Patterns of preventable death and government response compliance across Australian coronial jurisdictions: a natural language processing analysis of 9833 findings

ObjectivesTo quantify patterns of preventable death in Australian coronial findings, measure government compliance with coroner recommendations across jurisdictions, and identify case characteristics associated with recommendation issuance and acceptance. DesignCross-sectional computational text analysis of publicly available coronial findings using unsupervised topic modelling and rule-based classification. SettingAustralian coronial system, all eight state and territory jurisdictions, findings published on the Australasian Legal Information Institute (AustLII) database, 2000-2024. Participants9833 coronial findings and 2040 linked government responses. Main outcome measuresDeath-type topic prevalence, recommendation rate by jurisdiction, government response acceptance rate. ResultsTwenty-six death-type topics were identified, with medical/surgical deaths (12.5%), men-tal health (10.4%), and deaths in custody (8.6%) most prevalent. Overall, 45.6% of published find-ings contained formal recommendations (95% CI, 44.6-46.5%). Of 2040 government responses, 43.0% were unclassifiable (predominantly Victorian administrative cover letters). Among classifi-able responses, 53.2% were accepted (implemented, already implemented, or partially accepted), ranging from 26.0% (Western Australia) to 88.0% (Queensland). Multivariable logistic regression showed that jurisdiction was the strongest measured predictor of acceptance (pseudo R2 0.13 vs 0.14 with all covariates), though most variance remained unexplained. Among published findings, In-digenous Australians were represented in 10.1% (2.7 times the 3.8% population share).12 Findings involving medication errors had the highest recommendation rate (55.1%) but among the lowest acceptance rates (26.4%). ConclusionsAmong publicly available coronial findings, fewer than half contain formal recom-mendations. Government acceptance is low and structurally determined by jurisdiction rather than case characteristics, suggesting that legislative reform is needed to improve the systems preventive effectiveness.