± HEALTH Adequate self-harm surveillance is a key part of suicide prevention. Our previous research demonstrated that…± HEALTH Breast cancer remains the most commonly diagnosed malignancy among women globally, with disproportionately hi…+ HEALTH Background The prevalence of depression is higher among older adults with chronic diseases and cognitive impa…± HEALTH Purpose To compare the information quality, accuracy, and readability of patient-directed responses generated…± HEALTH Objective Comparative evaluations of machine learning (ML) and logistic regression (LR) for clinical predicti…± HEALTH Introduction Alzheimer's disease (AD) is a progressive neurodegenerative disorder for which early diagnosis-p…+ HEALTH Background Accurate identification of ganglionated bowel is essential during laparoscopic pull-through for Hi…+ CLIMATE Accurate prediction of ash fusion temperatures (AFTs) is crucial for ensuring the operational efficiency and…
TruaceTracing the truth around AISunday, September 13, 2026
Subscribe
Lead story · Health
Temporal and cross-site validation of an AI system for self-harm detection

AI system for self-harm detection in emergency department triage notes

Abstract: Adequate self-harm surveillance is a key part of suicide prevention. Our previous research demonstrated that an artificial intelligence (AI)-based system could effectively detect self-harm in emergency department triage notes. However, the system was developed using data from a single hospital, raising concerns about its generalisability. Read more →

PLOS Digital Health
Gain

An AI system combining text normalisation with 1931 features maintained stable self-harm detection in prospective validation at its development metropolitan hospital, achieving PR AUC 0.84 over 329,655 triage notes in the following four years.

Problem

When applied to a regional hospital 150 km outside Melbourne, the same AI system's ability to distinguish self-harm cases declined to PR AUC 0.78, with instability linked to linguistic domain shift and different self-harm presentations.

Reading of the Month

When the builders say slow down

The companies building the world's most powerful AI systems are beginning to say something unusual: they may not be ready for what comes next.

Mosisa Saba · September 2026Read the essay →

Top stories

View all →

Health

All health →
Temporal and cross-site validation of an AI system for self-harm detection
Both readings

Temporal and cross-site validation of an AI system for self-harm detection

Adequate self-harm surveillance is a key part of suicide prevention. Our previous research demonstrated that an artificial intelligence (AI)-based system could effectively detect self-harm in emergency department triage notes. However, the system was developed using data from a single hospital, raising concerns about its generalisability. Here, we aim to validate the system prospectively and externally to better understand its portability across hospitals. We leveraged emergency department data from two Australi…

PLOS Digital Health

Explainable machine learning for breast cancer prediction in resource-constrained settings: A multi-algorithmic framework integrating shap-based transparency with clinical decision support

Breast cancer remains the most commonly diagnosed malignancy among women globally, with disproportionately higher mortality rates in low- and middle-income countries (LMICs) where diagnostic delays and limited specialist pathology capacity are widespread. While machine learning (ML) approaches achieve strong predictive performance for cancer classification, algorithmic opacity and absence of interpretability frameworks tailored to resource-constrained environments have impeded clinical adoption. This study bridg…

Explainable machine learning for breast cancer prediction in resource-constrained settings: A multi-algorithmic framework integrating shap-based transparency with clinical decision support

Machine learning identifies depression risk in older adults with chronic diseases: Clarifying shared risk factors stratified by cognitive impairment status

Background The prevalence of depression is higher among older adults with chronic diseases and cognitive impairment than the general population. The comorbidity of cognitive impairment and chronic diseases significantly impacts the lives of these patients. This study aims to develop machine learning models to identify depression risk among older adults with chronic illnesses across different levels of cognitive impairment. Methods Data were derived from the Chinese Longitudinal Healthy Longevity Survey (n = 5798…

Machine learning identifies depression risk in older adults with chronic diseases: Clarifying shared risk factors stratified by cognitive impairment status

Comparative quality, accuracy, and readability of large language model responses to patient questions about robotic-assisted total knee arthroplasty

Purpose To compare the information quality, accuracy, and readability of patient-directed responses generated by large language models (LLMs), including ChatGPT-o3, ChatGPT-5.2, Gemini 3, and DeepSeek, regarding robotic-assisted total knee arthroplasty (RA-TKA). Methods Thirty frequently asked patient questions were identified using LLM outputs and Google search queries. Responses were evaluated for information quality using the DISCERN and Quality Analysis of Medical Artificial Intelligence (QAMAI) instruments,…

Comparative quality, accuracy, and readability of large language model responses to patient questions about robotic-assisted total knee arthroplasty

Systematic Bias in Comparative Evaluations of Machine Learning Versus Logistic Regression for Clinical Prediction Models: A Meta-Research Analysis Using Trauma Mortality as an Empirical Case

Objective Comparative evaluations of machine learning (ML) and logistic regression (LR) for clinical prediction frequently report ML as superior, but the methodological framework producing those comparisons has received limited scrutiny. We aimed to quantify the apparent discrimination advantage of ML over LR using trauma mortality prediction as an empirical case, and to characterise the evaluation practices that shape it. Study design and setting Systematic review and random-effects meta-analysis combined with…

Systematic Bias in Comparative Evaluations of Machine Learning Versus Logistic Regression for Clinical Prediction Models: A Meta-Research Analysis Using Trauma Mortality as an Empirical Case

Predicting Conversion from Mild Cognitive Impairment to Alzheimer's Disease: A Systematic Review of Deep Learning Models for Early-Stage Disease Classification

Introduction Alzheimer's disease (AD) is a progressive neurodegenerative disorder for which early diagnosis-particularly the accurate prediction of conversion from mild cognitive impairment (MCI) to AD-is essential to enable timely and effective therapeutic interventions. Deep learning (DL) models have demonstrated substantial promise in this domain; however, critical challenges persist, including multiclass staging of disease progression, longitudinal data modeling, and effective multimodal data integration. Th…

