Decision tree outcome prediction of acute acetaminophen exposure in the United States: A study of 30,000 cases from the National Poison Data System

Machine learning for predicting medical outcomes associated with acute lithium poisoning

The use of machine learning algorithms and artificial intelligence in medicine has attracted significant interest due to its ability to aid in predicting medical outcomes. This study aimed to evaluate the effectiveness of the random forest algorithm in predicting medical outcomes related to acute lithium toxicity. We analyzed cases recorded in the National Poison Data System (NPDS) between January 1, 2014, and December 31, 2018. We highlighted instances of acute lithium toxicity in patients with ages ranging from 0 to 89 years. A random forest model was employed to predict serious medical outcomes, including those with a major effect, moderate effect, or death. Predictions were made using the pre-defined NPDS coding criteria. The model's predictive performance was assessed by computing accuracy, recall (sensitivity), and F1-score. Of the 11,525 reported cases of lithium poisoning documented during the study, 2,760 cases were categorized as acute lithium overdose. One hundred thirty-nine individuals experienced severe outcomes, whereas 2,621 patients endured minor outcomes. The random forest model exhibited exceptional accuracy and F1-scores, achieving values of 99%, 98%, and 98% for the training, validation, and test datasets, respectively. The model achieved an accuracy rate of 100% and a sensitivity rate of 96% for important results. In addition, it achieved a 96% accuracy rate and a sensitivity rate of 100% for minor outcomes. The SHapley Additive exPlanations (SHAP) study found factors, including drowsiness/lethargy, age, ataxia, abdominal pain, and electrolyte abnormalities, significantly influenced individual predictions. The random forest algorithm achieved a 98% accuracy rate in predicting medical outcomes for patients with acute lithium intoxication. The model demonstrated high sensitivity and precision in accurately predicting significant and minor outcomes. Further investigation is necessary to authenticate these findings.

Predictive modeling of methadone poisoning outcomes in children ≤ 5 years: utilizing machine learning and the National Poison Data System for improved clinical decision-making

The escalating therapeutic use of methadone has coincided with an increase in accidental ingestions, particularly among children ≤ 5 years. This study utilized machine learning (ML) methodologies on data from the National Poison Data System (NPDS) to predict pediatric methadone poisoning outcomes to enhance clinical decision-making. We analyzed 140 medical parameters from pediatric patient records. Pre-processing steps, including synthetic oversampling, addressed the imbalanced distribution of the outcome variable. We evaluated various ML models in multiclass classification tasks. Random forest showed versatility with an accuracy of 0.96 and a strong receiver operating characteristic area under the curve (ROC AUC) (0.98). Meanwhile, the support vector machine (SVM) had the highest negative predictive value (NPV) (0.64). Shapley Additive exPlanation (SHAP) analysis identified key predictors such as coma, cyanosis, respiratory arrest, and respiratory depression for predicting serious outcomes.

CONCLUSION

This research emphasizes the utility of ML in clinical settings for early detection and intervention in methadone poisoning events in children, highlighting the synergy between data science and clinical expertise.

WHAT IS KNOWN

• The increased use of methadone for treatment has been associated with a rise in accidental ingestions, particularly in children under five years old. • Methadone poisoning in young children can lead to severe outcomes, including respiratory depression and coma, requiring urgent medical intervention.

WHAT IS NEW

• Machine learning models, particularly Random Forest and Bagging, outperform traditional methods in predicting methadone poisoning outcomes in children. • SHAP analysis provides novel insights into key predictors of severe outcomes, enabling improved clinical decision-making and risk stratification.

The presence of abdominal pain associated with acetaminophen overdose does not predict severity of liver injury

AIM

Whilst it is known that abdominal pain is a common symptom in patients with acetaminophen overdose, its association with severity of liver injury has not been clearly defined. This study investigates the association between the symptom of abdominal pain on presentation to hospital and the degree of liver injury post-acetaminophen overdose.

METHODS

Admissions with acetaminophen poisoning, requiring treatment with acetylcysteine were identified and reviewed from a search of a large Australian tertiary hospital network from February 20th, 2014, to August 30th, 2018. Parameters such as presence of abdominal pain, time post-ingestion and peak ALT were collected. Single acute ingestions, staggered and repeated supratherapeutic ingestions were analysed.

