Background <p>Cognitive impairment (CI) frequently occurs in relapsing-remitting multiple sclerosis (RRMS) and is assumed to be irreversible. Recent longitudinal studies highlighted the heterogeneity of CI in RRMS, challenging the traditional view of inevitable progression. Longitudinal studies exploring the baseline determinants of future cognitive decline are limited, and none yet explored the predictive value of patient-reported outcome measures (PROMs).</p> Objective <p>To explore the evolutionary patterns of cognitive status in a cohort of RRMS patients initiating a new disease modifying treatment, and to determine whether PROMs may have a predictive value for future cognitive decline.</p> Methods <p>This prospective study is a 36-month follow-up of 59 RRMS patients who underwent yearly a comprehensive, multiparametric assessment combining clinical, neuropsychological, MRI-derived metrics and a set of self-reported questionnaires. Lesion load and brain volumes were analyzed and processed by the automated MSmetrix^® software (Icometrix^®, Leuven, Belgium). A longitudinal logistic regression analysis was performed to investigate baseline predictors of future cognitive decline.</p> Results <p>A total of 33 (56%) and 17 (35%) patients were defined as cognitively impaired at baseline and at the end of the study, respectively. Of these 33 patients, 20 showed either improvement and/or impairment in fewer cognitive domains at 36-month follow-up. Baseline physical disability as measured by EDSS was the best predictor for future cognitive decline (OR: 2.17; <i>p</i> = 0.03, 95% confidence interval = 1.07–4.38). None of the PROMs variables contributed to predict further cognitive decline.</p> Conclusions <p>Our findings highlight the importance of considering the evolution of CI in MS as a dynamic phenomenon with a substantial heterogeneity.</p>