Skip to main content

How few? Bayesian statistics in injury biomechanics.

Publication ,  Journal Article
Cutcliffe, HC; Schmidt, AL; Lucas, JE; Bass, CR
Published in: Stapp car crash journal
October 2012

In injury biomechanics, there are currently no general a priori estimates of how few specimens are necessary to obtain sufficiently accurate injury risk curves for a given underlying distribution. Further, several methods are available for constructing these curves, and recent methods include Bayesian survival analysis. This study used statistical simulations to evaluate the fidelity of different injury risk methods using limited sample sizes across four different underlying distributions. Five risk curve techniques were evaluated, including Bayesian techniques. For the Bayesian analyses, various prior distributions were assessed, each incorporating more accurate information. Simulated subject injury and biomechanical input values were randomly sampled from each underlying distribution, and injury status was determined by comparing these values. Injury risk curves were developed for this data using each technique for various small sample sizes; for each, analyses on 2000 simulated data sets were performed. Resulting median predicted risk values and confidence intervals were compared with the underlying distributions. Across conditions, the standard and Bayesian survival analyses better represented the underlying distributions included in this study, especially for extreme (1, 10, and 90%) risk. This study demonstrates that the value of the Bayesian analysis is the use of informed priors. As the mean of the prior approaches the actual value, the sample size necessary for good reproduction of the underlying distribution with small confidence intervals can be as small as 2. This study provides estimates of confidence intervals and number of samples to allow the selection of the most appropriate sample sizes given known information.

Duke Scholars

Published In

Stapp car crash journal

DOI

EISSN

2993-1940

ISSN

1532-8546

Publication Date

October 2012

Volume

56

Start / End Page

349 / 386

Related Subject Headings

  • Wounds and Injuries
  • Survival Analysis
  • Sample Size
  • Risk Assessment
  • Models, Statistical
  • Confidence Intervals
  • Biomechanical Phenomena
  • Bayes Theorem
  • 4003 Biomedical engineering
  • 0903 Biomedical Engineering
 

Citation

APA
Chicago
ICMJE
MLA
NLM
Cutcliffe, H. C., Schmidt, A. L., Lucas, J. E., & Bass, C. R. (2012). How few? Bayesian statistics in injury biomechanics. Stapp Car Crash Journal, 56, 349–386. https://doi.org/10.4271/2012-22-0009
Cutcliffe, Hattie C., Allison L. Schmidt, Joseph E. Lucas, and Cameron R. Bass. “How few? Bayesian statistics in injury biomechanics.Stapp Car Crash Journal 56 (October 2012): 349–86. https://doi.org/10.4271/2012-22-0009.
Cutcliffe HC, Schmidt AL, Lucas JE, Bass CR. How few? Bayesian statistics in injury biomechanics. Stapp car crash journal. 2012 Oct;56:349–86.
Cutcliffe, Hattie C., et al. “How few? Bayesian statistics in injury biomechanics.Stapp Car Crash Journal, vol. 56, Oct. 2012, pp. 349–86. Epmc, doi:10.4271/2012-22-0009.
Cutcliffe HC, Schmidt AL, Lucas JE, Bass CR. How few? Bayesian statistics in injury biomechanics. Stapp car crash journal. 2012 Oct;56:349–386.

Published In

Stapp car crash journal

DOI

EISSN

2993-1940

ISSN

1532-8546

Publication Date

October 2012

Volume

56

Start / End Page

349 / 386

Related Subject Headings

  • Wounds and Injuries
  • Survival Analysis
  • Sample Size
  • Risk Assessment
  • Models, Statistical
  • Confidence Intervals
  • Biomechanical Phenomena
  • Bayes Theorem
  • 4003 Biomedical engineering
  • 0903 Biomedical Engineering