Statistical analyses

In an effort to be fully transparent and accelerate progress our data is available for data sharing on our website: https://scicrunch.org/odc-tbi. All bar and line graphs were made using GraphPad Prism version 8.01 software for Mac (GraphPad Software) to include individual data points (red circles) along with group bars representing the estimated marginal means and the standard error in general linear models (GLM), which were calculated by IBM SPSS Statistics for Windows, version 25.0 (IBM Corp., Armonk, NY, USA). Two-way analysis of variance (ANOVA) was used for multiple longitudinal comparisons using JMP^® 13 (SAS Institute) with the main effects of group (polytrauma combined, ipsilateral polytrauma, contralateral polytrauma, fracture only, TBI only) and time considered. F-ratios are reported with numerator and denominator degrees of freedom in subscript for all effects meeting the type I error rate of p < 0.05. One-way ANOVA with post-hoc analysis was used for multiple conditional comparisons followed by Tukey-Kramer Honestly Significant Difference (HSD) test. Principal component analysis (PCA) was performed using eigenvalue decomposition of the cross-correlation matrix of all outcome measures over time in SPSS for syndromic analysis of neurotrauma as described previously^41,71–73. The outcome measurements from fracture callus in the TBI only group were input as a missing value rather than zero, and the brain lesion in the fracture only group were input as zero due to the intact brain for the validity of statistical analysis. Listwise deletion was used for missing value in the PCA. The syndromic outcome space was plotted using PC1-3 axes without the factor rotation using GPL code written within IBM SPSS v.25 syntax. Each PC reflects an orthogonal linear combination of the variables that accounts for the maximum amount of the total variance in all outcome measures. Number of principal components (PCs) were determined according to the criteria: (1) the Kaiser rule, retaining PCs with eigenvalues greater than 1; (2) the Cattell rule, retaining principal components above the elbow in the scree plot; (3) PC over-determination, retaining components with at least four PC loading values above |0.6|. PC scores were calculated using the regression method. All PC loading values above |0.3| were retained for PC interpretation. The validity of the PC loading pattern was assessed using GLM on the PCA derived scores for ANOVA followed by Tukey’s HSD post-hoc test. Effect size is reported as eta-squared and the precise observed power is reported for ANOVAs performed on PC scores. In all graphs, a statistically significant relationship among the groups for all outcome measures was indicated with a bar and an asterisk according to the following probabilities: *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.