Abstract 052: External Validation of the Prediction Model For Delayed Reperfusion in Patients with Incomplete Reperfusion: EXTEND‐PROCEED

Introduction The benefit of additional reperfusion attempts in patients with partial angiographic reperfusion (TICI2b) is unknown. The PROCEED model predicts subsequent favorable occurrence of complete reperfusion (i.e. delayed reperfusion [DR]) at 24 hours after initial incomplete angiographic reperfusion at the conclusion of thrombectomy. This study aims to externally validate the PROCEED model using pooled data from multiple international trials that systematically performed follow‐up perfusion imaging. Methods Individual patient data for external validation were obtained from the EXTEND‐IA, EXTEND‐IA TNK part 1 and 2 trials (clinicaltrials.gov, unique‐identifier: NCT01492725, NCT02388061 and NCT03340493). The model’s primary outcome of interest was the occurrence of DR, defined as the absence of any focal perfusion deficit on follow‐up CT or MRI perfusion imaging maps, despite initial incomplete angiographic reperfusion on the final thrombectomy angiography series. The updated model’s performance was evaluated with discrimination, calibration and clinical decision curves. Results We analyzed 267 patients for the external validation, with median age of 74 (IQR 64 – 80), 44.2% were female and 62% had DR. The externally validated model had good discrimination (C‐statistic 0.81, 95% CI 0.72 – 0.86) and was well calibrated (intercept 0.22, 95% CI 0.18‐0.33 and slope 0.96, 95% CI 0.81‐1.23). With threshold probability of R=12% (i.e. 88% chance of having DR), pursuing additional reperfusion attempts to pursue complete angiographic reperfusion in a patient with high‐likelihood of DR were seven times worse (Cost:Benefit Ratio 1:7, Figure 1) than no further endovascular maneuver. In terms of standardized net reduction, the PROCEED model could reduce one in five unnecessary interventions without missing an intervention for any patient who would eventually have DR. Across a wide range of threshold probabilities, the model outperformed the scenario of the typical decision‐making process in the angiography suite, based on the current treatment guidelines. Conclusion The externally validated model had good predictive accuracy and discrimination. Depending on the acceptable risk threshold, the model may compliment clinical judgment of the treating physicians and inform on natural progression of untreated incomplete reperfusion.