You are here

  • Home
  • Archive
  • Volume 58, Issue Suppl 1
  • 2.20 Statistical modelling of blood biomarkers of sport-related concussion: can we improve estimate precision by using all sample replicates instead of means and >20%CV data exclusion?

Article Text

Article menu
Download PDFPDF
Second Round Abstract Submissions
2) Sideline Screening
2.20 Statistical modelling of blood biomarkers of sport-related concussion: can we improve estimate precision by using all sample replicates instead of means and >20%CV data exclusion?
  1. Jason Tabor1,
  2. Linden Penner1,
  3. Jean-Michel Galarneau1,
  4. Jennifer Cooper2,
  5. Mohammad Ghodsi2,
  6. Douglas Fraser3,
  7. Carolyn Emery1,
  8. Cheryl Wellington2,
  9. Chantel Debert4
  1. 1Sport Injury Prevention Research Centre, University of Calgary, Calgary, Canada
  2. 2Department of Pathology and Laboratory Medicine, University of British Columbia, Vancouver, Canada
  3. 3Western University, London, Canada
  4. 4Department of Clinical Neurosciences, University of Calgary, Calgary, Canada

Abstract

Objective To investigate agreement between plasma biomarker measurements run in duplicate and explore the precision of statistical modelling approaches using all replicate data vs standard of practice sample means and >20%CV data exclusion.

Design Cohort study.

Setting Canadian high-school and community sport settings.

Participants Healthy youth athletes participating in the SHRed Concussions study (n=149, 48% female, ages 11–17).

Assessment of Risk Factors Previous concussion (Yes/No), age (years), and sex (M/F).

Outcome Measures Plasma GFAP, NF-L, UCH-L1, T-tau, and ptau-181 concentrations (SIMOA;Quanterix) were examined. 95% limits of agreement (95%LOA) between biomarker replicates were assessed using Bland-Altman (B-A) analysis. Three multivariable regression models (α=0.05) were performed to assess estimates of association between independent variables and natural-log (ln) transformed biomarker levels. Model 1 (M1): multi-level model, all replicate data; Model 2 (M2): means of replicates; Model 3 (M3): means of replicates, excluding pairs >20%CV.

Main Results B-A showed 95%LOA in GFAP (-17.74;18.2 pg/mL), UCH-L1 (-13.77;14.82 pg/mL), NF-L (-1.871;1.801 pg/mL), ptau-181 (-0.5314;0.5391 pg/mL) and T-tau (ln-T-tau back-transformed ratio 65.27%;150.03%). Biomarker-specific regression β-estimates differed between M1–3 for each biomarker (largest difference in M3 given data exclusion >20%CV). Age was associated with ln-UCH-L1 in M2 (β=0.0485; 95%CI:0.0006,0.0964). Sex was associated with ln-NF-L [(M1:β=0.1940; 95%CI:0.0135,0.3745); (M2:β=0.1945; 95%CI:0.0105,0.3786)] and ln-GFAP [(M1:β=-0.1676; 95%CI:-0.3037,-0.0315); (M2:β=-0.1668; 95%CI:-0.3066,-0.0270) in M1–2. M1 displayed narrowest 95%CIs (highest estimate precision).

Conclusions Wide 95%LOA in baseline blood biomarkers suggest the mean of sample duplicates may not reflect true population values. Future pediatric sport-related concussion studies may benefit from statistical modelling incorporating all replicate biomarker data to increase estimate accuracy and precision while considering the importance of age and sex.

https://doi.org/10.1136/bjsports-2023-concussion.211

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.