Improving environmental cleaning quality and consistency in a multi-campus hospital using a computer vision-assisted mobile application: a prospective interventional study

Abstract

Abstract Background Consistent environmental cleaning is essential for preventing healthcare-associated infections, yet achieving uniform standards across multi-campus hospital systems is difficult. Conventional methods, such as manual visual inspection and traditional fluorescent marking, are inherently subjective, qualitative, and prone to observer bias. This study assessed whether a computer vision-assisted mobile application embedded in a technology-empowered Plan-Do-Check-Act (T-PDCA) quality improvement model improves cleaning performance and promotes homogeneity across campuses. Methods We conducted a 17-week prospective interventional study (June–October 2025) in a tertiary hospital system with four geographically dispersed campuses in Southwest China. A smartphone application generated an algorithm-derived fluorescent marker removal rate (FMRR) score by analyzing ultraviolet fluorescence images captured before and after routine cleaning and provided real-time quantitative feedback. The intervention was implemented within a T-PDCA framework, including standardized training, blinded fluorescent marking, algorithm-based performance assessment, and data-driven feedback. Cleaning performance was graded using predefined FMRR thresholds. Temporal trends, between-campus homogeneity, and differences by unit type and surface material were evaluated using nonparametric statistics. Results A total of 2,457 valid fluorescent marker samples were analyzed during the study period. The overall weekly mean FMRR was 83.91% in Week 1 and remained above 90% from approximately Week 4 onward, with a significant upward trend across the study period (Spearman’s ρ = 0.615, P = 0.01). Performance variability decreased substantially, with the weekly standard deviation declining from 29.72 to 9.59%, indicating improved homogeneity. Campuses with lower baseline performance showed improvement, while high-performing campuses maintained relatively stable cleaning quality. Significant differences in FMRR were observed across hospital units and surface materials; notably, intensive care units, outpatient departments, and surfaces made of matte plastic or wood demonstrated lower and more variable performance. Conclusions Integrating a computer vision-assisted mobile monitoring system into a T-PDCA quality improvement framework was associated with improved environmental cleaning performance and enhanced homogeneity across a multi-campus hospital system. This approach may support standardized, real-time monitoring of targeted high-touch surfaces and provide a practical tool for improving environmental cleaning management in multi-campus hospital systems.