Effect of data censoring on thoracic injury risk curves

Objective

The first objective was to evaluate the effect of using less censored (i.e., exact and interval-censored) data on thoracic injury risk curves and the resulting injury probabilities. The second objective was to generate new injury risk curves to predict Abbreviated Injury Scale (AIS) 3+ rib fractures based on chest deflection.

Methods

Two data sets consisting of postmortem human surrogate (PMHS) tests with multipoint chest deflection measurements were compiled: A less censored data set consisting of exact and interval-censored data and a doubly censored data set consisting of left- and right-censored data. Chest deflection data from both data sets were processed in a consistent manner to calculate the maximum deflections at different locations across the chest. Survival analysis methods were used to generate nonparametric and parametric injury risk curves for serious skeletal injury. The total sample sizes and proportions of less censored data used to generate the risk curves were varied for each curve to evaluate the effects of sample size and less censored data on risk curve shape and predicted injury thresholds.

Results

Increasing the proportion of less censored data resulted in steeper injury risk curves and a higher predicted risk for a given amount of deflection. Differences in injury risk were more pronounced in the upper half of the injury risk curves. Introducing less censored data also produced narrower confidence intervals. At a total sample size of 79, increasing the percentage of less censored data from 0 to 30 had minimal effect on the shape of the risk curve.

Conclusions

Doubly censored chest deflection data have historically been used to generate thoracic injury risk curves for frontal motor vehicle crash events. This study found that incorporating less censored data into thoracic injury risk curves meaningfully affected the shape of the injury risk curves and their resulting injury risk predictions. All of the injury risk curves generated in the study predicted a lower threshold for serious rib fracture injury compared to previously developed injury risks curves that are currently in use in the field. Based on the results of this study, adding less censored data to injury risk curves should be strongly considered to improve thoracic injury risk curve prediction and confidence, especially for smaller sample sizes.