Accuracy of decrease in blood flow in predicting hemodialysis graft thrombosis

We recently showed that a single low graft blood-flow measurement (Qa) does not accurately predict graft thrombosis. In this study, we prospectively determined whether percentage of decrease in Qa (ΔQa) or adjustment of Qa for mean arterial pressure (Qa/MAP; Δ(Qa/MAP)) provides greater predictive accuracy than a single Qa. We monitored 83 grafts from 80 patients for thrombosis over periods up to 12 months. Qa (by ultrasound dilution) and MAP were measured monthly during the study. Receiver operating characteristic curves were used to determine whether Qa, ΔQa, Qa/MAP, or Δ(Qa/MAP) provided the combination of high sensitivity (>80%) and low false-positive rate (FPR; <20%) needed for clinical use. This level of predictive accuracy requires an area under the curve (AUC) of approximately 0.90. We analyzed the four predictors by a number of criteria and found that all AUCs were less than 0.90 and adjustment for MAP reduced the AMC. In predicting thrombosis within 1 month, for example, AUCs for Qa and net ΔQa (over 3 months) were 0.84 and 0.82, respectively, whereas AUCs for Qa/MAP and net Δ(Qa/MAP) were 0.78 and 0.75, respectively. At a sensitivity of 80%, FPRs for all predictors were at least 30%. Thus, a high sensitivity always required a high FPR. These results show that ΔQa and adjustment for MAP are not more accurate than a single low Qa in predicting thrombosis. None of these predictors provide enough predictive accuracy to be the sole criterion for clinical decision making. A successful monitoring and intervention program will likely require the inclusion of other predictors that, together with Qa, may provide the needed accuracy. (C) 2000 by the National Kidney Foundation, Inc.