Background: The Gait Deviation Index (GDI) and the Gait Profile Score (GPS) are the most used scores to sum up gait deviations and are used as primary outcomes in many clinical studies. They are considered as equivalent scores. The computation of these scores is based on a reference dataset but often no description is provided. Among other characteristics, the number of observations needed and its possible influence on the computation of the scores remains unknown.

Research question: Define the number of observations needed in the reference dataset to compute consistent and reliable GDI and GPS.

Methods: Fifty individuals with cerebral palsy (CP) were randomly selected from our laboratory database. Both scores were computed based on the reference dataset of Schwartz et al. (2008). A bootstrap analysis was performed, for every individual, to assess the effect of the number of observations on both scores. N number of observations were randomly selected, with replacement, from the reference dataset. This procedure was repeated 2000 times for every individual and every N and performed from N = 5 to N = 165 with an increment of 5. The 95 % of the absolute error distribution was considered for every individual and every N. The smallest detectable change (SDC) for both scores was considered as a threshold (GDI: 10.8; GPS:1.3°) to determine the minimum N required.

Results and significance: A minimum of 90 and 20 observations are required to compute consistent GDI and GPS, respectively. The number of observations has a higher impact on the GDI than the GPS, mainly because the GPS calculation does not rely on the standard deviation (SD). Furthermore, the GDI absolute error seems to be higher in individuals with greater gait deviations, i.e. lower GDI value. This effect was not observed on the GPS. In the case of a small reference dataset, the GPS should therefore be preferred.