NIH AI IPV Predictor Hits 0.73 AUROC on 9M EHRs

NIH AI IPV Predictor Hits 0.73 AUROC on 9M EHR Cohort

NIH researchers developed an AI IPV predictor that analyzes electronic health records (EHRs) from 9,034,287 patients across 24 PCORnet systems. The model forecasts intimate partner violence (IPV) risk up to two years before diagnosis. It relies on routine clinical notes, ICD-10 codes, and visit patterns.

Lead author Benjamin C. Yan, MD, MPH, from the National Institute on Minority Health and Health Disparities, published findings in JAMA Network Open (2024; doi:10.1001/jamanetworkopen.2024.33281). The XGBoost machine learning model achieved an area under the receiver operating characteristic curve (AUROC) of 0.73 overall. Validation held steady across sexes (AUROC 0.72 for females, 0.74 for males), ages, races, and U.S. regions.

External testing on 2 million additional patients confirmed AUROC at 0.72. Sensitivity hit 0.70; specificity 0.68. This outperforms prior logistic regression baselines by 15%.

Study link.

IPV Accelerates Epigenetic Aging by 2-5 Years

Intimate partner violence induces chronic stress, spiking cortisol and inflammation. A JAMA Psychiatry meta-analysis by Devries et al. (2019; 20 studies, n>10,000) linked IPV to 1.5-2x higher cardiovascular disease risk (RR 1.52, 95% CI 1.41-1.64).

Telomere length shortens 10-15% faster in survivors, according to Humphreys et al. in Psychoneuroendocrinology (2020; prospective cohort, n=1,200 women). Epigenetic clocks, like Horvath's, advance 2-5 years prematurely. Peter Attia, MD, notes in Outlive (2023) that such stress directly caps healthspan at 20-30% all-cause mortality increase.

CDC data (2023) confirms IPV survivors lose 5-15 years of life expectancy. Violence drives 28% higher cancer rates and 52% elevated stroke risk.

CDC IPV facts.

XGBoost Extracts IPV Signals from Unstructured EHR Text

The NIH AI IPV predictor processes unstructured clinical notes using natural language processing. It flags injury codes (ICD-10 S00-T88), anxiety/depression diagnoses (F32-F33), and frequent ER visits. Demographics, medications, and social determinants integrate as features.

XGBoost ensemble trees handled 1,200+ variables with low overfitting. Training split 70/15/15; federated learning across PCORnet preserved privacy. No surveys needed—scales to population level.

Performance: Positive predictive value 0.65; negative predictive value 0.85. Model code releases open-source via NIH. Primary care integration pilots start Q1 2025.

Longevity Protocols Layer IPV AI with VO2 Max, NAD+

Healthspan optimization tracks VO2 max, NAD+ levels, hsCRP, and epigenetic age. IPV evades these biomarkers yet slashes expectancy by 5-15 years (CDC, 2023). AI screens in EHRs enable proactive flags.

Interventions—counseling, safety plans, trauma therapy—reverse 30-50% epigenetic acceleration, per pilot RCTs (n=500; JAMA Pediatrics, 2022). Fountain Life clinics deploy similar AI atop wearables. Oura Ring HRV (<50 ms RMSSD) aligns with IPV stress signals.

Biohackers stack rapamycin or metformin with violence prevention for compounded gains. Longevity models now quantify social risks alongside genetics.

$15.4B AI Funding Powers IPV-Longevity Integration

Digital health AI secured $15.4 billion in 2023 venture funding, per Rock Health Q4 report. Longevity biotech Rejuvenate Bio raised $100 million (Series A, 2023) for inflammation therapies targeting stress pathways.

PCORnet's federated platform, funded by $500 million NIH grants (2020-2024), enables such models. FDA cleared comparable sepsis AI (e.g., Epic Sepsis Model, 2022; AUROC 0.82). Unity Biotechnology (NASDAQ: UBX) partners AI firms for senescence linked to chronic stress.

NIH release.

Rock Health report.

EHR AI refines longevity valuations. Early IPV intervention extends healthspan 5+ years, boosting biotech pipelines and fund returns.