2.3. Statistical Analysis

Statistical analysis was performed using GraphPad Prism 6 (GraphPad Software, La Jolla, CA, USA). Pearson’s correlation coefficients were calculated to assess the correlation between the different MRI scores and neurodevelopmental outcomes measured by the cognitive composite score of the Bayley Scales of Infant and Toddler Development. Correlation coefficients were categorized into (1) weak correlation (0.2–0.39); (2) moderate correlation (0.4–0.59); (3) strong correlation (0.6–0.79); and (4) very strong correlation (0.8–1). R squared was calculated to analyze the association between the scoring system and the Bayley Scales of Infant and Toddler Development. As described by Cohen et al., the association was described as very weak (r² < 0.02), weak (r² 0.02–0.13), moderate (r² 0.13–0.26), and substantial (r² ≥ 0.26) [23].

A p-value of ≤0.05 was considered to be significant. Cronbach alpha was performed to assess the intra-rater agreement by rescoring a subset of 28 MRI scans. The intra-rater agreement was performed by the reader (2). The scoring date was separated in time by more than two months (time between the scorings). The number of patients who were rescored was randomly selected.

Each scoring system was categorized into 3 severity grades (mild, moderate, and severe) of MRI abnormalities by dividing the maximum possible points of each MRI score into three subgroups. For example, for the Rutherford score, a mild grade was defined by 0 to 3 points, a moderate grade by 4 to 7 points, and a severe grade by 8 to 11 points, as the maximum of 11 points was divided into three subgroups. Abnormalities were detected to study the frequency of detecting abnormalities using the different scoring systems. Every given point results in “detection”, and zero points in each scoring system means “no detection”. An exception was processed for the Weeke score as this is known to be of non-linear distribution, as shown in the study by Szakmar et al. [24].