> 574y #bjbj;({{u
8V4@yyy $|n/
yyyyy/
Dy
y PsEvZ0X
yyyyyyy//yyyyyyyyyyyyyyyy$:A rescue strategy for handling unevaluable patients in Simons two-stage designs
L.Belin1,2,3 , Y.De Rycke3,4 , P.Brot2,5,6
1-Clinical Research Department, Institut Curie, Paris (France). 2-INSERM-U669,Villejuif (France). 3-INSERM-U900, Paris (France). 4- Public Health Department, Institut Curie, Paris (France). 5-JE2492, Universit Paris-Sud, Villejuif.
6-Paul Brousse Hospital, AP-HP, Villejuif (France).
For phase II oncology trials, Simons two-stage design is the most commonly used strategy. When clinically unevaluable patients occur, the total number of patients included at each stage differs from that initially planned. All the design properties (such as controlling type I and type II error rates) may no longer hold. Several questions arise: how to take into account unevaluable patients, whether there is a need to recalculate stopping boundaries and what are the operating characteristics of the new design. Various ad hoc strategies are used for handling these unevaluable patients: (i) Consider unevaluable patients as non-responders, (ii) Consider unevaluable patients as responders, (iii) Exclude the unevaluable patients, (iv) Include new patients in order to achieve the planned sample size. To the best of our knowledge, these ad hoc strategies are used without any knowledge of their consequences for type I and type II error rates.
The principal objective of this work is to evaluate how the overall type I and type II error rates are affected by these four strategies. We also propose a fifth strategy in which the critical stopping rules are reset in order to take into account patient availability at the study endpoint. The proposed strategy is a rescue strategy. It tries to minimize the deviation from the planned type I and type II error rates as much as possible without changing the initial sample size. The rescue strategy is based on the conditional probability of responding at an evaluation time for an evaluable patient. Estimate this probability allows building a new Simon design with parameter taking into account the probability to be unevaluable. It allows furnishing new stopping boundaries adapted to the number of observed unevaluable patients at each stage.
Simulation shows that every strategy provides a biased estimate of response rate. The bias is negative for all the strategy except the one which consider unevaluable patients as responders. These biases lead to deviations for the planned type I and type II error rates Assuming 20% of unevaluable patients, the power is reduced from 90% to 78% or 82% for the exclusion and the replacement strategy respectively. Type I error rate is reduced from 0.1 to 0.047 and 0.034 respectively. For these strategy type I error rate is conservative but power is strongly reduced which is a major drawback in a phase II trial. However, our proposed strategy is the one which best approaches the target error rates requirement (the power is 88% and type I error rate is inflated to 0.11 assuming 20% of unevaluable patients). Even if the rescue strategy assumes a Weibull latent failure times and uniform censoring times, simulation show performance of the proposed strategy are robust to this assumption.
We exemplify the interest of this rescue strategy by the re-analysis of a Phase II trial planned at the Curie Institute. The trial included breast cancer patients with cerebral metastases and showed 14% of unevaluable patients at each stage.
The rescue strategy represents a practical solution when a Simon phase II trial has not been implemented as planned. It performs well, provides meaningful information and is simple to implement without requiring much patients. It could be recommended for handling the occurrence unevaluable patients.
