Missing data in clinical research: an integrated approach

The clinical study with no missing data has yet to be conducted – and never will be! Yet, despite its ubiquity, missing data and the issues they raise are still too often brushed under the carpet or inappropriately handled. Despite the ready availability of software, this situation has changed surprisingly slowly over the last decade.
Fortunately, the careful handling and reporting of missing data in the STOP GAP trial, reported in this issue, provides a ready model for researchers.5 STOP GAP, a multicentre, observer-blinded, randomized controlled trial, compared the cost-effectiveness of ciclosporin with prednisolone-initiated treatment for pyoderma gangrenosum over a period of 24 weeks. Data on healthcare resource use was collected at 8 and 24 weeks by clinic visits, telephone, trial drug logs and patient diaries. Following the Consolidated Standards of Reporting Trials (CONSORT) recommendations, their Table 1 includes the completeness of data, reporting of the number of participants per group for all analyses. The primary analysis (referred to by the authors as the base-case analysis) included multiple imputation to account for missing data, assuming data were missing at random (explained below).