SAS生存時間解析プロシジャの 最新の機能拡張

1 SAS What s new in SAS survival procedures. Chikuma Hamada Tokyo University of Science 1 2 3 Kyoto Heart Study Non-ARB LIFETEST kernel phreg CONTRAST ESTIMATE LSMEANS LSMESTIMATE HAZARDRATIO RANDOM LIFETEST phreg LSMESTIMATE 4 LIFETEST kernel phreg ESTIMATE LSMESTIMATE LSMEANS (LIFETEST HAZARDRATIO RANDOM ASSESS 5 6 kernel RANDOM ESTIMATE LSMESTIMATE LSMEANS HAZARDRATIO ASSESS 22110exp)()

2 (zzuththi (pointwise) CL:Confidence Limit (simultaneous) : CB:Confidence Band 7 GreenwoodtSVtSVtS: plots=s(cl) proc lifetest data=gehan plots=s(cl strata=panel test atrisk=0 to 40 by 5) method=km; time week*remiss(0); strata drug; run; 9 10 (0~1) tSgtSxg tSgtSg tSgtSgg 1g-1(x) tS xg11 LIFETEST procedure 5 CONFTYPE= LINEAR:NOTRANS LOG LOGLOG ASINSQRT LOGIT xxg1sin xxgloglog xxg xxglog xxxg 1log12 95% CL 95% LINEAR LOG LOGLOG ASINSQRT LOGIT ShSh 1 13 GEHAN 5 LINEAR LOG LOGLOG ASINSQRT LOGIT 14 LOGLOG 95% [ , ] 90% 95% 100% 50 100 200 400 NONTRANS LOG LOGLOG ASINSQRT LOGIT [n] 15 LOGLOG 95% [n=50, LOGLOG] 16 (2011) SAS Forum LIFETEST LOGLOG HW 17 EP(Equal Precision) HW(Hall-Wellner) tSntnaahtStStSntnaahtSSULSUL 1, 1, 2/122/12 tStaaetStStStaaetSSULSUL , , :: 2ttSS W.)

3 J. Hall and Jon A (1980) Nair V. N. (1984) 18 plots=s(cb) proc lifetest data=gehan plots=s(cb strata=panel test atrisk=0 to 40 by 5) method=km; time week*remiss(0); strata drug; run; 19 20 HW LOGLOG (2012) SAS Forum plots=s(cb cl) 21 proc lifetest data=gehan method=pl nelson plots(only)=hazard(kernel=b bw=5); time week*remiss(0); strata drug; run; 22 kernel( : ) bw(band width)=5 Nelson-Aalen di : i ni : i kernel( ) )(tH iittiindthndtHi ,23 Nelson ( ) WEEK Nelson-Aalen 0 0 0.

4 0 21 .. 1 20 2 19 .. 3 18 4 17 5 16 .. 6 15 7 14 .. 8 13 9 12 .. 10 11 .. 11 10 .. 12 9 13 8 2: drug = CONTROL 24 ttiiind212192212 Nelson ( ) Control 142162171192212 124826211312141 iindth 25 kernel-smoothed hazard Gehan 5 28 CONTROL 6-MP biweight (kernel ) iniitHbttKbth~1 1 2211615xxK (-1 x 1) K(x) th b 29 2211615xxK 5 =10 30 Control 5 31 34 FU (6 vs. 12 X-ACT 3 JFMC7-1,JFMC7-2,JFMC15 III overall survival, OS disease-free survival, DFS X-ACT JFMC7-1 JFMC7-2 JFMC15 5-FU/LV 5-FU,HCFU 6 12 983 1004 405 532 543 190 451 461 215 35 6 vs.)

5 12 (DFS ) 0~ : 12 32% 1 2 3 4 5 1 2 3 4 5 Logrankp<.0001 Logrank p< 6 12 6 12 b 1 36 KM 6 vs. 12 (OS ) 0~1 1 : 12 14% 1 2 3 4 5 Logrankp= 1 2 3 4 5 6 6 12 12 b 1 37 KM 38 kernel RANDOM ESTIMATE LSMESTIMATE LSMEANS HAZARDRATIO ASSESS 22110exp)()(zzuththi VA Lung Cancer data( phreg ) ESTIMATE LSMESTIMATE LSMEANS Veterans Administration lung cancer trial presented in Kalbfleisch and Prentice(1980) Time( ) Therapy(2) ( : standard or test) (i) Cell(4) ( : adeno, large, small, squamous) (j) Prior(2) (.)

6 0=no, 10=yes) Age( ) ( ) Duration ( ) ( ( ) Kps ( ) (Karnofsky performance scale)0~100 39 KM (VA Lung Cancer data) Cell (adeno-large ) adeno small large squamous 40 310: H KM (VA Lung Cancer data) Therapy (standard-test ) standard test 41 210: Hcell type =adeno large small squamous test standard 42 KM Cell Therapy 32310: H Cell Therapy Standard Test Adeno 11 12 1. Large 21 22 2. Small 31 32 3. Squamous 41 42 4..1 .2 .. 1 2 3 4 1 2 model Time*Status(0) = Cell Therapy Cell*Therapy; 321 120 210 220 311 110 410 420 43 ESTIMATE LSMESTIMATE Cell(i) small , (j) Standard Test H0 jiijjiijijijjiijthththth :),(log)(log)exp ()()(00 3231 44 phreg ESTIMATE LSMESTIMATE proc phreg data=VALung; class Prior Cell Therapy/param=glm; model Time*Status(0) = Cell Therapy Cell*Therapy; estimate "3small 1standard vs 2test" Cell*Therapy 0 0 0 0 1 -1 0 0/e exp; lsmestimate Cell*Therapy "3small 1standard vs 2test" 0 0 0 0 1 -1 0 0/e exp;run; 45 3231 2 Pr > ChiSq Cell adeno 1 <.

