Statistics Collaborative - Design and analysis for biomedical research


Missing data

Missing data are unavoidable in clinical trials. In preparing a report for a DMC, the programming staff at Statistics Collaborative, Inc. (SCI) search the interim database to be sure of presenting as accurately as possible any important safety outcome (such as death) that may be captured in multiple locations. For clinical study reports, SCI has experience with a variety of statistical methods, including carry-forward approaches, multiple imputation, and longitudinal models. SCI also has experience with sensitivity analyses for a range of diseases and types of data.

SCI encourages clients to distinguish between stopping study therapy and withdrawing consent. We work with clients to develop protocols that explicitly make this distinction and that encourage follow-up of patients who stop study medication.

Examples of SCI's work with missing data:

  • Quality of life: For a trial evaluating changes in quality of life over time, before we analyzed the data, we evaluated various methods to use in addressing missing data, including multiple imputation.
  • Early discontinuation: For a study of a novel formulation of a pain therapy that used a categorical outcome scale, we performed a thorough sensitivity analysis, including a range of imputation approaches and alternative analyses, to address a larger-than-anticipated proportion of participants who had discontinued early.
  • Using missing data and patterns of missingness to glean information: A client was conducting a trial of a novel delivery mechanism for a treatment for a pulmonary disease. SCI collaborated with the client's statisticians in writing the analysis plan in such a way as to develop prospectively an imputation algorithm that would account for those participants who discontinued taking the study drug, as well as those who continued taking it but were too ill to contribute data for the assessment of the primary outcome.
  • Multiple imputation: Trials of interventions for psychiatric diseases, especially psychoses, are notorious for the high proportion of patients who stop their study medication. SCI did multiple imputation for a trial of an intervention for schizophrenia.