The methods for handling missing data in clinical trials influence sample size requirements

J Clin Epidemiol. 2004 May;57(5):447-53. doi: 10.1016/j.jclinepi.2003.09.012.

Abstract

Objective: Results of studies estimating osteoarthritis progression may be affected by missing values. In clinical trials assessing disease-modifying osteoarthritis drugs, sample sizes should be calculated using close estimates of outcome variables.

Study design and setting: Supposing a two-parallel group design in hip osteoarthritis clinical trials, we estimated sample sizes using the joint space width (JSW), number of patients with JSW progression >0.5 mm (JSN), time to total hip arthroplasty (THA), and time to JSN or THA using several approaches to deal with missing data.

Results: Three-year clinical trials testing a treatment effect of 50%, with a power of 80%, could require sample sizes of 121 patients for JSW, 57 for JS progression using multiple imputation for handling missing values; 200 for THA; and 47 for JSN or THA. These numbers vary greatly depending on the approach chosen for handling missing data.

Conclusions: These results can help investigators plan clinical trials to select the primary outcome and a priori specify the way missing data will be handled.

MeSH terms

  • Aged
  • Antirheumatic Agents / therapeutic use*
  • Data Interpretation, Statistical
  • Disease Progression
  • Female
  • Humans
  • Male
  • Middle Aged
  • Multicenter Studies as Topic / methods
  • Osteoarthritis, Hip / drug therapy*
  • Osteoarthritis, Hip / pathology
  • Randomized Controlled Trials as Topic / methods*
  • Research Design
  • Sample Size
  • Treatment Outcome

Substances

  • Antirheumatic Agents