2007; 132: 372–383.

31.

Tan

. Hypothermia advocates functional mitochondria and alleviates oxidative stress to combat acetaminophen-induced hepatotoxicity. Cells 2020; 9: 2354.

An artificial intelligence algorithm for analyzing acetaminophen-associated toxic hepatitis