- Email: [email protected]
·P-58 Power and sample size calculations: A review and computer program
Abstracts
301
=nd~catedthree propert=es (vahd~ty,efhc=ency and complexity of the method). Nevertheless no decision procedure =s available Cons~denng the need of such a tool, we developed a computenzed model slmulat=ng chn=cal tnals to rule the we=ght of the d=fferent factors on the chmcal tr=al final results The factors accounted for the model are sample s~ze, number of part=c=pat=ngunits, patient charactenst=cs, and center charactenst=cs Bas=c pnnc=ples of th~s s~mulat~on are (1) the inclus=on of pat=ents ~n a chmcal trial and the charactenst~cs of each patient are random events s~mulated by the d=rect Monte-Carlo method, (2) each factor ~ss~mulated w~th a random multdevel d~screte variable, the allocated treatment being a b~nary variable, and (3) the response =s calculated for each s~mulated patient as the result of a hnear equat=on of covanables we=ghted w=th parameters Results: One hundred populat=ons of 100 s=mulated pat=ents were generated us=ng the SAS software random function. Four d~fferent allocation procedures (d~rect randomization w=th or w~thout blocks, strat=ficat~on w~th or w~thout blocks) were apphed to these populations One hundred chn~cal trials were s~mulated w=th each procedures Comparisons between procedures used the d=stance between the est~mat=onof the treatment effect obta=ned from s=mulated thai results and the theoret=cal one cons=denng the d~fferent values of the parameters Th~s model allows to evaluate the "error" generated by us=ng an =nappropr=ate random=zat~on procedure
P-57 SAMPLE SIZE FOR TRIALS WHERE SURVIVAL DEPENDS ON FIRST RESPONSE RATE Benny Zee and Joseph Pater
Nabonal Cancer Institute of Canada Chmca/ Tna/s Group K~ngston, Ontano, Canada A chmcal thai has been designed to compare ~nterferon versus no additional treatment ~n multiple myeloma patients who have responded to Melphalan and Predn~sone (MP) All patients w~ll be treated with MP once they are registered ~n the trial Random allocation of e~ther ~nterferon or no additional treatment wdl be made only ~f patients have responded to MP, which ~s characterized by a 50% drop ~n serum M-prote~n and 10% drop ~n urine M-prote~n on at least two separate measurements taken at least 4 weeks apart The objective of th~s paper ~s to present an approach to estimate the necessary number of registered patients to detect a g~ven d~fference ~n survival We assumed that patient's hfetime after randomization follows an exponential d~stnbut~onand ~talso depends on the t~me to response to MP, the latter ~s assumed to be exponentially d~stnbuted Th~s model allows adm~mstrat~ve censoring to occur ~n both t~me to response to MP and patient's I~fet~meafter randomization The parameters for estimating sample s~ze using th~s method are s~mdar to other methods except that th~s method requires the knowledge of the median time to response to MP, which can be obtained from prewous trials
• P-58 POWER AND SAMPLE SIZE CALCULATIONS: A REVIEW AND COMPUTER PROGRAM William D. Dupont and Walton D. Plummer, Jr.
