Table 1.
Description of SDoH focus issue publications
| Author | Research area | Study target | Study population | Study data | Contribution | Shared task |
|---|---|---|---|---|---|---|
| Sajdeya et al4 | Lexicon development and NLP extraction | Preoperative cannabis use status | UF SH surgery patients ≥65 years old (2018–2020) | UF EHR notes (all types) and MIMIC-III notes | Cannabis lexicon, annotated corpus, and LLM extraction | No |
| Wang et al5 | Ontology adaptation and NLP extraction | Social, behavioral/lifestyle, and economic factors related to suicide | National suicide victims (2003–2019) | Death investigation narratives from NVDRS | Suicide-specific SDoH-ontology, annotated corpus, and LLM classifier | No |
| Lituiev et al6 | Ontology development and NLP extraction | Social support, relationship status, finances, food security, transportation, housing, and insurance | UCSF ISS patients with chronic low back pain (2017–2020) | UCSF EHR notes, patient instructions, and telephone encounters | SDoH ontology, annotated corpus, and LLM classifier | No |
| Yao et al7 | NLP extraction | Eviction status | VHA patients with homeless program, social work, or mental health notes (2016–2021) | VHA EHR homeless program, social work, and mental health notes; MIMIC-III | Annotated corpus and prompt-based LLM extraction approach | |
| Lybarger et al8 | NLP extraction | Substance use, employment, and living situation | Patients in MIMIC-III (2001–2012) and at UW (2008–2019) | n2c2/UW SDoH Challenge data (notes from MIMIC-III and UW) | Overview of n2c2/UW SDoH Challenge task and results | Yes |
| Romanowski et al9 | NLP extraction | Substance use, employment, and living situation | Patients in MIMIC-III (2001–2012) and at UW (2008–2019) | n2c2/UW SDoH Challenge data (notes from MIMIC-III and UW) | LLM seq2seq SDoH event extractor | Yes |
| Zhao et al10 | NLP extraction | Substance use, employment, and living situation | Patients in MIMIC-III (2001–2012) and at UW (2008–2019) | n2c2/UW SDoH Challenge data (notes from MIMIC-III and UW) | LLM multistage SDoH event extractor | Yes |
| Richie et al11 | NLP extraction | Substance use, employment, and living situation | Patients in MIMIC-III (2001–2012) and at UW (2008–2019) | n2c2/UW SDoH Challenge data (notes from MIMIC-III and UW) | LLM multistage SDoH event extractor | Yes |
| Lybarger et al12 | NLP extraction and EHR case study | Substance use, employment, and living situation | UW population, including all medical specialties (2021) | n2c2/UW SDoH Challenge data; UW EHR notes and structured data | LLM extractor, and large-scale EHR case study of narrative SDoH information | No |
| Hartzler et al14 | Ethical use of NLP extraction | Ethical considerations for SDoH extraction system design | Marginalized and underrepresented populations emphasized | Perspective article does not use patient data | Ethical guidance for SDoH extraction system design using AI4People framework | No |
EHR: electronic health record; LLM: large language models; seq2seq: sequence-to-sequence; UW: University of Washington; VHA: Veterans Health Administration; NVDRS: National Violent Death Reporting System; UF SH: University of Florida Shands Hospital; UCSF ISS: University of California at San Francisco Integrated Spine Service.