Insights into the science and the dynamics associated with the stimulus-injury relationship of the human system are often gained by focusing on the results obtained from human surrogate testing. These results are commonly mapped into an injury risk paradigm for the purpose of characterizing the stimulus injury response: the injury risk curve. However; the quality and quantity of the data along with the experimental design are critical factors when considering the value of the estimated risk curve. Presented in this article is an analysis of injury risk curves in terms of their usability and appropriateness with respect to the sample size, stimulus distribution over the critical range, censoring, shape of the underlying risk function, and the inclusion of "actual" (uncensored) along with censored data. The results from this analysis indicate that for "large" biomechanical injury data sets there is no advantage to using actual data; censored data will yield the same injury risk curve as actual data. However; for "small" biomechanical injury data sets the inclusion of actual data will significantly improve the quality of the resulting risk curve. In addition, the results show that the amount of injury data needed to generate a risk curve with a given confidence bound is not only dependent on the relative contribution of the censored data and actual data but also on the shape of the risk function along with the stimulus distribution over the critical range. Confidence intervals are presented for the thoracic injury risk and the head injury risk to show the influence of data distribution on the goodness of the risk function estimation. (Author/publisher).
Abstract