Development of a Statistical Methodology for Improved Analysis of Workplace Injuries.

Abstract

Alternative models for analyzing occupational injuries are developed to examine the shortcomings of conventional assumptions. Two models are presented: a Homogeneous Exposure Model and a Heterogeneous Exposure Model. In the Homogeneous Exposure Model, workers are assumed to be assigned to jobs of comparable physical stresses. This model assumes that some fraction of the worker population will not be injured. This is in contrast to conventional methods which assume an equal and constant risk of injury for all workers. A mixed Weibull distribution is used to describe the time to injury, where in conventional methods a simple exponential distribution is assumed. In the Heterogeneous Exposure Model, workers are allowed to switch jobs of different physical stresses. This model assumes a semi-Markov process for transitions between heterogeneous jobs and one injury state. For both models, maximum likelihood estimates for parameters were attained using a numerical nonlinear minimization based on a quasi-Newton method. Further, single censoring of data is taken into account for both models. The fidelity of the two models is illustrated using the injury data for 1,564 workers in a southwestern industrial plant. The constant injury rates obtained by conventional methods were in most cases significantly underestimated in comparison with the Homogeneous Exposure Model. This result underscores the importance of electing realistic modelling assumptions. There appeared to be no significant differences in the risk of major injury between high and low stress jobs. However, workers who were not matched to their jobs experienced higher major injury rates than those who were well matched. Further, workers with frequent minor injuries showed a greater probability of suffering a major injury compared to those with few minor injuries. This suggests the possibility of using minor injury experience to predict subsequent major injury. The Homogeneous Exposure Model also appeared to be useful in examining workers' injury recidivism or injury proneness tendency. Similar results were found with the Heterogeneous Exposure Model in terms of job exposure effects and the risk of injury.