Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-0817
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-08-18T06:04:21.309440Z
status: published
lens: trace
sector: health
headline: Artificial Intelligence and Machine Learning-based prediction of tuberculosis treatment failure: A systematic review and meta-analysis
dek: Tuberculosis (TB) remains a leading cause of infectious disease mortality worldwide, and treatment failure contributes to ongoing transmission, drug resistance, and poor clinical outcomes. Artificial intelligence (AI) and machine learning (ML) approaches have attracted growing interest in predicting TB treatment outcomes, but the literature is heterogeneous and lacks a comprehensive synthesis. We systematically searched PubMed/MEDLINE and Embase (January 2000-October 2025) for studies developing or validating AI…
gain_title: Meta-analysis of 19 studies with 100,790 participants found AI/ML models achieved pooled discrimination of 0.836 for predicting tuberculosis treatment failure.
problem_title: Most models lacked external validation, only one was low risk of bias, publication bias was detected, and performance dropped in HIV-positive populations, leaving models not ready for routine clinical implementation.
trace_subject: AI/ML models predicting tuberculosis treatment failure
gain_reading: Meta-analysis of 19 studies with 100,790 participants found AI/ML models achieved pooled discrimination of 0.836 for predicting tuberculosis treatment failure.
gain_evidence: The pooled AUC was 0.836 (95% CI 0.799-0.868) | AI/ML models for predicting TB treatment failure show promising discrimination
problem_reading: Most models lacked external validation, only one was low risk of bias, publication bias was detected, and performance dropped in HIV-positive populations, leaving models not ready for routine clinical implementation.
problem_evidence: Only eight studies (23.5%) performed external validation, and only one study (2.9%) was rated low risk of bias overall | are not yet ready for routine clinical implementation | Egger's test suggested publication bias (p = 0.024)
quick_read: By August 2026, a systematic review and meta-analysis of 34 studies evaluated AI and machine learning models to predict tuberculosis treatment failure. Nineteen studies with 100,790 participants were pooled, yielding an AUC of 0.836 with high heterogeneity, with tree-based and multimodal approaches common and most publications appearing after 2019.

The findings matter because treatment failure drives transmission and drug resistance, yet the review found only 23.5% of studies had external validation and 2.9% were low risk of bias, with lower performance in HIV-positive groups and evidence of publication bias. This leaves uncertainty about calibration, generalizability to high-burden and pediatric populations, and readiness for clinical deployment.
limitation: 
tag: Dual reading
key_points: Systematic review included 34 studies from 2000-2025, with 91% published from 2019 onwards. | Tree-based methods predominated at 52.9% and multimodal models using >=3 data types were used in 41.2% of studies. | Subgroup analysis found lower discrimination when HIV-positive participants were included, AUC 0.748 versus 0.924 when excluded.
rundown: The review searched PubMed/MEDLINE and Embase from January 2000 to October 2025 and registered as PROSPERO CRD420251101443. Two reviewers extracted AUC, sensitivity, specificity and confidence intervals, estimating missing standard errors from sample sizes and event rates, and assessed bias with PROBAST.

Nineteen studies contributed to random-effects meta-analysis. Authors reported funnel plots, Egger's test, trim-and-fill, subgroup analyses and meta-regression to explore heterogeneity, and identified gaps including high-burden countries, social determinants, pediatric TB and extrapulmonary disease.
sources:
- peer_reviewed | PLOS One | https://doi.org/10.1371/journal.pone.0355124 | 2026-08-17
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

114c4d940cee07be2e4735193a73ebdb8b48c7649117213f62ff3e59657c9388

previous

0000000000000000000000000000000000000000000000000000000000000000