Minimizing Missing Data in Clinical Trials: Design, Operation, and Regulatory Considerations
- PDF / 9,744,141 Bytes
- 8 Pages / 612 x 792 pts (letter) Page_size
- 49 Downloads / 240 Views
DATA
485
Minimizing Missing Data in Clinical Trials: Design, Operation, and Regulatory Considerations
Guoxing (Greg) Soon, PhD Lead Mathematical Statistician,Division of Biometrics l y Office of Biostatistics. Office of Translational Science. Center for Drug Evaluation and Research, Food and Drug Administration, Silver Spring, Maryland
Key Words Clinical trial; Benefit/risk. Minimizing missing data; MCAR; MAR; Offtreatment follow-up; Sensitivity analysis; Labeling; Statistical methods Corrrspondrntr Address Guoxing (Greg) Soon, Division of Biometrics W, Office of Biostatistics. Bldg 21, 10993 New Hampshire Avenue, Silver Spring, MD 20993-0002 (email: [email protected]). Presented on April 16, 2008, at the 2nd Annual FDA/DIA Statistics Forum, Bethesda, Maryland.
The ultimate question to be answered in a clinical trial is "What benefivrisk has the drug caused to the patient?" To answer this question, it is m'tical to avoid missing data on outcome that meusurn benejit and risk j h the beginning. This article emphasizes that when a patient discontinues the study treatment, insteud of trying to impute the missing outcome values, effortshould be made to fdlinv the patient until the endpoint or the end of the trial. The information in the fdlmup period can be used to describe the benefithk to those patients who would discontinuetreatment in the a d clinical pmctice, while imputing values that are not observed, yet may never exist, can only answer
INTRODUCTION The presence of missing data in clinical trials has often been dealt with as an analysis issue, and various statistical methods like last observation carried forward (LOCF) or mixed model repeated measures (MMRM) have been proposed for this purpose (1-4). These statistical approaches rely on strong and sometimes unrealistic assumptions like missing completely at random (MCAR), missing at random (MAR) (S),or particular models for the missing mechanism, with the replacement values for the missing data being predicted based on the observed portion of the data. MCAR assumes patients who have missing data behave the same way as patients who do not have missing data in the study, while MAR assumes MCAR within subgroups defined by selected observed values. These approaches tend to answer a question like, "What would have happened had the patient remained in the study?" Often the type of patients whose measurements or outcomes missing in the clinical trial will also have missing value in actual practice when the drug is marketed, so that the "what if" conjecture is neither realistic nor necessary. This article is organized as follows. In the next
the unrealisticquestion, "What would have happened had the subject not discontinued?"Minimizing missing data is possible through thejoint efforts of drug companies and regulatory agencies. Drug companies can improve their efforts in study design and opemtion, including oftreatment fdlmup, prion'tization of infmation cdledion, endpoint seledion, and detailed documentation of reasons for missing data. Regulatory agencies
Data Loading...
