Analyzing continuous injury outcomes with quantile regression methods

Background

Public health researchers use hospital charges to quantify injury severity and healthcare system burden. Charges tend to be right-skewed and violate statistical assumptions of ordinary least-squares (OLS) methods. Modeling conditional percentiles of hospital charges and other continuous injury outcomes provides insight into these outcomes and makes no distributional assumptions.

Methods

We analyzed total hospital charges for motorcyclists involved in Utah crashes between 1996 and 2004 using quantile regression. Data were obtained from crash records probabilistically matched to hospital records. Non-linked motorcyclist charges were $0. Multiple imputation methods were used to account for uncertainty in identifying matches and for missing data. We estimated the reduction in median hospital charges for helmeted vs. unhelmeted motorcycle operators. Results were compared to OLS regression results. The effect of helmet use on upper percentiles of hospital charges was investigated with quantile regression methods.

Results

After adjusting for year, urban/rural location, and age, helmet use was associated with a 46% (95% CI: 14%, 66%) reduction in median hospital charges for motorcyclists involved in a crash. OLS regression estimated a 49% (95% CI: 20%, 68%) reduction in mean charges. The effect of helmet use was similar for upper percentiles of charges with a 46% reduction in both the 75th and 90th percentiles of charges. The effects of year and location were similar for median and upper percentile charges, while the effect of age varied more across percentiles.

Conclusions

Quantile regression methods can be used to describe and analyze hospital charges and other continuous injury outcomes.