Sample size calculations in studies using the EuroQol 5D

Abstract

Health-related quality of life (HRQoL) instruments are increasingly used as outcome variables in clinical trials, leading to a requirement for sample size calculations based on these variables. This paper aims to provide a guide to sample size calculations for use with the EuroQol-5D. The paper focuses on sample sizes required for comparative studies, and uses scores from two reference groups of general population and critically ill patients to determine sample sizes using the three parts of the EQ-5D (descriptive system, visual analogue scale (VAS), and EQ-5D index). The effect on sample sizes of different methods of categorising the three variables are compared, and comparisons are also made between sample sizes using parametric and non-parametric methods. Sample sizes required when the EQ-5D descriptive system is used as a binary variable (problems/no problems) are higher than or equal to those required when each dimension is categorised in three levels of severity (no problems, moderate problems, extreme problems). The use of three categories is appropriate in ill populations, though in more healthy populations two categories should be used. Due to the slight skewness of VAS data, and the equality of results using parametric and non-parametric methods, sample size calculations using the VAS should be based on a parametric approach. Sample sizes were considerably higher for the EQ-5D index when predefined intervals, as opposed to a score frequency based categorisation, were used with the general population reference group. Using the EQ-5D index in ill populations, it is recommended that sample size calculations are based on parametric methods, whilst in healthier populations non-parametric methods should be used.