Panpan Zhao Lin Shi Gui Mei Zhang Chunxiao Wei Weijie Zhai Yanxin Shen Yongchun Wang Zicheng Wang Li Sun ^*

• Department of Neurology and Neuroscience Center, The First Hospital of Jilin University, Jilin University, Changchcun, China

Many predictive models for cognitive impairment after mild stroke and transient ischemic attack are based on cognitive scales at a certain timepoint. We aimed to develop two easy-to-use predictive models based on longitudinal cognitive trajectories to facilitate early identification and treatment. This was a prospective cohort study of 556 patients, followed up every three months. Patients with at least two cognitive scales within 2.5 years were included in the latent class growth analysis (LCGA) . The patients were categorized into two groups based on the LCGA. First, a difference analysis was performed, and further univariate and stepwise backward multifactorial logistic regression was performed. The results were presented as nomograms, and receiver operating characteristic curve analysis, calibration, decision curve analysis, and cross-validation were performed to assess model performance. The LCGA eventually included 255 patients, and the ‘22’ group was selected for further subgroup analysis. Among them, 29.8% were included in the cognitive impairment trajectory. Model 1, which incorporated baseline Montreal Cognitive Assessment, ferritin, age, and previous stroke, achieved an area under the curve (AUC) of 0.973, and Model 2, which incorporated age, previous stroke, education, and ferritin, with an AUC of 0.771. Decision curve analysis and cross-validation showed excellent clinical applicability. Here, we developed two simple and easy-to-use predictive models of post-stroke cognitive trajectories based on a LCGA, which are presented in the form of nomograms suitable for clinical application. These models provide a basis for early detection and prompt treatment.