This prospective cohort study aimed to determine whether pre-injury characteristics and performance on baseline concussion assessments predicted future concussions among collegiate student-athletes. Participant cases (concussed = 2,529; control = 30,905) completed pre-injury: demographic forms (sport, concussion history, sex), Immediate Post-Concussion Assessment and Cognitive Test (ImPACT), Balance Error Scoring System (BESS), Sport Concussion Assessment Tool (SCAT) symptom checklist, Standardized Assessment of Concussion (SAC), Brief Symptom Inventory-18 item (BSI-18), Wechsler Test of Adult Reading (WTAR), and Brief Sensation Seeking Scale (BSSS). We used machine-learning logistic regressions with area under the curve (AUC), sensitivity, and positive predictive values (+PV) statistics for univariable and multivariable analyses. Primary sport was determined to be the strongest univariable predictor (AUC = 64.3% +/- 1.4, sensitivity = 1.1% +/- 1.4, +PV = 4.9% +/- 6.5). The all-predictor multivariable model was the strongest (AUC = 68.3% +/- 1.6, sensitivity = 20.7% +/- 2.7, +PV = 16.5% +/- 2.0). Despite a robust sample size and novel analytical approaches, accurate concussion prediction was not achieved regardless of modeling complexity. The strongest +PV(16.5%) indicated only 17 out of every 100 individuals flagged would experience a concussion. These findings suggest pre-injury characteristics or baseline assessments have negligible utility for predicting subsequent concussion. Researchers, healthcare providers, and sporting organizations therefore should not use pre-injury characteristics or baseline assessments for future concussion risk identification at this time.