Intent-to-randomize corrections for missing data resulting from run-in selection bias in clinical trials for chronic conditions.

National Cancer Institute, Bethesda, Maryland, USA.
Journal of Biopharmaceutical Statistics (Impact Factor: 0.72). 03/2011; 21(2):263-70. DOI: 10.1080/10543406.2011.550107
Source: PubMed

ABSTRACT In many clinical trials for chronic conditions a run-in period is used prior to randomization. Often, only those participants who meet certain criteria during the run-in phase go on to get randomized. The others, along with the information that they might have provided, are excluded from the study. This exclusion of the relevant response data from any subsequent study analysis can be considered as resulting in missing data; although quite common in practice, this approach has expectedly been shown to create a bias in favor of the active treatment when this active treatment is used during the run-in. Hence, many randomized clinical trials report overly optimistic results, with the extent of the bias depending in large part on how many otherwise eligible subjects were excluded due to the use of the run-in. If these biased trials are to contribute valid information to medical decision making, then the biases need to be corrected, and this involves accounting for all participants who were intended to be randomized. We propose specific imputation methods for doing so.

  • Contemporary clinical trials 01/2012; 33(1):12. DOI:10.1016/j.cct.2011.09.014 · 1.99 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Although it is generally acknowledged that electronic monitoring of adherence to treatment improves blood pressure (BP) control by increasing patients' awareness to their treatment, little information is available on the long-term effect of this intervention. In this observational study among a total of 470 patients with mild-to-moderate hypertension, adherence was measured in 228 patients by means of both the Medication Event Monitoring System (MEMS) and pill count (intervention group), and in 242 patients by means of pill count alone (control group). During a follow-up period of 1 year consisting of seven visits to the physician's office, BP measurements were performed and medication adjusted based on the achieved BP. In addition, at each visit adherence to treatment was assessed. On the basis of pill counts, median adherence to treatment did not differ between the intervention group and the control group (96.1% vs. 94.2%; P = 0.97). In both groups, systolic and diastolic BP decreased similarly: 23/13 vs. 22/12 mm Hg in the intervention and control group respectively. Drug changes and the number of drugs used were associated with BP at the start of study, but not with electronic monitoring. In this study, electronic monitoring of adherence to treatment by means of MEMS did not lead to better long-term BP control nor did it result in less drug changes and drug use.
    American Journal of Hypertension 01/2012; 25(1):54-9. DOI:10.1038/ajh.2011.153 · 3.40 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In active run-in trials, where patients may be excluded after a run-in period based on their response to the treatment, it is implicitly assumed that patients have individual treatment effects. If individual patient data are available, active run-in trials can be modelled using patient-specific random effects. With more than one trial on the same medication available, one can obtain a more precise overall treatment effect estimate. We present a model for joint analysis of a two-sequence, four-period cross-over trial (AABB/BBAA) and a three-sequence, two-period active run-in trial (AB/AA/A), where the aim is to investigate the effect of a new treatment for patients with pain due to osteoarthritis. Our approach enables us to separately estimate the direct treatment effect for all patients, for the patients excluded after the active run-in trial prior to randomisation, and for the patients who completed the active run-in trial. A similar model approach can be used to analyse other types of run-in trials, but this depends on the data and type of other trials available. We assume equality of the various carry-over effects over time. The proposed approach is flexible and can be modified to handle other designs. Our results should be encouraging for those responsible for planning cost-efficient clinical development programmes.
    Clinical Trials 02/2012; 9(2):164-75. DOI:10.1177/1740774511430714 · 1.94 Impact Factor