1. Introduction Subtle changes in upper body motion during gait may be a marker of incipient pathology, intervention response and disease progression in Parkinson's disease (PD) . It is unknown whether variables obtained from the upper body during gait are merely a reflection of lower body mechanics (as measured by spatiotemporal characteristics) or can provide novel additional information about PD. If unique information can be gained, it may improve the objective characterisation of PD gait in the early stages of the disease and so inform intervention strategies and accurately quantify their effect. 2. Research Question Can measuring upper body motion during gait provide unique information about PD gait from spatiotemporal characteristics; and can this information help classify people with early stage PD and age-matched controls independently of traditional spatiotemporal gait measurements? 3. Methods Seventy people with PD (69.2 ± 9.9 yr, Female: 23, UPDRS III: 36.9 ±12.3) and 64 age-matched controls (71.6± 6.8 yr, Female: 29) walked for two minutes around a 25m circuit. Sixteen spatiotemporal variables were measured using a 7m meter Gaitrite mat located along the circuit, and were selected a priori according a five-domain (pace, rhythm, variability, asymmetry and postural control) validated model of gait . We measured upper body motion using three inertial sensors (128 Hz, APDM) located at the head, neck and pelvis. A broad range of characteristics were chosen based on previous gait literature and included the magnitude, smoothness, harmonicity, attenuation, regularity and symmetry of upper body motion. Upper body characteristics were calculated for anterio-posterior, mediolateral and vertical directions. This process resulted in 78 upper body variables [1,3]. Pearson's correlations were calculated to test how strongly upper body and spatio-temporal characteristics are correlated. Univariate receiver operator characteristic (ROC) curves were used to quantify how well each upper body and spatiotemporal characteristic could discriminate people with PD from controls. Binary logistic regression analysis was performed to determine whether the upper body variables provide additional discriminative information when combined with spatiotemporal characteristics. 4. Results The spatiotemporal characteristics relating to pace were strongly correlated with regularity (r = 0.72), and mildly correlated with symmetry (r = 0.55). Rhythm was mildly correlated with magnitude and smoothness (r = 0.33, for both). Apart from these exceptions, upper and lower body gait domains did not significantly correlate. The univariate analysis showed that 44 of the 78 upper body variables significantly discriminated PD from control participants (p <.05). When the 16 spatiotemporal characteristics were entered (forward stepwise) into a binary logistic regression, the model classified group membership with 74% accuracy. Upper body variables resulted in a model with 83% accuracy. When spatiotemporal characteristics entered the model first and upper body variables were added as a second step, the latter variables significantly contributed (p <.001) to an increase of 16% in the accuracy of the prediction model (from 74% to 90%). 5. Discussion Most upper body variables provided additional and unique information about PD gait with respect to traditional spatiotemporal variables obtained from the lower body. Multivariate analyses showed that this additional information was beneficial in discriminating early PD from controls. We recommend measuring upper body motion in conjunction to traditional spatiotemporal characteristics will help characterise PD gait in a more holistic manner.