Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1528
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-16T01:10:32.237162Z
status: published
lens: trace
sector: labor
headline: Development and internal validation of a machine learning model for work-related sick leave in Brazilian workers with mental disorders
dek: Background: Mental disorders are a leading cause of work-related sick leave in Brazil, yet routinely collected surveillance data are seldom used to build prediction tools for occupational health services. We aimed to develop and internally validate machine learning models predicting work-related sick leave among individuals notified with work-related mental disorders.
Methods: We analyzed 21,601 notifications of work-related mental disorder (ICD-10 F99) from Brazil's national surveillance system (SINAN, 2006-202…
gain_title: Models trained on routine national surveillance notifications gave moderate-discrimination prediction of sick leave, with positive net benefit across roughly 30-70% decision thresholds.
problem_title: The champion model discriminated markedly worse for Black workers and workers with secondary education, so deployment would need recalibration and fairness auditing.
trace_subject: Machine learning prediction of work-related sick leave among Brazilian workers with mental disorders
gain_reading: Models trained on routine national surveillance notifications gave moderate-discrimination prediction of sick leave, with positive net benefit across roughly 30-70% decision thresholds.
gain_evidence: Decision curve analysis showed positive net benefit over default strategies for threshold probabilities of approximately 30-70%. | Routinely collected surveillance data can support moderate-discrimination, potentially useful prediction of work-related sick leave in mental disorders
problem_reading: The champion model discriminated markedly worse for Black workers and workers with secondary education, so deployment would need recalibration and fairness auditing.
problem_evidence: markedly lower AUC-ROC for Black individuals (0.620) and workers with secondary education (0.664) than overall (0.697) | deployment as a decision-support tool would require targeted recalibration and fairness auditing
quick_read: Researchers used 21,601 work-related mental disorder notifications from Brazil's national surveillance system to build three machine learning models predicting whether workplace absence would be granted. A random forest performed best on precision-recall, and absence was granted in about 64 percent of cases.

The work suggests existing public health records could feed occupational health decision tools, but performance was only moderate and uneven across demographic groups. Because only internal validation was done, it remains unknown whether the model would hold up in other settings or prospectively.
limitation: Only internal validation was performed, so the model is not ready for deployment and needs external, prospective validation, plus recalibration and fairness auditing for disadvantaged subgroups.
tag: Dual reading
key_points: Data covered 21,601 notifications of work-related mental disorder (ICD-10 F99) in Brazil's SINAN system from 2006 to 2025. | Workplace absence was granted in 13,896 notifications, or 64.3 percent. | The random forest champion had test AUC-ROC 0.697 and AUC-PR 0.802, similar to XGBoost and better than elastic net. | Notification year, employment status, psychotropic medication use, region, and confirmed work-relatedness contributed most to predictions. | Authors say external, prospective validation is required before any clinical or operational use.
rundown: The team compared elastic net, random forest, and XGBoost models, tuned with 5-fold cross-validation on a 75% training split and tested on the remaining 25%. The champion was the model with the highest test-set AUC-PR, and it was examined using SHAP for variable importance.

Fairness checks across sex, race/ethnicity, education, and region showed discrimination varied substantially across subgroups. AUC-ROC was 0.620 for Black individuals and 0.664 for workers with secondary education, compared with 0.697 overall.

The authors conclude that internal validation alone does not support deployment, and refer to TRIPOD + AI guidance for the need for external, prospective validation.
sources:
- peer_reviewed | International Journal of Medical Informatics | https://doi.org/10.1016/j.ijmedinf.2026.106736 | 2026-09-23
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

5966b6b792b5b7dc2cd1ef43aca042be83d662fbf5a862185a1b492ae6357600

previous

0000000000000000000000000000000000000000000000000000000000000000