Key factors and temporal stability of severe laboratory accidents in Chinese universities and research institutes

Laboratory safety is a prerequisite for teaching and experimental activities in universities. However, factors associated with severe laboratory accidents remain insufficiently quantified, and publicly available accident data are often constrained by small sample sizes, class imbalance and sparse observations. This study analysed 225 publicly reported laboratory accidents in Chinese universities and research institutions from 1999 to 2025. Standard binary logistic regression (SBLR) was used as the benchmark model, and Firth’s penalised logistic regression (FPLR) was used as the primary model to reduce estimation bias. Adjusted predicted probabilities (APPs) were estimated, typical high-risk combinations were identified, and key factor substitution analysis was conducted. In addition, robustness was assessed using an alternative dependent variable and extended model specifications, and temporal stability was examined across three historical periods. The results show that biological accidents and explosion accidents had relatively high predicted probabilities of severe laboratory accidents, as did experimental operations and chemical/material experiments. Risks were concentrated in specific combinations of accident type, activity context and experimental field, with explosions during experimental operations in chemical/material experiments showing the highest predicted probability. These findings provide quantitative evidence that may help identify higher-priority risk scenarios and inform differentiated laboratory safety governance.