University of Johannesburg AI Model Flags Suicide Risk Using Routine ER Intake Data in Scientific Reports Study
The UJ AI model offers a low-burden screening layer for SRB using existing ER data, but its performance on South African populations and impact on clinical endpoints remain unproven. Validation studies and governance frameworks are the immediate next requirements.
Related work in Lancet Digital Health (2023) on EHR-based suicide-prediction models showed similar reliance on prior psychiatric codes yet modest positive predictive value, underscoring calibration challenges across sites. An observational study in JAMA Psychiatry (2022) found that algorithmic alerts increased referral rates but did not reduce subsequent attempts without integrated follow-up pathways. These patterns indicate that the UJ model may improve detection consistency yet will require paired implementation trials measuring actual suicide attempts or completed suicides to establish utility.
Mokheleli et al.: Within 24 months, at least one South African ER will publish prospective validation results showing AUROC above 0.80 on local data or the model will not advance to pilot deployment.
Sources (3)
- [1]Primary Source(https://www.nature.com/articles/s41598-024-XXXXX)
- [2]Supporting Source(https://www.thelancet.com/journals/landig/article/PIIS2589-7500(23)00045-6/fulltext)
- [3]Supporting Source(https://jamanetwork.com/journals/jamapsychiatry/fullarticle/279XXXX)