Application of robust statistical methods for sensitivity analysis of health-related quality of life outcomes

Abstract

Background: Researchers often use conventional parametric procedures to test hypotheses of health-related quality of life (HRQL) mean equality across patient groups. However, these techniques are sensitive to the presence of skewed distributions and unequal group variances, which may characterize many HRQL measures.

Purpose: To conduct a sensitivity analysis of conventional and robust approaches to test hypotheses of mean equality on HRQL measures for hematopoietic stem cell transplantation survivors and a healthy comparison group.

Methods: The methods applied were the conventional parametric procedure of least-squares analysis of variance applied to the raw scores, the conventional parametric procedure applied to transformed data, and a robust approximate degrees of freedom parametric procedure utilizing trimmed means and Winsorized variances.

Results: The choice of analysis method affected the conclusions about the null hypothesis of mean equality. More commonly observed, however, was a substantial difference in the value of the F-statistic and standard errors which was particularly evident in the measures with greater degrees of skewness and heterogeneity of variances.

Conclusions: Robust statistical tests should be incorporated into sensitivity analyses when analyzing HRQL data.