Construction and Preliminary Evaluation of a Precise Perioperative Glycemic Management Model for Kidney Transplant Recipients Using Healthcare Failure Mode and Effect Analysis (HFMEA) Combined with Continuous Glucose Monitoring (CGM) and Artificial Intelligence: A Retrospective Cohort Study.

OBJECTIVES: Post-transplantation diabetes mellitus (PTDM) is a common complication after kidney transplantation and contributes to graft dysfunction and adverse cardiovascular outcomes. We conducted a preliminary evaluation of an integrated perioperative glycemic management model combining Healthcare Failure Mode and Effect Analysis (HFMEA), continuous glucose monitoring (CGM), and artificial intelligence (AI)-driven prediction in routine clinical practice. DESIGN: Single-center retrospective cohort study with historical controls. METHODS: We included 120 adult kidney transplant recipients from 2021-2024 and divided them into a historical control group (n = 60, conventional intermittent capillary glucose monitoring) and an intervention group (n = 60, HFMEA-CGM-AI integrated management). Primary outcomes were perioperative CGM-derived glycemic metrics over 14 days (time in range TIR, coefficient of variation CV, time below range TBR, mean amplitude of glycemic excursions MAGE) and 3-month PTDM incidence. Secondary outcomes included hospital length of stay and postoperative complications; patient satisfaction was assessed descriptively in the intervention group. RESULTS: The intervention group was associated with substantially tighter perioperative glycemic control, with higher TIR (80.0 ± 9.0% vs 63.8 ± 11.6%) and lower CV, TBR, and MAGE (all p < 0.001). TIR in the intervention group exceeded recommended CGM targets (>70%). Three-month PTDM incidence was numerically lower (10.0% vs 20.0%; p = 0.201; 18 events), consistent with limited power for this endpoint and an exploratory role of PTDM in this study. Hospital stay was shorter with the integrated model (12.4 ± 2.2 vs 14.2 ± 3.3 days, p < 0.001), and postoperative complications were numerically reduced but non-significant. Patient satisfaction in the intervention group was high (mean 8.4 ± 1.1 on a 10-point scale). CONCLUSION: In this retrospective historical cohort, the HFMEA-CGM-AI model was feasibly implemented and was associated with improved perioperative CGM metrics and shorter hospital stay. PTDM and complication outcomes remained exploratory due to limited power and the non-randomized design. Prospective, adequately powered randomized trials are required to determine the causal impact and cost-effectiveness of this integrated HFMEA-CGM-AI management approach.