Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-0891
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-08-26T06:05:04.300321Z
status: published
lens: g_space
sector: health
headline: Assessing the Utility of Social Determinants of Health Data in Suicide Prediction Models
dek: Objective This study seeks to explore the utility of social determinants of health (SDoH) variables in suicide prediction models. We aim to assess the impact of individual- and geographic-level SDoH factors on improving the performance of suicide prediction models and the identification of individuals at high risk for suicide. Methods A retrospective sample of 1214 deaths by suicide and 815,544 living patients was identified in the Maryland Suicide Data Warehouse (MSDW) and linked to census tract data through ge…
gain_title: Adding individual-level social determinants of health and clinical features to machine learning models improved suicide risk prediction over demographic-only baselines in a Maryland sample of 1214 suicide deaths and 815,544 living patients.
problem_title: (none)
trace_subject: (none)
gain_reading: Adding individual-level social determinants of health and clinical features to machine learning models improved suicide risk prediction over demographic-only baselines in a Maryland sample of 1214 suicide deaths and 815,544 living patients.
gain_evidence: Models incorporating clinical information demonstrated substantial improvements in predictive performance compared with demographic-only baselines across all algorithms. | In the absence of clinical data, the addition of individual level SDoH significantly improved performance compared to baseline models | 73% of individual level SDoH factors were identified as significant.
problem_reading: (none)
problem_evidence: (none)
quick_read: A retrospective study in the Maryland Suicide Data Warehouse tested whether social determinants of health improve machine learning suicide prediction. Using 1214 suicide deaths and 815,544 living patients linked to census tract data, three algorithms were trained and validated across demographic, clinical, and individual and geographic SDoH inputs.

The findings matter for clinical risk stratification because individual-level SDoH added predictive value beyond demographics and, in some configurations, beyond clinical and demographic information, while geographic-level SDoH did not. What remains uncertain is generalizability beyond Maryland, prospective performance, and how such models would be implemented without introducing bias or privacy harms.
limitation: 
tag: Evidence-backed gain
key_points: Retrospective study used Maryland Suicide Data Warehouse with 1214 deaths by suicide and 815,544 living patients linked to census tract data through geo-coding. | Three machine learning algorithms were trained and validated with cross-validation across combinations of demographics, clinical features, and individual and geographic SDoH. | Individual-level SDoH showed strong signal with 73% of factors significant, while geographic-level SDoH showed only 7% significant and limited improvement.
rundown: Researchers built and cross-validated three machine learning algorithms on data from the Maryland Suicide Data Warehouse, comprising 1214 deaths by suicide and 815,544 living patients linked to census tract data through geo-coding. They compared performance across data category combinations including demographics, clinical features, and individual and geographic level SDoH.

Results showed clinical information drove substantial gains over demographic-only baselines, and individual-level SDoH significantly improved performance when clinical data were absent, with most individual factors ranking among the most important risk factors. Geographic-level SDoH provided limited added value, with only 7% of variables showing significance marginally.
sources:
- peer_reviewed | Archives of Suicide Research | https://doi.org/10.1080/13811118.2026.2722144 | 2026-08-25
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

1b006aa05e50699ce1af38d581a74fabe48ad2e1f25dd14dcb6f470bf340c9a8

previous

0000000000000000000000000000000000000000000000000000000000000000