A review of RCTs in four medical journals to assess the use of imputation to overcome missing data in quality of life outcomes

Background: Randomised controlled trials (RCTs) are perceived as the gold-standard method for evaluating healthcare interventions, and increasingly include quality of life (QoL) measures. The observed results are susceptible to bias if a substantial proportion of outcome data are missing. The review aimed to determine whether imputation was used to deal with missing QoL outcomes. Methods: A random selection of 285 RCTs published during 2005/6 in the British Medical Journal, Lancet, New England Journal of Medicine and Journal of American Medical Association were identified. Results: QoL outcomes were reported in 61 (21%) trials. Six (10%) reported having no missing data, 20 (33%) reported <= 10% missing, eleven (18%) 11%-20% missing, and eleven (18%) reported >20% missing. Missingness was unclear in 13 (21%). Missing data were imputed in 19 (31%) of the 61 trials. Imputation was part of the primary analysis in 13 trials, but a sensitivity analysis in six. Last value carried forward was used in 12 trials and multiple imputation in two. Following imputation, the most common analysis method was analysis of covariance (10 trials). Conclusion: The majority of studies did not impute missing data and carried out a complete-case analysis. For those studies that did impute missing data, researchers tended to prefer simpler methods of imputation, despite more sophisticated methods being available.