Sensitivity Analysis by Worst and Best Case Assessment: Is it Really Sensitive?
- PDF / 303,743 Bytes
- 5 Pages / 504 x 720 pts Page_size
- 68 Downloads / 176 Views
0092-861 5\99 Copyright 0 1999 I h g Inlomiation Associarion Inc.
SENSITIVITY ANALYSIS BY WORST AND BEST CASE ASSESSMENT IS IT REALLY SENSITIVE? KRISTINA UNNEBRINK,
PHD, AND
JURGEN WINDELER,
PHD
Department of Medical Biometry, University of Heidelberg, Heidelberg. Germany
The question of lion to perform an nnalvsis according t o ititention-to-trecit in the presence of missing information is oj‘mujor importunce. Several procedures have been suggested, but there is no generally accepted rule. Sensitivity analysis has been recommended as one solution. Extreme cuse analysis, consisting of worst case und best cuse analysis, has been widely requested to assess the sensitivit?:of results to the method OJ handling missing vnlues. In a simulation study the consequences of worst case and best case atiulwis on @pe I error and power of clinicul trials uith continuous endpoints ore elwluuted. The strutegies of worst case and best cuse urralysis are too extreme for a meaning/iul sensitivity analysis. Only .for very small dropout rates ure the two strntegies nble to discriminnte at all between different situations. The demand for sensitiviy anulysis is no sutisfacto ry .solution to the intention-to-trent problem unless its methods and application nre evaluated carefully, it is specified in advance, and its interpretution is clear-cut. Key Words: Sensitivity analysis; Missing values; Worst case analysis; Best case analysis; Intention-to-treat
INTRODUCTION SINCE APPEARING IN the 1960s, the principle of intention-to-treat ( I n ) has become widely accepted for the analysis of controlled clinical trials. It is now part of several guidelines and recommendations (1,2). Based on the general view that randomization is an important design principle for clinical trials, analysis by ITT preserves the balances achieved between treatment groups. The necessity to include all patients requires adequate information about all patients. Therefore, the question of how to perform an analysis according to ITT in the
Reprint address: Dr. Kristina Unnebrink, Department of Medical Biometry, University of Heidelberg. Im Neuenheimer Feld 305. D-69120 Heidelberg, Germany. E-mail: [email protected].
presence of missing information is of major importance. Several procedures have been suggested, but there is no generally accepted rule. The Committee for Proprietary Medicinal Products working party ( 2 ) and the International Conference on Harmonization E9 Guideline ( 3 ) state that: “unfortunately, no universally applicable methods of handling missing values can be recommended.” They therefore demand that: “an investigation should be made concerning the sensitivity of the results of analysis to the method of handling missing values, especially if the number of missing values is substantial.” Sensitivity analyses have also been recommended by several other authors (4.5). The rationale behind sensitivity analysis is to use at least two different methods of data analysis, one of which is likely to give an overoptimistic result. while the other i
Data Loading...
