Truace
Subscribe

Simple model matched 14 AI methods at predicting deaths from poisoning in 556 young adults

Study title: Comparison of Machine Learning and Logistic Regression in Predicting Mortality from Acute Poisoning in Young Adults: A Multicenter Study Identifying Herbicide Exposure as the Predominant Risk Determinant

A multicenter study of 556 young adults with acute poisoning compared 14 machine learning algorithms to traditional logistic regression for predicting 28-day mortality. Using six LASSO-selected factors including herbicide poisoning, white blood cell count and shock, logistic regression matched the best machine learning model on discrimination and calibration, leading authors to build a nomogram for early risk stratification.

Read the study abstract

Objective Compare the effectiveness of machine learning algorithms and traditional logistic regression in predicting the mortality risk of young patients with acute poisoning, and establish a risk stratification nomogram. Methods This multicenter retrospective study derived a derivation cohort of 406 young adults with acute poisoning from Wenzhou and an external validation cohort of 150 patients from Lishui.LASSO regression was used to screen predictive factors from 43 candidate variables. Compare the predictive…

By the Truvace desk1 min read1 sourceEvidence: Moderate · 60/100 — Observational study · n = 556 · 1 independent sourcePeer-reviewedTRV-2026-0641· cite & verify
Comparison of Machine Learning and Logistic Regression in Predicting Mortality from Acute Poisoning in Young Adults: A Multicenter Study Identifying Herbicide Exposure as the Predominant Risk Determinant
Acute septic ostitis in children and young people by Royal College of Surgeons of England. Public domain

In brief

The finding matters because it suggests interpretable regression can provide comparable prognostic performance to complex models in this emergency toxicology population, potentially aiding triage and counseling. What remains uncertain from the text is prospective performance, generalizability beyond the two Chinese centers, and whether use of the nomogram changes clinical decisions or outcomes.

Main points

  1. Multicenter retrospective study with derivation cohort of 406 young adults from Wenzhou and external validation cohort of 150 patients from Lishui.
  2. LASSO regression screened 43 candidate variables down to six independent predictors: white blood cell count, creatinine, herbicide poisoning, invasive mechanical ventilation, liver dysfunction, and shock.
  3. 14 machine learning algorithms including RandomForest, XGBoost, CatBoost, LightGBM, SVM compared to logistic regression on 7:3 random split with AUC, sensitivity, specificity and calibration.
  4. Herbicide exposure identified as predominant risk determinant among young adults with acute poisoning.

The problem

Logistic regression matched 14 machine learning models at predicting poisoning mortality, so complex AI added no predictive advantage over a simple model in this population.

The rundown

Researchers used LASSO regression to select six predictors from 43 candidates and compared logistic regression to RandomForest, XGBoost, CatBoost, LightGBM, SVM and nine other algorithms on a 7:3 training split, evaluating AUC, sensitivity, specificity and calibration with internal and external verification.

The final logistic model showed internal AUC 0.885 and external AUC 0.971 with Brier scores 0.058-0.082 and Hosmer Lemeshow P > 0.05, supporting construction of a nomogram described as a simple and effective early risk stratification tool for clinical decision-making assistance.

Sources

  1. Peer-reviewedToxicology Mechanisms and Methods2026-08-03

The debate