Abstract Background and aims The treatment benefit of thrombectomy vary across ischemic stroke patients. MR PREDICTS is an 11-variable clinical/imaging model, using last-seen-well (LSW) to-groin-puncture-time as one of its variables, to provide individualized treatment benefit predictions of thrombectomy in the early window (0-6 hours). However, its performance in the late window (6-24 hours) has not been tested. Moreover, for unknown stroke onset, first-noticed-symptoms (FNS) to-groin-puncture time may perform better than LSW. We aimed to externally validate MR PREDICTS in the late-treatment window and additionally compare its performance using FNS in place of LSW. Methods We used data from the MR CLEAN-LATE trial. Treatment benefit of thrombectomy was defined as the difference in probability of functional independence with and without thrombectomy (modified Rankin Scale (mRS) 0-2 at 90 days). Model performance was evaluated by discrimination and calibration. We additionally examined performance by replacing time-from-LSW-to-groin-puncture with FNS. Results MR PREDICTS yielded moderate discriminative power (C-statistic) 0.78(95%CI:0.73-0.83), but poor calibration in the late window (17% functional independence predicted versus 37% observed), indicating systematic underestimation of the treatment benefit. Using FNS-to-groin time (median: 265 minutes; IQR:185-437) instead of LSW (median:735 minutes; IQR:576-936), the C-statistic remained similar, while calibration improved substantially (33% functional independence predicted versus 37% observed). Using FNS-to-groin-puncture-time, median treatment benefit of thrombectomy is 8.0% (IQR: 3.1-12.9). Conclusions MR PREDICTS shows limited performance in late-presenting patients. Replacing LSW with FNS, substantially improved model performance, indicating that FNS time might be more accurate in guiding patient selection for thrombectomy in the late window. Conflict of interest Le Nguyen: nothing to disclose
Nguyen et al. (Fri,) studied this question.