Abstract: Objective  To evaluate the recognition of key and standard behaviors for infection prevention, the consistency with the effects of manual monitoring, and the clinical effects of the hemodialysis infection prevention and control monitoring system using computer vision.  Methods We randomly recruited 198 nurse-manipulations in the nurses responsible for the connecting to and disconnecting from the dialyzers in the hemodialysis center from September to December 2022 as the research objects. Systematic and real-time recordings for standard behaviors were set as the computer group, and manual review of the systematic video recordings for standard behaviors were defined as the manual group. The key behaviors systematically recorded during connecting to and disconnecting from the dialyzers before and after use of the system in a nurse were divided into behaviors before the intervention and behaviors after the intervention. The first-grade operation indexes and the second-grade indexes  were used to examine the consistency of standard recognition between computer group and manual group. The standard rates of the key behaviors recorded during connecting to and disconnecting from the dialyzers in the same nurse were compared before and after use of the system.  Results  A total of 198 nurse-manipulations during connecting to and disconnecting from the dialyzers were observed. Comparing the first-grade indexes between the two groups found that there was a strong consistency in the recognition of standard internal fistula/catheter connecting to and disconnecting from the dialyzers (kappa value of internal fistula connecting to the dialyzers =0.718; kappa value of internal fistula disconnecting from the dialyzers =0.714; kappa value of catheter connecting to the dialyzers =0.788; kappa value of catheter disconnecting from the dialyzers=0.712). Comparing the second-grade indexes between the two groups found that a less consistency in the recognition of catheter disconnection (kappa value=0.173), an average consistency in wearing gloves (kappa value =0.243) and catheter connection (kappa value =0.305), and a moderate consistency in internal fistula/catheter disinfection (kappa value =0.556) and hand hygiene execution (kappa value =0.590). After the intervention, standard disinfection manipulations during connecting to dialyzers (χ2=8.156, P=0.004), waiting time (χ2=30.462, P<0.001), hand hygiene (χ2=21.023, P<0.001) and standard rate of access evaluation (χ2=23.522, P < 0.001) were better in the computer group than in the control group; the standard rates of wearing gloves (χ2=1.823, P=0.177) and tubes connecting to the dialyzers (χ2=0.410,           P=0.520) had no statistical differences. After the intervention, the standard rates of disinfection for disconnecting from the dialyzers (χ2=4.444, P=0.035) and hand hygiene execution (χ2=17.770, P<0.001) were better in the computer group than in the control group; but the standard rates of wearing gloves (χ2=1.309, P=0.253), catheter disconnections (χ2=3.220, P=0.073) and waiting time (χ2=2.254, P=0.133) had no statistical differences between the two groups.  Conclusion  Hemodialysis infection prevention and control monitoring system using computer vision can record and recognize the standard manipulations of nurses during connecting to and disconnecting from the dialyzers, which is relatively consistent with the results of manual observation. This system increases the standard rates of key behaviors for infection control during connecting to and disconnecting from the dialyzers, and prevents the hemodialysis-related infection events from occurrence.

Key words: Computer vision, Prevention and control of hemodialysis infection, Clinical application, Effect evaluation