RESULTS

539 cases were identified in the study period, 79% female, with mean age 25 (17-43) years. Patients presenting to the emergency department with abdominal pain post-acetaminophen overdose had a similar risk of developing hepatotoxicity or acute liver injury compared to patients without abdominal pain regardless of time to presentation. Patients presenting <8-h post-overdose with abdominal pain were as likely to develop hepatotoxicity (1/46, 2.2%) compared to those without abdominal pain (1/54 [1.9%]; OR = 1.18 [0.07 to 19.4]). Those presenting >8-h post-overdose with abdominal pain were as likely to develop hepatotoxicity (13/92, 14.1%) compared to those without abdominal pain (4/35 [11.4%]; OR = 1.28 [0.39 to 4.21]).

CONCLUSIONS

The presence of abdominal pain after acetaminophen overdose was not predictive of the development of liver injury in patients receiving acetylcysteine treatment. Further prospective studies are required to confirm this finding. The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Prediction of acute methanol poisoning prognosis using machine learning techniques

Methanol poisoning is a global public health concern, especially prevalent in developing nations. This study focuses on predicting the severity of methanol intoxication using machine learning techniques, aiming to improve early identification and prognosis assessment. The study, conducted at Loghman Hakim Hospital in Tehran, Iran. The data pertaining to individuals afflicted with methanol poisoning was retrieved retrospectively and divided into training and test groups at a ratio of 70:30. The selected features were then inputted into various machine learning methods. The models were implemented using the Scikit-learn library in the Python programming language. Ultimately, the efficacy of the developed models was assessed through ten-fold cross-validation techniques and specific evaluation criteria, with a confidence level of 95%. A total number of 897 patients were included and divided in three groups including without sequel (n = 573), with sequel (n = 234), and patients who died (n = 90). The two-step feature selection was yielded 43 features in first step and 23 features in second step. In best model (Gradient Boosting Classifier) test dataset metric by 32 features younger age, higher methanol ingestion, respiratory symptoms, lower GCS scores, type of visual symptom, duration of therapeutic intervention, ICU admission, and elevated CPK levels were among the most important features predicting the prognosis of methanol poisoning. The Gradient Boosting Classifier demonstrated the highest predictive capability, achieving AUC values of 0.947 and 0.943 in the test dataset with 43 and 23 features, respectively. This research introduces a machine learning-driven prognostic model for methanol poisoning, demonstrating superior predictive capabilities compared to traditional statistical methods. The identified features provide valuable insights for early intervention and personalized treatment strategies.

The value of machine learning for prognosis prediction of diphenhydramine exposure: National analysis of 50,000 patients in the United States

Background

Diphenhydramine (DPH) is an antihistamine medication that in overdose can result in anticholinergic symptoms and serious complications, including arrhythmia and coma. We aimed to compare the value of various machine learning (ML) models, including light gradient boosting machine (LGBM), logistic regression (LR), and random forest (RF), in the outcome prediction of DPH poisoning.

Materials and Methods

We used the National Poison Data System database and included all of the human exposures of DPH from January 01, 2017 to December 31, 2017, and excluded those cases with missing information, duplicated cases, and those who reported co-ingestion. Data were split into training and test datasets, and three ML models were compared. We developed confusion matrices for each, and standard performance metrics were calculated.

Results

Our study population included 53,761 patients with DPH exposure. The most common reasons for exposure, outcome, chronicity of exposure, and formulation were captured. Our results showed that the average precision-recall area under the curve (AUC) of 0.84. LGBM and RF had the highest performance (average AUC of 0.91), followed by LR (average AUC of 0.90). The specificity of the models was 87.0% in the testing groups. The precision of models was 75.0%. Recall (sensitivity) of models ranged between 73% and 75% with an F1 score of 75.0%. The overall accuracy of LGBM, LR, and RF models in the test dataset was 74.8%, 74.0%, and 75.1%, respectively. In total, just 1.1% of patients (mostly those with major outcomes) received physostigmine.

Conclusion

Our study demonstrates the application of ML in the prediction of DPH poisoning.

Deep learning neural network derivation and testing to distinguish acute poisonings

INTRODUCTION

Acute poisoning is a significant global health burden, and the causative agent is often unclear. The primary aim of this pilot study was to develop a deep learning algorithm that predicts the most probable agent a poisoned patient was exposed to from a pre-specified list of drugs.