International Biometric Society
International Biometric Conference, Florence, ITALY, 6 11 July 2014
PQRTY^_ackmnpqx}~ ʾճճէq_qPqPhs6CJ]aJmHsH"hh6CJ]aJmHsH"hh'-6CJ]aJmHsH"h'-h'-6CJ]aJmH sH h')h
ph')h
pCJaJh')h')CJH*aJh')h6XCJaJh')h'-CJH*aJh')h'-CJaJh')h')CJaJh
ph`N#h
p5:CJ\aJh6X5:CJ\aJQR~i T
U
~$G]Ga$gd99$G]Ga$gdC$G]Ga$gd')$G]Ga$gdC ? f g h i j
!)?\{S
wowgwg_Wg_g_gOWhCCJaJhCCJaJhWCJaJh 3CJaJhGnCJaJhCJaJh')h
pmHsH"h')h
p6CJ]aJmHsHh'-6CJ]aJmHsHh')6CJ]aJmHsHhs6CJ]aJmHsH%h'-h'-6CJ\]aJmHsH"h')h')6CJ]aJmHsH"h'-h'-6CJ]aJmHsHS
T
u
y
569IsdsXdh99CJaJmH sH h:h:CJaJmH sH h:h99CJaJmH sH h "CJaJmH sH h(CJaJmH sH hGnCJaJmH sH hH^CJaJmH sH h(hCJaJmH sH h(h(CJaJmH sH h "CJaJh(CJaJhZCJaJh 3CJaJhCCJaJh
pCJaJAknrGK\pstʾvggXvIvIhwFh^CJaJmH sH hwFhCJaJmH sH hwFhsCJaJmH sH h}-CJaJmH sH hsCJaJmH sH h7CJaJmH sH h) eCJaJmH sH h?(CJaJmH sH h"CJaJmH sH h:dDCJaJmH sH h:CJaJmH sH h:h:CJaJmH sH h99CJaJmH sH h:h:dDCJaJmH sH $-FGKQaej
ʻ~znbbbh99CJaJnHtHh}jh996CJaJh99hCh99CJaJh99CJaJh[6jh[6UhwFh99CJaJmH sH hwFhB }CJaJmH sH hwFhi,cy$gc.bAn̖kեC A0vs<>6r=*%@&ٖ6دk.~8$
6./lS_//"JFV!zn^3ੵ,_wr%8/[k.$eVi2[Psjcs7f
WKՆ7dgȘJj|j(KD-
dXiJ؇2x$(
:'vɗsKZj{Zb&^>{?=уG]8 ˻^|ɟ>D<Ϫ~j ̜?_|+!D7!18fZNFl;I(q
=9S̲8vt
xur1xEu5$ IA.N&)f^݈8f2((}B*C.X;u0ɐjmڦ1eZ3ۉm-r"+00~Hƫxp\%rcVu*#Sq= !a"e՞-%ƪL.R(_%:漌iv@}OCbU|Cp0ݷ)q}2ܢcҬ@ULC5@W4;fj8?|
Tބӫg"qzr7$%G[r~KS,bm3v1el\fp}X̕4A6{ASLt(Q%\rl_٫fS_B,sHvx`Wr~(BshUr)
64Cebs)bMjBUkp:6GyZL3E2rQ$)9Gv$5˳:Y+x%vYRnNöj.7=.Yo|%lٟ̦gl厹MPW6s;<
liGY Dk/7!@Ί5(
Z2_]Zѱ_3*E
шMR*uahG.9gMW~#fpv4;
!6o%JۍsgwŴ9R.+<+FS*BiDpT߅PT/ȁa.jHP8T$Z2wzvvY,d*dL#r@Ps>=A6hן=Qr9LV638h&T-;ƬFtExZƚxY(wHH>vu[xA@U_iG08E[LZ
m6:9Oc֖&gv1^<`gvbm2{Eai_dLbd_.$z~30%M1T=0}o5PK!
ѐ'theme/theme/_rels/themeManager.xml.relsM
0wooӺ&݈Э5
6?$Q
,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{P8g/]QAsم(#L[PK-![Content_Types].xmlPK-!֧60_rels/.relsPK-!kytheme/theme/themeManager.xmlPK-!Ttheme/theme/theme1.xmlPK-!
ѐ' theme/theme/_rels/themeManager.xml.relsPK]
#7( --tttw S
##
8@0(
B
S ?ae
x !$Z^}!$3333!$<*u<W("C "`N#?(7*'-}- 3[67::dD=G6X:YZH^^
c) ei