Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-0897
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-08-26T06:06:25.793384Z
status: published
lens: g_space
sector: health
headline: Adaptation of a fair individualized polysocial risk score for hospitalization risk prediction
dek: Objectives We adapted the individualized polysocial risk score (iPsRS), a machine learning model originally developed for patients with type 2 diabetes, to evaluate its generalizability in predicting 1-year hospitalization risk in a disease-agnostic adult cohort, with attention to fairness and explainability. Materials and methods The study utilized de-identified electronic health record data from a retrospective cohort of 17 857 adult patients at the University of Florida Health. The original iPsRS framework wa…
gain_title: Adapted iPsRS model using 13 individual-level SDoH predictors predicted 1-year all-cause hospitalization in a general adult cohort with AUROC up to 0.671, enabling equity-aware risk stratification for clinical care.
problem_title: (none)
trace_subject: (none)
gain_reading: Adapted iPsRS model using 13 individual-level SDoH predictors predicted 1-year all-cause hospitalization in a general adult cohort with AUROC up to 0.671, enabling equity-aware risk stratification for clinical care.
gain_evidence: achieving an AUROC of up to 0.671 with XGBoost and Normal sampling | The iPsRS was successfully adapted beyond diabetes, maintaining moderate predictive performance and meaningful risk stratification
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers adapted the individualized polysocial risk score, originally built for type 2 diabetes, to a disease-agnostic cohort of 17,857 adults at University of Florida Health. Using 13 individual-level social determinants of health, they trained XGBoost and logistic regression models to predict all-cause hospitalization within one year, testing multiple fine-tuning levels and sampling strategies.

The adapted model reached moderate discrimination with AUROC up to 0.671 and highlighted social factors like food insecurity, marital status, and financial constraints alongside age as key predictors. The work suggests SDoH-informed models can extend beyond diabetes for equity-aware risk stratification, though performance remains moderate and validation is limited to a single retrospective EHR dataset as of the August 2026 publication.
limitation: Model achieved only moderate predictive performance with AUROC up to 0.671, limiting clinical discrimination, and was evaluated in a single health system retrospective cohort.
tag: Evidence-backed gain
key_points: Study used de-identified EHR data from 17,857 adult patients at University of Florida Health to reimplement iPsRS originally for type 2 diabetes. | Models were XGBoost and Logistic Regression trained with 13 individual-level SDoH predictors and fine-tuned at 5 levels with 3 sampling strategies. | SHAP analysis showed age, food insecurity, marital status, and financial constraints as consistently influential predictors of hospitalization.
rundown: Researchers reimplemented the individualized polysocial risk score framework using XGBoost and Logistic Regression, training on 13 individual-level SDoH predictors to predict all-cause hospitalization within 1 year. Fine-tuning was tested at 0%, 10%, 20%, 50%, and 70% with none, oversampling, and undersampling strategies, evaluated primarily by AUROC.

Interpretability was assessed via SHapley Additive exPlanations and causal structure learning, which identified age, race, and employment as proximal factors. Fairness was examined through false negative rate disparities across age, sex, and racial/ethnic subgroups to support equity-aware prediction.
sources:
- peer_reviewed | JAMIA Open | https://doi.org/10.1093/jamiaopen/ooag161 | 2026-08-24
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

0070a3772adb05d8657faf16df3d605b3bcf35a0fa780a14f9b525ad1b7abe4a

previous

0000000000000000000000000000000000000000000000000000000000000000