5.1 Participants
Participant eligibility and enrolment information are visualized in Figure 1. From August 2015 – February 2018, 113 athletes with SRC were eligible for study enrollment. From this, 70 athletes were enrolled, and 49 consented to blood draw. Due to an inability to draw blood (n = 4), medications that interfered with inflammatory processes (n = 2), and an absence of SCAT symptom information (n – 3), 40 interuniversity athletes with a clinician diagnosed SRC were enrolled (n = 20 male (M), n = 20 female (F)), from nine sport teams: basketball (M & F), field hockey (F), football (M), ice hockey (M & F), lacrosse (M & F), mountain biking (F), rugby (M & F), soccer (F), volleyball (M & F). This cohort was analyzed in a previously published study by our group [6]. Concussion diagnosis and medical clearance decisions were made by a staff physician at the university sport medicine clinic in accordance with the Concussion in Sport Group guidelines [1]. Prior to enrollment, all participants provided written informed consent; all study procedures were in accordance with the declaration of Helsinki, and approved by the Health Science Research Ethics Board, University of Toronto (protocol reference # 27958).
5.2 Blood Biomarkers
Blood was sampled from athletes within a range of 2 - 7 days after an SRC (males, median = 4 days; females, median = 5 days). Athletes were excluded if they were taking medications other than birth control, or if they were currently symptomatic as a result of a known infection, illness or seasonal allergies. Venous blood was drawn into a 10-mL K2EDTA tube and was equilibrated for approximately one hour at room temperature before a two min centrifugation using a PlasmaPrep 12TM centrifuge (Separation Technology Inc., FL, USA). Plasma supernatant was then aliquoted and frozen at -80°C until analysis.
Nineteen cytokines and chemokines were analyzed by immunoassay using Meso Scale Diagnostics 96-well MULTI-SPOT® technology: interferon (IFN)-g, interleukin (IL)-1b, -2, -4, -6, -8, -10, -12p70, -13, tumor necrosis factor (TNF)-a, eotaxin, eotaxin-3, interferon gamma-induced protein (IP)-10, monocyte chemoattractant protein (MCP)-1, -4, macrophage-derived chemokine (MDC), macrophage inflammatory protein (MIP)-1a, -1b, and thymus and activation-regulated chemokine (TARC). Myeloperoxidase (MPO) was run as a single-plex assay. All assays were run according to manufacturer’s instructions, with individual samples run in duplicate.
5.3 Symptoms
On the day of the blood draw, athletes’ concussion symptoms were ascertained via a 22-item post-concussion symptom scale using a seven-point Likert rating as part of the Sport Concussion Assessment Tool (SCAT). The SCAT is the most widely used tool to assist in the diagnosis, management, and prognosis of individuals with concussion [1], and has shown reliability and validity for the assessment of both symptom presence and severity [63, 64]. A total symptom score was obtained by summing the presence or absence of each symptom irrespective of severity, with a maximum value of 22; symptom severity was evaluated by summing the rated symptom score for each symptom. In addition, four distinct symptom clusters were obtained by combining and summing the scores of SCAT symptoms related to somatic complaints (headache, pressure in head, neck pain, nausea/vomiting, dizziness, blurred vision, balance problems, sensitivity to light, sensitivity to noise), cognition (feeling slowed down, feeling in a fog, don’t feel right, difficulty concentrating, difficulty remembering, confusion), sleep (fatigue/low energy, drowsiness, trouble falling asleep), and emotion (more emotional, irritability, sadness, nervous/anxious). This approach has been previously employed by our group [6, 65].
5.4 Statistical Analysis
Prior to statistical analysis, we applied a previously published set of biomarker detection criteria [6], retaining only biomarkers that contained individual values for >80% of subjects per biomarker. Biomarker values were removed if they 1) did not fall within the manufacturer provided limits of detection, or 2) displayed a > 25% coefficient of variation between individual sample replicates. Hence, 10 of 20 inflammatory biomarkers satisfied these criteria and were evaluated in the current study. See Supplementary Table 1 for biomarker detection data in both male and female participant groups.
All variables (biomarkers and symptoms) were tested for deviations from normality by calculating sample skewness and kurtosis, with empirical p-values obtained by comparison against a simulated null distribution (random gaussian noise, 1000 simulated samples). In males, skewness ranged from 0 (p = 0.448) to 1.5 (p = 0.003), and kurtosis ranged from 3.2 (p = 0.721) to 9.4 (p = 0.001). In females, skewness ranged from – 0.9 (p = 0.949) to 2.2 (p < 0.001), and kurtosis ranged from 3.1 (p = 0.796) to 10.6 (p < 0.001). Hence, prior to all statistical comparisons, missing values were imputed separately in male and female subjects using the variable median, followed by rank transformation for symptom variables and a two-tailed winsorization (5%) of biomarkers that significantly deviated from normality.
Univariate comparisons of demographic variables and biomarkers between male and female athletes with SRC (Tables 1 & 2) were evaluated by calculating the mean differences, followed by bootstrapping of the mean difference scores (1000 resamples) to obtain standardized effect size in terms of bootstrap ratios (BSR; mean / standard error) and empirical p values based on the bootstrap estimates of the standard error, which were corrected at a false discovery rate (FDR) of 0.05. Univariate correlation analyses were conducted between calculated symptom severity and days to recovery via Spearman correlation with bootstrapping used to obtain BSRs and empirical p values.
The primary aim of the study was to test for associations between symptom reporting and inflammatory biomarker profiles in male and female athletes after SRC using a statistical framework designed to elucidate the potential complexity of these relationships. To accomplish this, a partial least squares (PLS) correlation analysis was employed [66]. PLS is a multivariate data reduction technique that creates orthogonal latent variables describing the maximal covariance between a set of predictor (biomarkers) and response (symptoms) variables [66]. In the current study, PLS was used in a bootstrap resampling framework (5000 iterations) to generate sets of variable loadings (i.e., weighted combinations of biomarkers / symptom clusters), along with corresponding BSRs and empirical p-values based on the bootstrap estimates of the standard error. For analyses evaluating the correlation between inflammatory biomarkers and calculated symptom severity - a single continuous predictor variable - an out-of-sample, leave-two-out cross correlation (R2) value was calculated on the PLS model. PLS plots represent the mean and standard deviation (SD) of the resamples for each variable. All statistical analyses were completed with R (RStudio, version 1.1.456, Boston, United States). Graphs were made with GraphPad Prism (version 7.0d, GraphPad Inc., La Jolla, CA, United States).