Prioritizing key performance indicators for maintenance management in public hospitals: a hybrid SEM–machine learning approach

Abstract

Purpose Public hospital maintenance operates under growing service pressure, yet existing frameworks often rely on data-intensive systems or broader facility outcomes, offering limited guidance on which maintenance key performance indicators (KPIs) should be prioritized under resource constraints. This study aims to develop a KPI-based decision-support framework for hospital facilities maintenance management (HFMM) and to identify the critical few KPIs most strongly associated with maintenance performance. Design/methodology/approach An explanatory mixed-method design was adopted in Henan Province, China. Evidence was drawn from a pilot survey of 55 hospitals, a main survey of 283 hospitals and 11 expert interviews. PLS-SEM was used to test the relationships between four KPI domains and maintenance performance, and machine learning with SHAP was used to assess indicator-level importance. Findings All four KPI domains showed positive relationships with maintenance performance. Across the machine learning models, age-based maintenance planning, immediate corrective maintenance and continuous improvement had the largest relative predictive contributions to maintenance performance. Practical implications The framework helps hospital managers allocate limited maintenance resources more selectively by reducing diffuse attention across an overly broad KPI set. It also offers a practical basis for performance review and resource allocation in data-constrained settings. Originality/value This study advances HFMM by moving from broad KPI coverage to strategic KPI prioritization. It offers a data-efficient governance framework that integrates PLS-SEM with machine learning and SHAP to connect construct-level validation with indicator-level prioritization.