Medical Equipment Maintenance Records for Priority Assessment in a Tertiary Hospital

9,924 medical devices, including 8,967 with preventive maintenance and 957 with corrective maintenance records, were used to develop a machine learning-assisted priority evaluation system. The study, authored by Chenjian Ye and published on figshare in 2026, details the deployment and performance of models like XGBoost in a real-world hospital setting. Results from a controlled implementation in 2025 showed a lower maintenance cost ratio in the intervention campus.

Use Cases

• Training ML models for maintenance priority classification based on device risk profiles.
• Comparing the effectiveness of preventive versus corrective maintenance strategies based on recorded outcomes.
• Analyzing the impact of a data-driven maintenance framework on cost and failure rates, as shown in the controlled deployment.
• Benchmarking performance of algorithms like logistic regression, decision trees, and XGBoost on structured maintenance data.

Strengths

• Includes records for 9,924 distinct medical devices, providing a substantial sample.
• Describes a real-world deployment with a controlled comparison, showing a 4.8% vs. 7.3% maintenance cost ratio.
• Model performance metrics are reported, with XGBoost achieving testing accuracies of 0.9379 and 0.8646 for different maintenance types.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying tabular data is unknown, which may limit suitability assessment.
• The dataset is presented as a 1.3 MB PDF, which may require extraction to access structured data.

Provenance

Source

figshare, authored by Chenjian Ye.

Collection Method

Developed and implemented in a tertiary hospital, using machine learning models trained on maintenance records.

Time Range

Deployment phase covered 2025-01-02 to 2025-12 25.

Freshness

Last updated 2026-04-14 05:38:55; freshness should be verified.

Geography

Likely from a tertiary hospital; specific location is not stated.