Informatics

The rapid adoption of generative artificial intelligence (AI) systems has transformed health information seeking, raising questions about their role as intermediaries in non-professional health self-consultation. This study compares Google Search and ChatGPT as paradigmatic models of algorithmic mediation of health information, focusing on accuracy, biases, information quality and potential harms. A scoping review was conducted following the PRISMA-ScR framework. Empirical studies published between 2023 and 2025 were retrieved from PubMed/MEDLINE, Web of Science (WoS) and Scopus. After screening and eligibility assessment, 63 original empirical studies were included. The results indicate that ChatGPT consistently outperforms Google Search in terms of factual accuracy and information quality, achieving moderate to high DISCERN scores (4–5 out of 5) and showing moderate to strong correlations with expert clinical evaluations. Users also tend to value ChatGPT responses positively due to their clarity, coherence and perceived empathy. However, these advantages coexist with significant structural limitations. Hallucinations are reported in an estimated 31–45% of references, source provenance remains opaque, linguistic complexity is high, and actionability is limited, with only around 40% of responses providing clearly actionable guidance. In contrast, Google Search offers greater source traceability and verifiability, but at the cost of fragmented information and higher exposure to commercial content. The review identifies critical research gaps related to behavioural impacts, critical health literacy, equity of access, professional integration and vulnerable contexts. Overall, the findings highlight the need for hybrid human–AI models, professional mediation and critical AI literacy to ensure safe, equitable and trustworthy use of generative AI in public health communication.

Background: The COVID-19 pandemic revealed persistent gaps in local health department (LHD) health informatics capacity. This study examines organizational characteristics of LHDs associated with the adoption of six health information systems: electronic case reporting (eCR), electronic disease reporting systems (EDRS), electronic health records (EHR), electronic lab reporting (ELR), health information exchange (HIE), and immunization registries (IR). Methods: We used a mixed-methods design, including multinomial or binary logistic regression analyses of quantitative data from the 2022 NACCHO National Profile of Local Health Departments (n = 441) and thematic analysis of semi-structured interviews with five LHD staff members. Results: About half (49.9%) of LHDs had implemented eCR, while higher proportions had implemented EDRS (78.0%), EHR (62.4%), ELR (57.2%), HIE (92.6%), and IR (92.6%). Workforce size was associated with the implementation of eCR, EHR, and IR. The number of vacant staff positions was associated with a lower odds of IR implementation; compared with medium-sized LHDs, both small and large LHDs had higher odds of IR implementation. Shared-governance LHDs had higher odds of adopting ELR and HIE than state-governed LHDs. Qualitative themes highlighted challenges, including staff burnout, high turnover, pay inequities, role ambiguity, political pressures, rapid changes in informatics, and interoperability problems. Conclusions: Findings underscore the need to improve LHD workforce capacity and governance structures to support a resilient public health informatics infrastructure.

Automating ICD-10 coding from discharge summaries remains demanding because coders analyze clinical narratives while justifying decisions. This study compares three automation patterns: PLM-ICD as a standalone deep learning system emitting 15 codes per case, LLM-only generation with full autonomy, and a hybrid approach where PLM-ICD drafts candidates for an agentic LLM audit to accept or reject. All strategies were evaluated on 19,801 MIMIC-IV summaries using four LLMs spanning compact (Qwen2.5-3B-Instruct, Llama-3.2-3B-Instruct, Phi-4-mini-instruct) to large-scale (Sonnet-4.5). Precision guided evaluation because coders still supply any missing diagnoses. PLM-ICD alone reached 55.8% precision while always surfacing 15 suggestions. LLM-only generation lagged severely (1.5–34.6% precision) and produced inconsistent output sizes. The agentic audit delivered the best trade-off: compact LLMs reviewed the 15 candidates, discarded weak evidence, and returned 2–8 high-confidence codes. Llama-3.2-3B-Instruct, for example, improved from 1.5% as a generator to 55.1% as a verifier while trimming false positives by 73%. These results show that positioning LLMs as quality controllers, rather than primary generators, yields reliable support for clinical coding teams, while formal recall/F1 reporting remains future work for fully autonomous implementations.