Vanderbllt Umverslty School of Medlclne Nashville, Tennessee Methods of sample s~ze and power calculations are rewewed for the most common study designs. The sample s~ze and power equations for these designs are shown to be special cases of two generic formulae for sample s~ze and power calculations. An interactive computer program is available that can be used for studies w~th d~chotomous, continuous, or survival response measures The alternative hypotheses of ~nterest may be specified either m terms of dlffenng response rates, means, or survival times; or In terms of relative nsks or odds ratios. Studies w~th d~chotomous or continuous outcomes may ~nvolve e~ther a matched or ~ndependent study design The program can determine the sample s~ze needed to detect a specified alternative hypothes~s with the required power, the power with which a specific alternative hypothes~s can be detected w~th a g~ven sample s~ze, or the specific alternative hypotheses that can be detected w~th a g~ven power and sample s~ze. The program can generate help messages on request that facd~tate the use of th~s software It wntes a log file of all calculated estimates and can produce an output file for plotting power curves It ~s written ~n FORTRAN-77 and ~s ~n the public domain
301
=nd~catedthree propert=es (vahd~ty,efhc=ency and complexity of the method). Nevertheless no decision procedure =s available Cons~denng the need of such a tool, we developed a computenzed model slmulat=ng chn=cal tnals to rule the we=ght of the d=fferent factors on the chmcal tr=al final results The factors accounted for the model are sample s~ze, number of part=c=pat=ngunits, patient charactenst=cs, and center charactenst=cs Bas=c pnnc=ples of th~s s~mulat~on are (1) the inclus=on of pat=ents ~n a chmcal trial and the charactenst~cs of each patient are random events s~mulated by the d=rect Monte-Carlo method, (2) each factor ~ss~mulated w~th a random multdevel d~screte variable, the allocated treatment being a b~nary variable, and (3) the response =s calculated for each s~mulated patient as the result of a hnear equat=on of covanables we=ghted w=th parameters Results: One hundred populat=ons of 100 s=mulated pat=ents were generated us=ng the SAS software random function. Four d~fferent allocation procedures (d~rect randomization w=th or w~thout blocks, strat=ficat~on w~th or w~thout blocks) were apphed to these populations One hundred chn~cal trials were s~mulated w=th each procedures Comparisons between procedures used the d=stance between the est~mat=onof the treatment effect obta=ned from s=mulated thai results and the theoret=cal one cons=denng the d~fferent values of the parameters Th~s model allows to evaluate the "error" generated by us=ng an =nappropr=ate random=zat~on procedure
P-57 SAMPLE SIZE FOR TRIALS WHERE SURVIVAL DEPENDS ON FIRST RESPONSE RATE Benny Zee and Joseph Pater
Nabonal Cancer Institute of Canada Chmca/ Tna/s Group K~ngston, Ontano, Canada A chmcal thai has been designed to compare ~nterferon versus no additional treatment ~n multiple myeloma patients who have responded to Melphalan and Predn~sone (MP) All patients w~ll be treated with MP once they are registered ~n the trial Random allocation of e~ther ~nterferon or no additional treatment wdl be made only ~f patients have responded to MP, which ~s characterized by a 50% drop ~n serum M-prote~n and 10% drop ~n urine M-prote~n on at least two separate measurements taken at least 4 weeks apart The objective of th~s paper ~s to present an approach to estimate the necessary number of registered patients to detect a g~ven d~fference ~n survival We assumed that patient's hfetime after randomization follows an exponential d~stnbut~onand ~talso depends on the t~me to response to MP, the latter ~s assumed to be exponentially d~stnbuted Th~s model allows adm~mstrat~ve censoring to occur ~n both t~me to response to MP and patient's I~fet~meafter randomization The parameters for estimating sample s~ze using th~s method are s~mdar to other methods except that th~s method requires the knowledge of the median time to response to MP, which can be obtained from prewous trials
• P-58 POWER AND SAMPLE SIZE CALCULATIONS: A REVIEW AND COMPUTER PROGRAM William D. Dupont and Walton D. Plummer, Jr.
Vanderbllt Umverslty School of Medlclne Nashville, Tennessee Methods of sample s~ze and power calculations are rewewed for the most common study designs. The sample s~ze and power equations for these designs are shown to be special cases of two generic formulae for sample s~ze and power calculations. An interactive computer program is available that can be used for studies w~th d~chotomous, continuous, or survival response measures The alternative hypotheses of ~nterest may be specified either m terms of dlffenng response rates, means, or survival times; or In terms of relative nsks or odds ratios. Studies w~th d~chotomous or continuous outcomes may ~nvolve e~ther a matched or ~ndependent study design The program can determine the sample s~ze needed to detect a specified alternative hypothes~s with the required power, the power with which a specific alternative hypothes~s can be detected w~th a g~ven sample s~ze, or the specific alternative hypotheses that can be detected w~th a g~ven power and sample s~ze. The program can generate help messages on request that facd~tate the use of th~s software It wntes a log file of all calculated estimates and can produce an output file for plotting power curves It ~s written ~n FORTRAN-77 and ~s ~n the public domain