7 0001 Cell large 1 Cell small 1 <.0001 Cell squamous 0 0 .. Therapy standard 1 Therapy test 0 0 .. Cell*Therapy adeno standard 1 Cell*Therapy adeno test 0 0 .. Cell*Therapy large standard 1 Cell*Therapy large test 0 0 .. Cell*Therapy small standard 1 Cell*Therapy small test 0 0 .. Cell*Therapy squamous standard 0 0 .. Cell*Therapy squamous test 0 0 .. 31 32 2 1 46 ESTIMATE LSMESTIMATE z Pr > |z| 3small 1standard vs 2test Non-est .. Least Squares Means Estimate z Pr > |z| 3small 1standard vs 2test ESTIMATE LSMESTIMATE 47 LSMESTIMATE E Coefficients for Cell*Therapy Least Squares Means Estimate cell type type of treatment Row1 Celladeno adeno Celllarge large Cellsmall small Cellsquamous squamous Therapystandard standard 1 Therapytest test -1 CelladenoTherapystandard adeno standard CelladenoTherapytest adeno test CelllargeTherapystandard large standard CelllargeTherapytest large test CellsmallTherapystandard small standard 1 CellsmallTherapytest small test -1 CellsquamousTherapystandard squamous standard CellsquamousTherapytest squamous test 1 2 31 32 48 3231213231 Cell C Standard( 31) Test ( 32)

8 3231213231322332311331 ijjiij49 31 32 3 1 2 ESTIMATE ( Therapy 1 -1 Cell*Therapy 0 0 0 0 1 -1 0 0 3231213231322332311331424132312221121121 4321,,,,,,,,,, 21 3231 50 ESTIMATE LSMESTIMATE estimate "3small 1standard vs 2test" Therapy 1 -1 Cell*Therapy 0 0 0 0 1 -1 0 0; or estimate "3small 1standard vs 2test" Therapy 1 -1 Cell*Therapy 0 0 0 0 1 -1; ( 0 lsmestimate Cell*Therapy "3small 1standard vs 2test" 0 0 0 0 1 -1 0 0; (Cell*Therapy 51 Adeno Large H0 222221121121 52 jiijjiijijijjiijthththth :),()(log)exp ()()(00 Cell A B 2222221121121222112112122222221122112211 2111111 53 11 12 21 22 1 2 ESTIMATE Cell 1 -1 0 0 Cell*Therapy .5 .5 0 0 0 0 222,,,,,,,,22211211212221121121424132312 2211211214321 22211211 54 ESTIMATE LSMESTIMATE estimate "1adeno-2large" Cell 1 -1 0 0 Cell*Therapy.)))

9 5 .5 0 0 0 0; or estimate "1adeno-2large" Cell 1 -1 Cell*Therapy .5 .5 ; ( 0 lsmestimate Cell 1adeno-2large 1 -1 0 0; (Cell 55 Therapy Standard Test H0 44423222124131211121 56 jiijjiijijijjiijthththth :),()(log)exp ()()(00 Cell A B 57 11 12 21 22 31 32 41 42 21 ESTIMATE LSMESTIMATE estimate "1standard-2test" Therapy 1 -1 Cell*Therapy .25 .25 .25 .25 ; lsmestimate Therapy 1standard-2test 1 -1; (Therapy 4241323122211211 21 58 ESTIMATE LSMESTIMATE ESTIMATE 0 E LSMEANS LSMESTIMATE 59 ESTIMATE LSMESTIMATE LSMEANS 60 ESTIMATE LSMESTIMATE LSMEANS ESTIMATE ( ) (0 LSMESTIMATE ( ) LSMEANS ( ) 61 : SCANCER 10nmol HR=1 90nmol:HR= 30nmol.))))

10 HR= 66 LIFETEST STRATA ADJ= ADJUST=BONFERRONI ADJUST=DUNNETT ADJUST=SCHEFFE ADJUST=SIDAK ADJUST=SIMULATE ADJUST=SMM | GT2 ADJUST=TUKEY 67 TEST=WILCOXON DOSE 1 10 30 11 19 2 30 30 19 11 3 90 30 20 10 Total 90 50 40 A,B,C A B C AB AC BC Tukey AB,AC,BC Dunnett AB,AC Scheffe AB,AC,BC,A-BC A-BC m Bonferroni p m Sidak SIMULATED 68 m = m=10 - )m )10 = = m=10 1- - )m m 1- )10 = 69 p 1mm)1()'()()1(1)(p)(1)1(11 mxmxFxfxxFmm 70 m p 1 2 3 4 5 10 )(pf1)1()( mpmpf71 p m p 1 2 3 4 5 m=10 )(pF Bonferronipm)(0)1(1)(m pFpSidakppF72 1 rVVVVVVVVVuuuTrrrrrrr:,::221222211121121 uVuVuVu 73 uVu T2 Dunnett ( 10nmol ods graphics on; proc lifetest data=scancer plots=survival(atrisk=0 to 80 by 10); time time*censor(0); strata dose / test=logrank adjust=dunnett; run; ods graphics off; 74 uVu T2 75 DOSE 10 30 90 1u2u3u DOSE 10 30 90 10 30 90 11V21V31V12V13V22V23V32V33V 2 Pr > Chi-Square 2 <.