Risk of Hemodialysis Graft Thrombosis: Analysis of Monthly Flow Surveillance

Background: During clinical application of flow surveillance of hemodialysis grafts, the risk of thrombosis is assessed month after month, rather than after one or several measurements, as has been done in published studies. Adequate assessment of risk should consider the many measurements obtained over time. Study Design: Prospective cohort diagnostic test study. Setting & Participants: 176 patients with hemodialysis grafts from 2 university-affiliated dialysis units during a 6-year period. Index Tests: Monthly measurement of graft blood flow or change in flow. Outcome: Graft thrombosis. Results: We used logistic regression analysis to compute the risk of thrombosis and used receiver operating characteristic (ROC) curves to assess the accuracy in predicting thrombosis within 1 month. Newer grafts were most likely to thrombose, whereas older grafts were unlikely to thrombose even at low flows or large decreases in flow. Areas under the ROC curves were 0.698 for flow and 0.713 for change in flow measured over 2 months. Flow predicted thrombosis with a sensitivity of 53% at a specificity of 79%, and change in flow had a sensitivity of 58% at a specificity of 75%. More than half the thromboses lacked a change in flow measurement, usually because thrombosis occurred before a change could be measured. Thus, the effective predictive accuracy of change in flow was much less than the ROC curves indicated because the curves do not consider missing measurements. Limitations: Performance characteristics of index tests may vary across patient populations. Conclusion: Flow and change in flow are inaccurate predictors of thrombosis. Many thromboses are not predicted, and intervention based on surveillance likely yields many unnecessary procedures. Thus, this study does not support routine application of surveillance to prevent thrombosis.