OBJECTIVE:
To systematically assess the quality of studies published between 2022 and 2025 that developed machine learning-based mortality risk prediction models for ischemic stroke, revealing the current status and shortcomings in reporting transparency and methodological rigor.

METHODS:
A systematic search of Chinese and English databases was conducted, identifying 58 relevant studies. The latest international standards, TRIPOD+AI, were used to assess reporting completeness, and PROBAST+AI was used to assess methodological quality (risk of bias and applicability). Descriptive statistical analysis was performed on reporting adherence rates and methodological quality.

RESULTS:
The overall reporting transparency of the included studies was moderate (median adherence rate 56.25%), with severe deficiencies in key areas: reporting rates were extremely low for sample size justification (2%), model fairness (2%), patient and public involvement (2%), code sharing (17%), and model specification (9%). Methodologically, more than half (55.2%) of the studies had a high risk of bias, primarily concentrated in the data analysis domain. Prominent issues included: insufficient sample size (24.1%), inappropriate handling of missing data (25.9%), and lack of recalibration after addressing class imbalance (76.2% of studies using such methods did not recalibrate).

CONCLUSIONS:
Current research on machine learning-based mortality risk prediction models for ischemic stroke is active, but widespread issues of incomplete reporting and methodological flaws seriously hinder model reproducibility, verifiability, and clinical translation potential. Future studies should strictly adhere to the TRIPOD+AI and PROBAST+AI frameworks, and journal reviewers should incorporate these standards to promote the development of high-quality, trustworthy prediction models.