OBJECTIVES:

Analysis of health-related quality of life (HRQoL) data is important to inform cost-effectiveness analyses (CEA) and missing data could create problems for the estimation of utility values. We conducted a targeted literature review (TLR) to identify methods of assessing and handling missing values in utility analysis.

METHODS:

Embase/Medline databases were searched for the last 10 years to identify published studies which handled missing values in their HRQoL analysis. We aimed only at the EQ-5D instrument, which is commonly used in various disease areas. The studies were critically appraised and relevant data on the assumptions and methodology were extracted.

RESULTS:

From the 476 records screened, we identified 47 publications, with most of them being CEA (22/47) and PRO (18/47) studies, along with some methodological, and observational studies (7/47). The most popular method (42/47) used to address missing values (either at domain- or index-level) was multiple imputation (MI), where the majority used the Multiple Imputation by Chained Equations (MICE) method (23/42) and produced the imputed datasets using the predictive mean matching (PMM) method. Naïve methods, such as Last Observation Carried Forward (LOCF) and single imputation (e.g., with mean) were employed in 6 publications. Regarding the missingness pattern, although only 3 studies assessed this in the missing data, most studies assumed that their EQ-5D data were missing at random (MAR) (22/47). There was also a considerable number of studies (15/47) that did not report details on the imputation methods used and the assumptions taken.

CONCLUSIONS:

Our research highlighted the preference for employing multiple imputation methods when handling missing EQ-5D data. Most of the studies assumed that the missing data are MAR, however, very few studies assessed the missingness pattern. This TLR aids future utility analyses by providing an overview of previously implemented methods to address missing values.