Dr. Lopes et al. developed a machine-learning algorithm to predict cognitive scores at 36 months after stroke in 72 participants with first-ever minor stroke by examining functional networks on MRI at 6 months They used a second independent data set of 40 patients to validate their results. They found that the machine-learning model was able to predict memory, attention, visuospatial perception, and language function at 36 months after stroke and was at least as accurate as models based on other functional networks or clinical data. In response, Dr. Lozada-Martínez et al. note that neuroplasticity after stroke is influenced by numerous underlying clinical, genetic, and psychosocial factors and argue that it is not possible to predict the speed and magnitude of recovery with neurorehabilitation, given the expected heterogeneity in these factors over long periods of follow-up. Responding to these comments, the authors agree that further studies seeking to predict long-term poststroke cognitive impairment could combine such additional variables to improve the precision of machine-learning approaches, but note that in their study, factors such as education and white matter hyperintensity burden had no effect on long-term cognitive status, and that their inclusion criteria sought to limit variability in relevant confounding factors. They also note that their network connectivity-based model was validated in a population with similar stroke characteristics but acknowledge that the generalizability of their findings is limited beyond patients with minor strokes and relatively small infarcts and white matter hyperintensity burdens. Indeed, sophisticated imaging studies are helping transform our understanding of poststroke cognitive outcomes. However, this exchange underscores the challenges involved in predicting long-term cognitive outcomes after ischemic stroke, particularly when seeking to extrapolate insights from relatively homogenous and high-functioning populations suitable for detailed imaging and cognitive studies, to the highly heterogeneous populations seen in practice. Dr. Lopes et al. developed a machine-learning algorithm to predict cognitive scores at 36 months after stroke in 72 participants with first-ever minor stroke by examining functional networks on MRI at 6 months They used a second independent data set of 40 patients to validate their results. They found that the machine-learning model was able to predict memory, attention, visuospatial perception, and language function at 36 months after stroke and was at least as accurate as models based on other functional networks or clinical data. In response, Dr. Lozada-Martínez et al. note that neuroplasticity after stroke is influenced by numerous underlying clinical, genetic, and psychosocial factors and argue that it is not possible to predict the speed and magnitude of recovery with neurorehabilitation, given the expected heterogeneity in these factors over long periods of follow-up. Responding to these comments, the authors agree that further studies seeking to predict long-term poststroke cognitive impairment could combine such additional variables to improve the precision of machine-learning approaches, but note that in their study, factors such as education and white matter hyperintensity burden had no effect on long-term cognitive status, and that their inclusion criteria sought to limit variability in relevant confounding factors. They also note that their network connectivity-based model was validated in a population with similar stroke characteristics but acknowledge that the generalizability of their findings is limited beyond patients with minor strokes and relatively small infarcts and white matter hyperintensity burdens. Indeed, sophisticated imaging studies are helping transform our understanding of poststroke cognitive outcomes. However, this exchange underscores the challenges involved in predicting long-term cognitive outcomes after ischemic stroke, particularly when seeking to extrapolate insights from relatively homogenous and high-functioning populations suitable for detailed imaging and cognitive studies, to the highly heterogeneous populations seen in practice.