A/IML. (998) A/DATA A/IML (998). (998) A B B X A Aj X j,, Bj A: X Aj ~ N, : X Bj ~ N, A B,, ; j,, B Z X A X B Z X A X j X Aj B X Bj j
A B Z,, Z A Z j ~ N, j B Z ~ N, H 0 : A B H A : A B H 0 : 0, H A : 0 c,, c H 0,, c,, c 0 c 0 H H P P Pr Pr c or c ad c or or c c ad ad c P P c,, c Eat Wag ad Tsats(987)oe parameter boudares c,, c c C C boudary shape parameter 0 0.5 0 O'Bree ad Flemg(979) 0. 5 Pocock(977) C c,, c P C ad C ad ad C 0 Pr uderh
998 Armtage et al.(969) Recursve Itegrato covoluto A/DATA C c,, c A/IML Recursve Itegrato 3. A/IML A/IML EQ EQCALE EQHIFT ell-defed EQ =0 EQCALE doma =0 EQHIFT 0 parametrzato EQ CALLEQ(prob, doma) < tscale, eps, de> prob stage sequtal test ( =0 (m+) ) (doma) ( m ) tscale stage eps de 3
parametrzato m c doma c c c c c C C C C C C.M.P prob parametrzato m m doma doma c doma c.m Pr c,, c, c Pr prob Pr Pr c Pr c, c Pr c,, c, c c Pr c, Pr c,, c, c Pr c, Pr c,, c, P [Pr Pr c ad c ad ad c c ad c ad ad c Pr c ad c ad ad c ] 3-Pocock Type I error 5 5% Pocock 0.5 c C.43 (,,3,4,5 ) Type I error A/IML A/IML 4
.43.43.43 3.43 4.43 5.43.43.43 3.43 4.43 5 PROB ALPHA Pocock EQCALE CALLEQCALE(prob, gscale, doma, level) < guess, tscale, eps, de> prob EQ stage sequtal test ( (m+) 5
) =0 (gscale) parameterzato C (doma) parameterzato (level) guess tscale stage eps de 3-Pocock Type I error 5 0.05 Pocock 0.5 c C (,,3,4,5 ) EQCALE oe parameter boudares C doma EQ 6
3 4 5 Wag ad Tsats(987)oe 3 4 5 parameter boudares CW=.4376 Pocock PROB PROB ALPHA EQHIFT CALLEQHIFT(prob, shft, doma, level) < guess, tscale, eps, de> prob EQ stage sequtal test ( (m+) ) (shft) parameterzato (doma) parameterzato C C C C C C (level) guess tscale stage eps de 3-3Pocock 5 5% Pocock 0.5 c C (,,3,4,5 ) 3- EQCALE EQHIFT c TypeII error - A B 0. 7
8 5.43 4.43 3.43.43.43 5.43 4.43 3.43.43.43 0.(plevel) DELTA= Pocock(977) (998) PROB PROB POWER 90% 4. hoefeld ad Rchter(98) 4-5.6%
P 3 A 7.8% e t P 0.0047 A 0.003 P 0.0047 5.6% 36 36 e 0. 56 Pr T P A ˆ ˆ T T TT P log N log, C =- C logˆ ˆ log C N0, C N log, 9
D 4 Z Z log () Z Z log = = =57 ~ 0.05 0. :F=3 :F=4 9 :A= 9 0
9 N N t 48 F t exp( t ) (,) P F 48 exp( 48 ) A F 48 exp( 48 ) 9 9 FEN N F 48 N F 48 N 3 exp( ) P H exp( 36) exp( ) A H exp( 36) A F A+F Pr T F A exp F A Pr T F A T A exp F PrT A exp A Pr T A PrT A PrT A Pr A 0 exp A d A A exp A d A exp A A
exp A Pr T F A exp F A exp A Pr T F A exp F A P A LEN N H N H FENLEN () D N=544 =7 5. 5-4- O'Bree-Flemg log log 5 FEN LEN N A
3 6. A/IML (998)
Wag ad Tsats(987) La ad DeMets(983) staggerd etry [] Armtage, P., McPherso, C.. ad Roe, B.C. (969). Repeated sgfcace tests o accumulatg data. Joural of the Royal tatstcal ocety A3, 35-44 [] La,.. ad DeMets, D.L. (983). Dscrete sequetal boudares for clcal trals. Bometrka 70, 659-663 [3] O'Bre, P.C. ad Flemg, T. R. (979). A multple testg procedure for clcal trals. Bometrcs 35, 549-556 [4] Pocock,. J. (977). Group sequetal methods the desg ad aalyss of clcal trals. Bometrka 64, 9-99 [5] A Isttute Ic (995). A/IML oftare: Chages ad Ehacemets through Release 6., Cary NC:A Isttute Ic. [6] hoefeld, D.A. ad Rchter, J. R. (98). Nomograms for calculatg the umber of patets eeded for a clcal tral th survval as a edpot. Bometrcs 38, 63-70 [7] (998). A/DATA, 7 A,77-0 [8] Wag,.. ad Tsats, A. A. (987). Approxmately optmal oe-parameter boudares for group sequetal trals. Bometrcs 43, 93-99 4