RESEARCH DESIGN & METHODS

Data were queried from the National Poison Data System (NPDS) from 2014 through 2018 for eight single-agent poisonings (acetaminophen, diphenhydramine, aspirin, calcium channel blockers, sulfonylureas, benzodiazepines, bupropion, and lithium). Two Deep Neural Networks (PyTorch and Keras) designed for multi-class classification tasks were applied.

RESULTS

There were 201,031 single-agent poisonings included in the analysis. For distinguishing among selected poisonings, PyTorch model had specificity of 97%, accuracy of 83%, precision of 83%, recall of 83%, and a F1-score of 82%. Keras had specificity of 98%, accuracy of 83%, precision of 84%, recall of 83%, and a F1-score of 83%. The best performance was achieved in the diagnosis of single-agent poisoning in diagnosing poisoning by lithium, sulfonylureas, diphenhydramine, calcium channel blockers, then acetaminophen, in PyTorch (F1-score = 99%, 94%, 85%, 83%, and 82%, respectively) and Keras (F1-score = 99%, 94%, 86%, 82%, and 82%, respectively).

CONCLUSION

Deep neural networks can potentially help in distinguishing the causative agent of acute poisoning. This study used a small list of drugs, with polysubstance ingestions excluded.Reproducible source code and results can be obtained at https://github.com/ashiskb/npds-workspace.git.

Paracetamol (acetaminophen) overdose and hepatotoxicity: mechanism, treatment, prevention measures, and estimates of burden of disease

INTRODUCTION

Paracetamol is one of the most used medicines worldwide and is the most common important poisoning in high-income countries. In overdose, paracetamol causes dose-dependent hepatotoxicity. Acetylcysteine is an effective antidote, however despite its use hepatotoxicity and many deaths still occur.

AREAS COVERED

This review summarizes paracetamol overdose and toxicity (including mechanisms, risk factors, risk assessment, and treatment). In addition, we summarize the epidemiology of paracetamol overdose worldwide. A literature search on PubMed for poisoning epidemiology and mortality from 1 January 2017 to 26 October 2022 was performed to estimate rates of paracetamol overdose, liver injury, and deaths worldwide.

EXPERT OPINION

Paracetamol is widely available and yet is substantially more toxic than other analgesics available without prescription. Where data were available, we estimate that paracetamol is involved in 6% of poisonings, 56% of severe acute liver injury and acute liver failure, and 7% of drug-induced liver injury. These estimates are limited by lack of available data from many countries, particularly in Asia, South America, and Africa. Harm reduction from paracetamol is possible through better identification of high-risk overdoses, and better treatment regimens. Large overdoses and those involving modified-release paracetamol are high-risk and can be targeted through legislative change.

Utility of support vector machine and decision tree to identify the prognosis of metformin poisoning in the United States: analysis of National Poisoning Data System

BACKGROUND

With diabetes incidence growing globally and metformin still being the first-line for its treatment, metformin's toxicity and overdose have been increasing. Hence, its mortality rate is increasing. For the first time, we aimed to study the efficacy of machine learning algorithms in predicting the outcome of metformin poisoning using two well-known classification methods, including support vector machine (SVM) and decision tree (DT).

METHODS

This study is a retrospective cohort study of National Poison Data System (NPDS) data, the largest data repository of poisoning cases in the United States. The SVM and DT algorithms were developed using training and test datasets. We also used precision-recall and ROC curves and Area Under the Curve value (AUC) for model evaluation.

RESULTS

Our model showed that acidosis, hypoglycemia, electrolyte abnormality, hypotension, elevated anion gap, elevated creatinine, tachycardia, and renal failure are the most important determinants in terms of outcome prediction of metformin poisoning. The average negative predictive value for the decision tree and SVM models was 92.30 and 93.30. The AUC of the ROC curve of the decision tree for major, minor, and moderate outcomes was 0.92, 0.92, and 0.89, respectively. While this figure of SVM model for major, minor, and moderate outcomes was 0.98, 0.90, and 0.82, respectively.

CONCLUSIONS

In order to predict the prognosis of metformin poisoning, machine learning algorithms might help clinicians in the management and follow-up of metformin poisoning cases.