ToxNet: an artificial intelligence designed for decision support for toxin prediction.

BACKGROUND: Artificial intelligences (AIs) are emerging in the field of medical informatics in many areas. They are mostly used for diagnosis support in medical imaging but have potential uses in many other fields of medicine where large datasets are available. AIM: To develop an artificial intelligence (AI) "ToxNet", a machine-learning based computer-aided diagnosis (CADx) system, which aims to predict poisons based on patient's symptoms and metadata from our Poison Control Center (PCC) data. To prove its accuracy and compare it against medical doctors (MDs). METHODS: The CADx system was developed and trained using data from 781,278 calls recorded in our PCC database from 2001 to 2019. All cases were mono-intoxications. Patient symptoms and meta-information (e.g., age group, sex, etiology, toxin point of entry, weekday, etc.) were provided. In the pilot phase, the AI was trained on 10 substances, the AI's prediction was compared to naïve matching, literature matching, a multi-layer perceptron (MLP), and the graph attention network (GAT). The trained AI's accuracy was then compared to 10 medical doctors in an individual and in an identical dataset. The dataset was then expanded to 28 substances and the predictions and comparisons repeated. RESULTS: In the pilot, the prediction performance in a set of 8995 patients with 10 substances was 0.66 ± 0.01 (F1 micro score). Our CADx system was significantly superior to naïve matching, literature matching, MLP, and GAT (p < 0.005). It outperformed our physicians experienced in clinical toxicology in the individual and identical dataset. In the extended dataset, our CADx system was able to predict the correct toxin in a set of 36,033 patients with 28 substances with an overall performance of 0.27 ± 0.01 (F1 micro score), also significantly superior to naïve matching, literature matching, MLP, and GAT. It also outperformed our MDs. CONCLUSION: Our AI trained on a large PCC database works well for poison prediction in these experiments. With further research, it might become a valuable aid for physicians in predicting unknown substances and might be the first step into AI use in PCCs.

An internet-based survey of 96 German-speaking users of "bath salts": frequent complications, risky sexual behavior, violence, and delinquency.

OBJECTIVE: To define the demographics of German-speaking "bath salt" users. DESIGN: Prospective web-based survey of volunteer users of "bath salts". Subject recruitment/exclusion: Participation was solicited by posts in web forums frequented by users of synthetic cathinones. An invitation to participate was also disseminated via regional drug information centers. Responses were discarded if participants refused data analysis, provided incomplete surveys, were under 18 years of age (five cases), and in case of clearly improbable answers (i.e., two cases with profanity typed in free-form input fields). Overall 96 out of 180 participants provided complete questionnaires. These were further analyzed. RESULTS AND CONCLUSIONS: 74% of respondents were male. 41% were under the age of 30 and a further 38% between 30 and 39 years old. Cathinones were used on more than 10 days in the preceding year by 62% of study subjects. The nasal and intravenous routes of administration were most often used. About 80% of respondents reported binge use. There were frequent co-administrations of opioids and opiates. The most common complication was prolonged confusion (47%). 16% had been involuntarily confined. One third had thoughts of violence and 16% acted on these thoughts either against themselves or others. About 44% reported high-risk sexual activity under the influence of cathinones. About 31% had driven or ridden a bike while intoxicated. About 6% had problems with law-enforcement for selling cathinones and 16% for crimes committed under the influence of cathinones. In conclusion, cathinone users are typically young males in their twenties and thirties. Most are experienced drug users, particularly of alcohol and opiates/opioids. The impact on society is tremendous as evidenced by high rates of self-reported violence, high-risk sexual activity, crimes, and traffic violations.

Independent validation of the ICU requirement score in a cohort of acutely poisoned adults.

OBJECTIVE: To independently validate the predictive value of the intensive care requirement score (IRS) in unselected poisoned patients. DESIGN: Retrospective chart review. PATIENTS AND METHODS: Five hundred and seventeen out of 585 admissions for acute intoxications could be analyzed. Eleven were excluded for a condition already requiring intensive care unit (ICU) support at admission (e.g., preclinical intubation). A further 57 admissions were excluded due to missing data. The IRS was calculated using a point-scoring system including age, Glasgow Coma Scale, heart rate, type of intoxication, and preexisting conditions. It was then compared to a composite endpoint indicating an ICU requirement (death in hospital, vasopressors, need for ventilation). The endpoint and the point-scoring system were identical to the original publication of the score. RESULTS AND CONCLUSION: Twenty-three out of 517 patients had a complicated clinical course as defined by meeting the endpoint definition. Twenty-one out of 23 complicated courses had a positive IRS (defined as greater or equal 6 points), as compared to 255/494 patients with an uncomplicated clinical course (p < .001, Fisher's exact test). One patient (with a positive IRS) died. The negative predictive value of the IRS was 0.99 (95% CI: 0.97-1), the sensitivity was 0.91 and the specificity 0.48. In conclusion, the IRS is significantly linked to outcome. While a negative IRS virtually excludes the need for ICU care, a positive IRS has a positive predictive value too low to be used for risk stratification. The IRS could also be applied to unselected admissions of poisoned patients.

Improved Stratification of Autonomic Regulation for risk prediction in post-infarction patients with preserved left ventricular function (ISAR-Risk).

AIMS: To investigate the combination of heart rate turbulence (HRT) and deceleration capacity (DC) as risk predictors in post-infarction patients with left ventricular ejection fraction (LVEF) > 30%. METHODS AND RESULTS: We enrolled 2343 consecutive survivors of acute myocardial infarction (MI) (<76 years) in sinus rhythm. HRT and DC were obtained from 24 h Holter recordings. Patients with both abnormal HRT (slope < or = 2.5 ms/RR and onset > or = 0%) and abnormal DC (< or =4.5 ms) were considered suffering from severe autonomic failure (SAF) and prospectively classified as high risk. Primary and secondary endpoints were all-cause, cardiac, and sudden cardiac mortality within the first 5 years of follow-up. During follow-up, 181 patients died; 39 deaths occurred in 120 patients with LVEF < or = 30%, and 142 in 2223 patients with LVEF>30% (cumulative 5-year mortality rates of 37.9% and 7.8%, respectively). Among patients with LVEF > 30%, SAF identified another high-risk group of 117 patients with 37 deaths (cumulative 5-year mortality rates of 38.6% and 6.1%, respectively). Merging both high-risk groups (i.e. LVEF < or = 30% and/or SAF) doubled the sensitivity of mortality prediction compared with LVEF < or = 30% alone (21.1% vs. 42.1%, P < 0.001) while preserving 5-year mortality rate (38.2%). CONCLUSION: In post-MI patients with LVEF>30%, SAF identifies a high-risk group equivalent in size and mortality risk to patients with LVEF < or = 30%.