医疗事故司法责任判定

: 司法和非司法因素影响研究

Abstract

This study provides an in-depth analysis of judicial decision-making patterns in medical malpractice cases and makes important contributions to the field of empirical legal research. The findings demonstrate that, among the explainable portion of verdict variation, judicial factors play a significantly more decisive role than non-judicial ones. Variables such as whether the patient died, the severity of the medical harm, and the conclusions of forensic evaluations emerged as strong predictors of liability assignment. In contrast, non-judicial factors—such as whether the plaintiff hired a lawyer, whether the lawyer came from a top-tier (Red Circle) firm, the name of the hospital, or the geographic location of the case—had comparatively weaker influence. This result reinforces the central role of legally structured information in judicial reasoning and affirms the foundational influence of the legal framework and procedural evidence in shaping court decisions.
Building on this foundation, the study further reveals that a substantial portion of judicial decisions in medical disputes remain non-quantifiable. Specifically, the R² value of the regression model was 0.2289, meaning that approximately 77.11% of variation in verdict outcomes could not be accounted for by the legal, medical, or socio-economic variables included in the model. This suggests that, beyond judicial and non-judicial factors, a significant portion of decisions are influenced by factors that are currently unmeasurable or difficult to model—what this study refers to as non-quantifiable factors. These may include courtroom dynamics, witness demeanor, subtle language cues in evidence, implicit biases, and institutional pressures, among others.
This finding challenges the traditional assumption that legal decisions are solely determined by objective facts and codified rules. It points instead to the presence of meaningful discretionary space in judges' decision-making processes, especially in high-risk, high-complexity contexts such as medical malpractice cases. It also highlights the limitations of current quantitative models in fully capturing the rich and complex nature of legal judgment.
To address these complexities, this study employs a Gradient Boosted Decision Tree (GBDT) model to analyze the relative importance of 64 variables without dimensionality reduction. The model effectively captures non-linear interactions and variable interdependencies, offering both high predictive accuracy and interpretable outputs. Unlike traditional legal studies that rely heavily on doctrinal analysis or qualitative case studies, this research adopts a data-driven methodology that enables the identification of latent decision patterns and the relative impact of diverse influencing factors. This approach demonstrates the potential of machine learning techniques in advancing the frontier of legal technology (Legal Tech) and empirical legal studies.
The study’s findings also carry several practical implications for policy, hospital administration, and judicial system reform. First, given the high degree of non-quantifiability in verdicts, it is critical to develop a unified and transparent case reference system, such as a national database of medical malpractice cases, to reduce inter-judge variation and promote consistency in judicial outcomes. Second, the prominent role of forensic evaluations in liability decisions underscores the need for standardization and professionalization in the medical appraisal system. Unified evaluation protocols and enhanced oversight of forensic institutions could significantly reduce inconsistencies in expert input.
From a hospital management perspective, the findings suggest that liability decisions hinge not only on the outcome (e.g., patient death) but also on the quality of documentation and legal preparedness. Hospitals should improve medical recordkeeping, ensure traceability, and establish stronger in-house legal teams to effectively respond to litigation risk. Additionally, the strong influence of patient death on court outcomes indicates the necessity for enhanced risk prevention mechanisms, particularly around high-risk procedures.
The research also lays a foundation for the development of Judicial Decision Support Systems (JDSS). Based on the modeling framework proposed in this study, AI-powered systems could be developed to assist judges in identifying key factors, comparing similar past cases, and promoting consistency in medical dispute adjudication. Such tools, if appropriately designed and ethically governed, could enhance both efficiency and transparency in judicial decision-making.
Looking ahead, several avenues for future research emerge. One direction involves cross-regional comparative analysis to examine whether verdict logic and the role of quantifiable versus non-quantifiable factors differ across provinces with varying levels of economic development, healthcare infrastructure, and judicial resources. Another promising direction is comparative legal studies: for example, contrasting medical malpractice litigation in civil law systems (which often emphasize expert opinion) with common law jurisdictions (which may rely more on jury reasoning or judicial discretion) to explore how systemic differences shape predictability and fairness in outcomes.
In terms of methodology, future studies may incorporate Natural Language Processing (NLP) to analyze the language of judicial opinions, extract reasoning structures, and identify latent linguistic patterns that affect liability determinations. Moreover, the integration of advanced causal inference techniques—such as synthetic control methods, instrumental variable approaches, or deep learning—may further enhance the explanatory and predictive power of legal decision models.
In summary, this study systematically reveals the structural composition of judicial decision-making in medical malpractice cases. It shows that judicial factors dominate the explainable portion of decisions, while non-quantifiable factors continue to play a substantial role in shaping outcomes. By applying machine learning techniques to a large-scale judicial dataset, the research expands the methodological toolkit of legal scholars, contributes to the growing field of empirical legal studies, and offers practical implications for legal policy and institutional reform.
As legal systems worldwide confront growing demands for transparency, efficiency, and fairness, the integration of data science and AI into legal analysis will become increasingly important. This study exemplifies how machine learning can be used not just to predict outcomes, but also to understand the underlying logic of legal reasoning. It provides a scalable framework for future research in other legal domains and contributes to the broader vision of a more transparent, evidence-based, and adaptive legal system.

Date of Award	18 Mar 2025
Original language	Chinese (Simplified)
Awarding Institution	China Europe International Business School
Supervisor	Meng Rui (Supervisor) & Kejia Hu (Supervisor)