Development of an injury risk curve for pelvic fracture in vertical loading environments

Abstract

Objective: Pelvis injury mechanisms are dependent upon loading direction (frontal, lateral, and vertical). Studies exist on the frontal and lateral modes; however, similar studies in the vertical mode are relatively sparse. Injury risk curves and response corridors are needed to delineate the biomechanical responses. The objective of the study was to derive risk curves for pelvis injuries using postmortem human subjects (PMHSs).Methods: Published data from whole-body PMHSs loaded axially through the pelvis were analyzed. Accelerometers were placed on the pelvis/sacrum and seat. Specimens were loaded along the inferior to superior direction using a horizontal sled or a vertical accelerator device. Specimens were positioned supine in the horizontal sled and seated upright on the vertical accelerator. Pre- and posttest images were obtained and autopsies were completed to document the pathology. Variables used in the development of risk curves included velocity, acceleration, time to peak acceleration, pulse duration of acceleration, and jerk for the seat and sacrum. Survival analysis was used for risk curves. To determine the best predictor of pelvis injury, the Brier Score metric (BSM) was used. The best parametric distribution was determined using the corrected Akaike information criterion (AICc). Injury data points were treated as either uncensored or left/interval censored. Noninjury data points were treated as right censored.Results: Twenty-four PMHS specimens were identified from 3 published data sets. Fifteen PMHS specimens sustained injuries and 9 remained intact. The BSM ranged from 1.24 to 24.75 and, in general, the BSMs for the seat metric-related scores were greater than the sacrum data. The sacrum acceleration was the optimal metric for predicting pelvis tolerance (lowest BSM). The Weibull distribution had the lowest AICc, with right and left/interval-censored data. This was also true when injury data were treated as exact (uncensored) observations. The 50% probability of injury was associated with 229 G for the uncensored analysis and 139 G for the censored analysis, and the quality indices in both cases were in the “good” range.Conclusions: Statistical determination of the best injury metric will help improve the accuracy of injury prediction, prioritize instrumentation choice in dummy development, and improve design criteria for crash mitigation. The present study showed that injury risk curves using response data are better biomechanical descriptors of human responses than exposure data. These data are important in automotive safety because complex loading of the pelvis, including submarining, occurs in frontal car crashes.