Predicting Conversion from Mild Cognitive Impairment to Alzheimer's Disease: A Systematic Review of Deep Learning Models for Early-Stage Disease Classification

Education

All education →
Students’ Acceptance of ChatGPT in Higher Education: An Extended Unified Theory of Acceptance and Use of Technology
Evidence-backed gain

Students’ Acceptance of ChatGPT in Higher Education: An Extended Unified Theory of Acceptance and Use of Technology

Abstract AI-powered chat technology is an emerging topic worldwide, particularly in areas such as education, research, writing, publishing, and authorship. This study aims to explore the factors driving students' acceptance of ChatGPT in higher education. The study employs the unified theory of acceptance and use of technology (UTAUT2) theoretical model, with an extension of Personal innovativeness, to verify the Behavioral intention and Use behavior of ChatGPT by students. The study uses data from a sample of 5…

Innovative Higher Education

Efficiency vs. safety in AI-enabled medical education: an ethical analysis of AI as a bridge or a wedge

Abstract: Artificial intelligence is rapidly changing medical education, promising faster workflows and richer learning resources while quietly reshaping how future clinicians think and act. This paper examines the central tension between efficiency and safety in AI-enabled medical education, asking when AI functions as a bridge that strengthens training and when it becomes a wedge that undermi...

Efficiency vs. safety in AI-enabled medical education: an ethical analysis of AI as a bridge or a wedge

Developing validity arguments for artificial intelligence-based assessment: Balancing affordances and threats

Background Artificial intelligence (AI) is increasingly used to generate, score and interpret educational assessment, yet these applications are being adopted in a largely unregulated environment. This creates a paradox: Whereas AI systems used in clinical care are subject to formal scrutiny for safety, performance and monitoring, AI systems used to inform consequential decisions about learner progression and future clinical practice are not. Existing validity frameworks remain useful but may not fully account f…

Developing validity arguments for artificial intelligence-based assessment: Balancing affordances and threats

Not quite eye to A.I.: student and teacher perspectives on the use of generative artificial intelligence in the writing process

Abstract Generative artificial intelligence (GenAI) can be used to author academic texts at a similar level to what humans are capable of, causing concern about its misuse in education. Addressing the role of GenAI in teaching and learning has become an urgent task. This study reports the results of a survey comparing educators’ (n = 68) and university students’ (n = 158) perceptions on the appropriate use of GenAI in the writing process. The survey included representations of user prompts and output from ChatGP…

Not quite eye to A.I.: student and teacher perspectives on the use of generative artificial intelligence in the writing process

Hands-on Artificial Intelligence Education for Radiology Residents: A Three-year Feasibility and Curriculum Implementation Study

Rationale and objectives Artificial intelligence (AI) has rapidly transformed radiology practice, yet structured and practical AI education remains inconsistently integrated into radiology residency training. We developed and implemented a hands-on AI curriculum designed to integrate core computational principles with clinically relevant imaging applications. This study describes the curriculum design and evaluates its feasibility, reproducibility, and preliminary educational outcomes over three consecutive year…

Hands-on Artificial Intelligence Education for Radiology Residents: A Three-year Feasibility and Curriculum Implementation Study

I’m a father of three who studies the impact of artificial intelligence: this is what parents need to know about AI

When I was growing up, my dad and I often talked about a story we wanted to write together. It was called “The Day Nobody Went to Disneyland”. One day, the story goes, the weather is so perfect that everyone in the world decides to stay away from Disneyland because they expect it to be far too busy. But my dad, who has the same thought as them, sees the disappointment on my face that morning – and decides to risk it. We set off. And when we arrive, the park is completely empty. Nobody else was brave enough to vi…

I’m a father of three who studies the impact of artificial intelligence: this is what parents need to know about AI

Media & Arts and the evidence leaderboard

Open the Index →

Top evidence-backed gains

  1. A SnSe/SnO2 p-n heterojunction single-sensor virtual electronic nose using a PCA-KNN framework enables room-temperature, non-contact detection and classification of volatile sulfur compounds released during methamphetamine production and trafficking, with high response to low-concentration H2S and 96.7% accuracy for H2Score 84 · 1 sources
  2. An interpretable multi-instance learning method for accurate differentiation of malignant and benign laryngeal lesions in laryngoscopy: Results IMIL-Net achieved the highest diagnostic performance, with a mean area under the curve (AUC) of 0.975 (95% CI 0.959-0.991), accuracy of 0.915 (95% CI 0.883-0.947), sensitivity of 0.876 (95% CI 0.803-0.949), and specificity of 0.945 (95% CI 0.910-0.980).Score 84 · 1 sources
  3. Systematic review of 35 studies found classical AI models for bipolar disorder achieved pooled AUC 0.80 for long-term maintenance response and 85%-97% accuracy for safety and dose optimization.Score 84 · 1 sources

Top evidence-backed problems

  1. Despite high awareness, most Chinese oncology professionals have not practically integrated medical AI, with only 17.7% reporting both hearing of and using it and 67.7% hearing but never using it.Score 82 · 1 sources
  2. In the same tertiary GDD cohort, the Youden-optimal model missed 33.9% of children who progressed to ID and had NPV 31.7%, limiting safe rule-out and indicating poor transportability beyond high-prevalence referral settings.Score 82 · 1 sources
  3. Even the domain-specific M4CXR model was inconsistent with reference findings in 25.2% of chest radiograph cases and did not significantly change RADPEER discrepancy rates versus original interpretation.Score 82 · 1 sources