Scholarlink

Background <p>Healthcare liability litigation has increased substantially, highlighting the need for structured and evidence-based medico-legal evaluations. While generative artificial intelligence (AI) has been extensively studied in clinical medicine, its application in medico-legal consultancy remains limited.</p> Objective <p>To assess whether two customized GPT-based models, simulating opposing medico-legal perspectives (“patient-oriented” and “hospital-oriented”), can analyze real clinical documentation and generate reports comparable to those produced by human experts.</p> Methods <p>Ten real healthcare liability cases were retrospectively analyzed. For each case, complete anonymized clinical records and human-authored medico-legal reports were available. Two GPT-based models were configured using domain-specific instructions and authoritative medico-legal references, without retraining. Both models generated simulated reports from identical documentation. Outputs were evaluated using structured qualitative scores assessing argumentative coherence, clinical synthesis, bibliographic accuracy and relevance, and concordance of permanent impairment estimates.</p> Results <p>No significant difference in overall argumentative coherence was observed between models (Wilcoxon test, <i>p</i> = 0.844). Orientation-consistent trends emerged after stratification by liability status. Agreement was substantial for clinical synthesis (Cohen’s κ = 0.804) but slight for permanent impairment estimates (κ = 0.167). The patient-oriented model produced significantly more fabricated or non-verifiable references (OR = 3.74; <i>p</i> = 0.0065), while citation relevance did not differ significantly between models (<i>p</i> = 0.195).</p> Conclusions <p>Generative AI can simulate divergent medico-legal reasoning and rapidly produce structured reports. However, expert human oversight remains essential, particularly for source verification and medico-legal interpretation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Generative artificial intelligence in forensic medicine: a pilot study on AI-simulated medico-legal reports in healthcare liability cases

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月2日

查看全文
  • Federica Ministeri,
  • Massimiliano Esposito,
  • Martina Francaviglia,
  • Lucio Di Mauro,
  • Grazia Giulia Pantè,
  • Monica Salerno,
  • Cristoforo Pomara,
  • Francesco Sessa

摘要

Background

Healthcare liability litigation has increased substantially, highlighting the need for structured and evidence-based medico-legal evaluations. While generative artificial intelligence (AI) has been extensively studied in clinical medicine, its application in medico-legal consultancy remains limited.

Objective

To assess whether two customized GPT-based models, simulating opposing medico-legal perspectives (“patient-oriented” and “hospital-oriented”), can analyze real clinical documentation and generate reports comparable to those produced by human experts.

Methods

Ten real healthcare liability cases were retrospectively analyzed. For each case, complete anonymized clinical records and human-authored medico-legal reports were available. Two GPT-based models were configured using domain-specific instructions and authoritative medico-legal references, without retraining. Both models generated simulated reports from identical documentation. Outputs were evaluated using structured qualitative scores assessing argumentative coherence, clinical synthesis, bibliographic accuracy and relevance, and concordance of permanent impairment estimates.

Results

No significant difference in overall argumentative coherence was observed between models (Wilcoxon test, p = 0.844). Orientation-consistent trends emerged after stratification by liability status. Agreement was substantial for clinical synthesis (Cohen’s κ = 0.804) but slight for permanent impairment estimates (κ = 0.167). The patient-oriented model produced significantly more fabricated or non-verifiable references (OR = 3.74; p = 0.0065), while citation relevance did not differ significantly between models (p = 0.195).

Conclusions

Generative AI can simulate divergent medico-legal reasoning and rapidly produce structured reports. However, expert human oversight remains essential, particularly for source verification and medico-legal interpretation.